id sid tid token lemma pos fcis-29830 1 1 frontiers frontier NOUN fcis-29830 1 2 in in ADP fcis-29830 1 3 computing computing NOUN fcis-29830 1 4 and and CCONJ fcis-29830 1 5 intelligent intelligent ADJ fcis-29830 1 6 systems system NOUN fcis-29830 1 7 issn issn VERB fcis-29830 1 8 : : PUNCT fcis-29830 1 9 2832 2832 NUM fcis-29830 1 10 - - SYM fcis-29830 1 11 6024 6024 NUM fcis-29830 1 12 | | NOUN fcis-29830 1 13 vol vol NOUN fcis-29830 1 14 . . PROPN fcis-29830 2 1 11 11 NUM fcis-29830 2 2 , , PUNCT fcis-29830 2 3 no no INTJ fcis-29830 2 4 . . NOUN fcis-29830 2 5 2 2 NUM fcis-29830 2 6 , , PUNCT fcis-29830 2 7 2025 2025 NUM fcis-29830 2 8 83 83 NUM fcis-29830 2 9 feature feature NOUN fcis-29830 2 10 recognition recognition NOUN fcis-29830 2 11 of of ADP fcis-29830 2 12 ultrasound ultrasound ADJ fcis-29830 2 13 breast breast NOUN fcis-29830 2 14 images image NOUN fcis-29830 2 15 based base VERB fcis-29830 2 16 on on ADP fcis-29830 2 17 improved improve VERB fcis-29830 2 18 dsa‐u++ dsa‐u++ DET fcis-29830 2 19 weichun weichun PROPN fcis-29830 2 20 guan guan PROPN fcis-29830 2 21 , , PUNCT fcis-29830 2 22 yuanyuan yuanyuan PROPN fcis-29830 2 23 zhang zhang PROPN fcis-29830 3 1 * * PROPN fcis-29830 3 2 , , PUNCT fcis-29830 3 3 xinshu xinshu PROPN fcis-29830 3 4 lv lv PROPN fcis-29830 3 5 , , PUNCT fcis-29830 3 6 shuyun shuyun PROPN fcis-29830 3 7 lv lv PROPN fcis-29830 3 8 , , PUNCT fcis-29830 3 9 mao mao PROPN fcis-29830 3 10 lin lin PROPN fcis-29830 3 11 , , PUNCT fcis-29830 3 12 zhengyang zhengyang PROPN fcis-29830 3 13 liu liu PROPN fcis-29830 3 14 university university PROPN fcis-29830 3 15 of of ADP fcis-29830 3 16 science science NOUN fcis-29830 3 17 and and CCONJ fcis-29830 3 18 technology technology NOUN fcis-29830 3 19 liaoning liaoning NOUN fcis-29830 3 20 , , PUNCT fcis-29830 3 21 anshan anshan PROPN fcis-29830 3 22 liaoning liaoning NOUN fcis-29830 3 23 , , PUNCT fcis-29830 3 24 114001 114001 NUM fcis-29830 3 25 china china PROPN fcis-29830 3 26 * * PUNCT fcis-29830 3 27 corresponding correspond VERB fcis-29830 3 28 author author NOUN fcis-29830 3 29 : : PUNCT fcis-29830 3 30 yuanyuan yuanyuan PROPN fcis-29830 3 31 zhang zhang PROPN fcis-29830 3 32 abstract abstract PROPN fcis-29830 3 33 : : PUNCT fcis-29830 3 34 the the DET fcis-29830 3 35 aim aim NOUN fcis-29830 3 36 of of ADP fcis-29830 3 37 this this DET fcis-29830 3 38 project project NOUN fcis-29830 3 39 is be AUX fcis-29830 3 40 to to PART fcis-29830 3 41 recognize recognize VERB fcis-29830 3 42 features feature NOUN fcis-29830 3 43 in in ADP fcis-29830 3 44 ultrasound ultrasound ADJ fcis-29830 3 45 breast breast NOUN fcis-29830 3 46 images image NOUN fcis-29830 3 47 . . PUNCT fcis-29830 4 1 and and CCONJ fcis-29830 4 2 an an DET fcis-29830 4 3 improved improved ADJ fcis-29830 4 4 dsa dsa NOUN fcis-29830 4 5 - - PUNCT fcis-29830 4 6 u++ u++ ADJ fcis-29830 4 7 model model NOUN fcis-29830 4 8 is be AUX fcis-29830 4 9 proposed propose VERB fcis-29830 4 10 based base VERB fcis-29830 4 11 on on ADP fcis-29830 4 12 the the DET fcis-29830 4 13 image image NOUN fcis-29830 4 14 classification classification NOUN fcis-29830 4 15 task task NOUN fcis-29830 4 16 , , PUNCT fcis-29830 4 17 the the DET fcis-29830 4 18 traditional traditional ADJ fcis-29830 4 19 u u NOUN fcis-29830 4 20 - - NOUN fcis-29830 4 21 net++ net++ PROPN fcis-29830 4 22 in in ADP fcis-29830 4 23 the the DET fcis-29830 4 24 encoder encoder NOUN fcis-29830 4 25 part part NOUN fcis-29830 4 26 in in ADP fcis-29830 4 27 the the DET fcis-29830 4 28 face face NOUN fcis-29830 4 29 of of ADP fcis-29830 4 30 ultrasound ultrasound ADJ fcis-29830 4 31 images image NOUN fcis-29830 4 32 there there PRON fcis-29830 4 33 is be VERB fcis-29830 4 34 a a DET fcis-29830 4 35 lack lack NOUN fcis-29830 4 36 of of ADP fcis-29830 4 37 feature feature NOUN fcis-29830 4 38 extraction extraction NOUN fcis-29830 4 39 , , PUNCT fcis-29830 4 40 to to PART fcis-29830 4 41 solve solve VERB fcis-29830 4 42 this this DET fcis-29830 4 43 problem problem NOUN fcis-29830 4 44 we we PRON fcis-29830 4 45 use use VERB fcis-29830 4 46 resnet50 resnet50 NOUN fcis-29830 4 47 as as ADP fcis-29830 4 48 the the DET fcis-29830 4 49 backbone backbone NOUN fcis-29830 4 50 of of ADP fcis-29830 4 51 the the DET fcis-29830 4 52 extraction extraction NOUN fcis-29830 4 53 , , PUNCT fcis-29830 4 54 in in ADP fcis-29830 4 55 order order NOUN fcis-29830 4 56 to to PART fcis-29830 4 57 further far ADV fcis-29830 4 58 enhance enhance VERB fcis-29830 4 59 the the DET fcis-29830 4 60 ability ability NOUN fcis-29830 4 61 of of ADP fcis-29830 4 62 the the DET fcis-29830 4 63 feature feature NOUN fcis-29830 4 64 learning learning NOUN fcis-29830 4 65 , , PUNCT fcis-29830 4 66 we we PRON fcis-29830 4 67 also also ADV fcis-29830 4 68 introduced introduce VERB fcis-29830 4 69 the the DET fcis-29830 4 70 aspp aspp NOUN fcis-29830 4 71 module module NOUN fcis-29830 4 72 , , PUNCT fcis-29830 4 73 to to PART fcis-29830 4 74 help help VERB fcis-29830 4 75 capture capture VERB fcis-29830 4 76 contextual contextual ADJ fcis-29830 4 77 information information NOUN fcis-29830 4 78 at at ADP fcis-29830 4 79 different different ADJ fcis-29830 4 80 scales scale NOUN fcis-29830 4 81 , , PUNCT fcis-29830 4 82 and and CCONJ fcis-29830 4 83 a a DET fcis-29830 4 84 module module NOUN fcis-29830 4 85 r r NOUN fcis-29830 4 86 - - PUNCT fcis-29830 4 87 as as ADP fcis-29830 4 88 is be AUX fcis-29830 4 89 designed design VERB fcis-29830 4 90 to to PART fcis-29830 4 91 enhance enhance VERB fcis-29830 4 92 the the DET fcis-29830 4 93 model model NOUN fcis-29830 4 94 multi multi ADJ fcis-29830 4 95 - - ADJ fcis-29830 4 96 scale scale ADJ fcis-29830 4 97 perception perception NOUN fcis-29830 4 98 ability ability NOUN fcis-29830 4 99 . . PUNCT fcis-29830 5 1 information information NOUN fcis-29830 5 2 to to PART fcis-29830 5 3 enhance enhance VERB fcis-29830 5 4 the the DET fcis-29830 5 5 model model NOUN fcis-29830 5 6 's 's PART fcis-29830 5 7 multi multi ADJ fcis-29830 5 8 - - ADJ fcis-29830 5 9 scale scale ADJ fcis-29830 5 10 perception perception NOUN fcis-29830 5 11 ability ability NOUN fcis-29830 5 12 and and CCONJ fcis-29830 5 13 a a DET fcis-29830 5 14 module module NOUN fcis-29830 5 15 r r NOUN fcis-29830 5 16 - - PUNCT fcis-29830 5 17 as as ADP fcis-29830 5 18 . . PUNCT fcis-29830 6 1 the the DET fcis-29830 6 2 output output NOUN fcis-29830 6 3 of of ADP fcis-29830 6 4 r r NOUN fcis-29830 6 5 - - PUNCT fcis-29830 6 6 as as ADV fcis-29830 6 7 after after ADP fcis-29830 6 8 five five NUM fcis-29830 6 9 stages stage NOUN fcis-29830 6 10 is be AUX fcis-29830 6 11 used use VERB fcis-29830 6 12 as as ADP fcis-29830 6 13 the the DET fcis-29830 6 14 encoding encoding NOUN fcis-29830 6 15 part part NOUN fcis-29830 6 16 of of ADP fcis-29830 6 17 u u NOUN fcis-29830 6 18 - - NOUN fcis-29830 6 19 net++ net++ X fcis-29830 6 20 , , PUNCT fcis-29830 6 21 and and CCONJ fcis-29830 6 22 the the DET fcis-29830 6 23 feature feature NOUN fcis-29830 6 24 information information NOUN fcis-29830 6 25 extracted extract VERB fcis-29830 6 26 by by ADP fcis-29830 6 27 the the DET fcis-29830 6 28 encoder encoder NOUN fcis-29830 6 29 is be AUX fcis-29830 6 30 reconstructed reconstruct VERB fcis-29830 6 31 and and CCONJ fcis-29830 6 32 enhanced enhance VERB fcis-29830 6 33 in in ADP fcis-29830 6 34 the the DET fcis-29830 6 35 decoder decoder NOUN fcis-29830 6 36 part part NOUN fcis-29830 6 37 . . PUNCT fcis-29830 7 1 in in ADP fcis-29830 7 2 order order NOUN fcis-29830 7 3 to to PART fcis-29830 7 4 reduce reduce VERB fcis-29830 7 5 the the DET fcis-29830 7 6 computational computational ADJ fcis-29830 7 7 complexity complexity NOUN fcis-29830 7 8 and and CCONJ fcis-29830 7 9 the the DET fcis-29830 7 10 number number NOUN fcis-29830 7 11 of of ADP fcis-29830 7 12 parameters parameter NOUN fcis-29830 7 13 , , PUNCT fcis-29830 7 14 and and CCONJ fcis-29830 7 15 to to PART fcis-29830 7 16 maintain maintain VERB fcis-29830 7 17 a a DET fcis-29830 7 18 certain certain ADJ fcis-29830 7 19 feature feature NOUN fcis-29830 7 20 extraction extraction NOUN fcis-29830 7 21 ability ability NOUN fcis-29830 7 22 , , PUNCT fcis-29830 7 23 we we PRON fcis-29830 7 24 replace replace VERB fcis-29830 7 25 the the DET fcis-29830 7 26 traditional traditional ADJ fcis-29830 7 27 convolution convolution NOUN fcis-29830 7 28 in in ADP fcis-29830 7 29 the the DET fcis-29830 7 30 decoder decoder NOUN fcis-29830 7 31 part part NOUN fcis-29830 7 32 of of ADP fcis-29830 7 33 u u NOUN fcis-29830 7 34 - - NOUN fcis-29830 7 35 net++ net++ PROPN fcis-29830 7 36 with with ADP fcis-29830 7 37 a a DET fcis-29830 7 38 depth depth NOUN fcis-29830 7 39 - - PUNCT fcis-29830 7 40 separable separable NOUN fcis-29830 7 41 convolution convolution NOUN fcis-29830 7 42 , , PUNCT fcis-29830 7 43 which which PRON fcis-29830 7 44 is be AUX fcis-29830 7 45 experimentally experimentally ADV fcis-29830 7 46 validated validate VERB fcis-29830 7 47 on on ADP fcis-29830 7 48 the the DET fcis-29830 7 49 ai ai ADJ fcis-29830 7 50 algorithm algorithm NOUN fcis-29830 7 51 elite elite ADJ fcis-29830 7 52 challenge challenge NOUN fcis-29830 7 53 dataset dataset NOUN fcis-29830 7 54 , , PUNCT fcis-29830 7 55 which which PRON fcis-29830 7 56 consists consist VERB fcis-29830 7 57 of of ADP fcis-29830 7 58 ultrasound ultrasound ADJ fcis-29830 7 59 breast breast NOUN fcis-29830 7 60 images image NOUN fcis-29830 7 61 with with ADP fcis-29830 7 62 four four NUM fcis-29830 7 63 features feature NOUN fcis-29830 7 64 , , PUNCT fcis-29830 7 65 namely namely ADV fcis-29830 7 66 , , PUNCT fcis-29830 7 67 orientation orientation NOUN fcis-29830 7 68 , , PUNCT fcis-29830 7 69 edges edge NOUN fcis-29830 7 70 , , PUNCT fcis-29830 7 71 calcifications calcification NOUN fcis-29830 7 72 , , PUNCT fcis-29830 7 73 and and CCONJ fcis-29830 7 74 shapes shape NOUN fcis-29830 7 75 , , PUNCT fcis-29830 7 76 and and CCONJ fcis-29830 7 77 is be AUX fcis-29830 7 78 trained train VERB fcis-29830 7 79 with with ADP fcis-29830 7 80 the the DET fcis-29830 7 81 dsa dsa NOUN fcis-29830 7 82 - - PUNCT fcis-29830 7 83 u++ u++ PROPN fcis-29830 7 84 model model NOUN fcis-29830 7 85 . . PUNCT fcis-29830 8 1 after after SCONJ fcis-29830 8 2 the the DET fcis-29830 8 3 dsa dsa PROPN fcis-29830 8 4 - - PUNCT fcis-29830 8 5 u++ u++ ADJ fcis-29830 8 6 model model NOUN fcis-29830 8 7 is be AUX fcis-29830 8 8 trained train VERB fcis-29830 8 9 , , PUNCT fcis-29830 8 10 it it PRON fcis-29830 8 11 improves improve VERB fcis-29830 8 12 on on ADP fcis-29830 8 13 several several ADJ fcis-29830 8 14 indexes index NOUN fcis-29830 8 15 , , PUNCT fcis-29830 8 16 not not PART fcis-29830 8 17 only only ADV fcis-29830 8 18 improving improve VERB fcis-29830 8 19 the the DET fcis-29830 8 20 recognition recognition NOUN fcis-29830 8 21 ability ability NOUN fcis-29830 8 22 of of ADP fcis-29830 8 23 subtle subtle ADJ fcis-29830 8 24 features feature NOUN fcis-29830 8 25 in in ADP fcis-29830 8 26 ultrasound ultrasound ADJ fcis-29830 8 27 breast breast NOUN fcis-29830 8 28 images image NOUN fcis-29830 8 29 , , PUNCT fcis-29830 8 30 but but CCONJ fcis-29830 8 31 also also ADV fcis-29830 8 32 effectively effectively ADV fcis-29830 8 33 improving improve VERB fcis-29830 8 34 the the DET fcis-29830 8 35 performance performance NOUN fcis-29830 8 36 of of ADP fcis-29830 8 37 image image NOUN fcis-29830 8 38 classification classification NOUN fcis-29830 8 39 . . PUNCT fcis-29830 9 1 keywords keyword NOUN fcis-29830 9 2 : : PUNCT fcis-29830 9 3 ultrasound ultrasound PROPN fcis-29830 9 4 breast breast NOUN fcis-29830 9 5 imaging imaging NOUN fcis-29830 9 6 ; ; PUNCT fcis-29830 9 7 resnet resnet NOUN fcis-29830 9 8 ; ; PUNCT fcis-29830 9 9 depth depth NOUN fcis-29830 9 10 - - PUNCT fcis-29830 9 11 separable separable NOUN fcis-29830 9 12 convolution convolution NOUN fcis-29830 9 13 ; ; PUNCT fcis-29830 9 14 u u NOUN fcis-29830 9 15 - - NOUN fcis-29830 9 16 net++ net++ PROPN fcis-29830 9 17 . . PUNCT fcis-29830 10 1 1 1 X fcis-29830 10 2 . . X fcis-29830 10 3 introduction introduction NOUN fcis-29830 10 4 : : PUNCT fcis-29830 10 5 in in ADP fcis-29830 10 6 2021 2021 NUM fcis-29830 10 7 , , PUNCT fcis-29830 10 8 breast breast NOUN fcis-29830 10 9 cancer cancer NOUN fcis-29830 10 10 has have AUX fcis-29830 10 11 surpassed surpass VERB fcis-29830 10 12 lung lung NOUN fcis-29830 10 13 cancer cancer NOUN fcis-29830 10 14 as as ADP fcis-29830 10 15 the the DET fcis-29830 10 16 most most ADV fcis-29830 10 17 frequently frequently ADV fcis-29830 10 18 diagnosed diagnose VERB fcis-29830 10 19 cancer cancer NOUN fcis-29830 10 20 worldwide worldwide ADV fcis-29830 10 21 , , PUNCT fcis-29830 10 22 and and CCONJ fcis-29830 10 23 globally globally ADV fcis-29830 10 24 , , PUNCT fcis-29830 10 25 especially especially ADV fcis-29830 10 26 in in ADP fcis-29830 10 27 women woman NOUN fcis-29830 10 28 early early ADJ fcis-29830 10 29 detection detection NOUN fcis-29830 10 30 and and CCONJ fcis-29830 10 31 accurate accurate ADJ fcis-29830 10 32 diagnosis diagnosis NOUN fcis-29830 10 33 are be AUX fcis-29830 10 34 important important ADJ fcis-29830 10 35 to to PART fcis-29830 10 36 improve improve VERB fcis-29830 10 37 patient patient ADJ fcis-29830 10 38 survival survival NOUN fcis-29830 11 1 [ [ X fcis-29830 11 2 1 1 NUM fcis-29830 11 3 ] ] PUNCT fcis-29830 11 4 . . PUNCT fcis-29830 12 1 and and CCONJ fcis-29830 12 2 the the DET fcis-29830 12 3 world world PROPN fcis-29830 12 4 health health PROPN fcis-29830 12 5 organization organization NOUN fcis-29830 12 6 's 's PART fcis-29830 12 7 international international ADJ fcis-29830 12 8 agency agency NOUN fcis-29830 12 9 for for ADP fcis-29830 12 10 research research NOUN fcis-29830 12 11 on on ADP fcis-29830 12 12 cancer cancer NOUN fcis-29830 12 13 has have AUX fcis-29830 12 14 released release VERB fcis-29830 12 15 shows show NOUN fcis-29830 12 16 that that SCONJ fcis-29830 12 17 it it PRON fcis-29830 12 18 is be AUX fcis-29830 12 19 expected expect VERB fcis-29830 12 20 to to PART fcis-29830 12 21 be be AUX fcis-29830 12 22 increasing increase VERB fcis-29830 12 23 by by ADP fcis-29830 12 24 2070 2070 NUM fcis-29830 13 1 [ [ X fcis-29830 13 2 2 2 NUM fcis-29830 13 3 , , PUNCT fcis-29830 13 4 3 3 NUM fcis-29830 13 5 ] ] PUNCT fcis-29830 13 6 . . PUNCT fcis-29830 14 1 among among ADP fcis-29830 14 2 the the DET fcis-29830 14 3 many many ADJ fcis-29830 14 4 imaging imaging NOUN fcis-29830 14 5 methods method NOUN fcis-29830 14 6 , , PUNCT fcis-29830 14 7 ultrasound ultrasound ADJ fcis-29830 14 8 imaging imaging NOUN fcis-29830 14 9 has have AUX fcis-29830 14 10 become become VERB fcis-29830 14 11 an an DET fcis-29830 14 12 important important ADJ fcis-29830 14 13 tool tool NOUN fcis-29830 14 14 for for ADP fcis-29830 14 15 the the DET fcis-29830 14 16 detection detection NOUN fcis-29830 14 17 of of ADP fcis-29830 14 18 breast breast NOUN fcis-29830 14 19 diseases disease NOUN fcis-29830 14 20 due due ADP fcis-29830 14 21 to to ADP fcis-29830 14 22 its its PRON fcis-29830 14 23 noninvasiveness noninvasiveness NOUN fcis-29830 14 24 , , PUNCT fcis-29830 14 25 real real ADJ fcis-29830 14 26 - - PUNCT fcis-29830 14 27 time time NOUN fcis-29830 14 28 nature nature NOUN fcis-29830 14 29 , , PUNCT fcis-29830 14 30 and and CCONJ fcis-29830 14 31 good good ADJ fcis-29830 14 32 discriminatory discriminatory ADJ fcis-29830 14 33 ability ability NOUN fcis-29830 14 34 for for ADP fcis-29830 14 35 dense dense ADJ fcis-29830 14 36 breast breast NOUN fcis-29830 14 37 tissue tissue NOUN fcis-29830 14 38 . . PUNCT fcis-29830 15 1 however however ADV fcis-29830 15 2 , , PUNCT fcis-29830 15 3 due due ADP fcis-29830 15 4 to to ADP fcis-29830 15 5 the the DET fcis-29830 15 6 inherent inherent ADJ fcis-29830 15 7 limitations limitation NOUN fcis-29830 15 8 of of ADP fcis-29830 15 9 ultrasound ultrasound ADJ fcis-29830 15 10 images image NOUN fcis-29830 15 11 themselves themselves PRON fcis-29830 15 12 and and CCONJ fcis-29830 15 13 the the DET fcis-29830 15 14 complexity complexity NOUN fcis-29830 15 15 of of ADP fcis-29830 15 16 breast breast NOUN fcis-29830 15 17 anatomy anatomy NOUN fcis-29830 15 18 , , PUNCT fcis-29830 15 19 its its PRON fcis-29830 15 20 application application NOUN fcis-29830 15 21 in in ADP fcis-29830 15 22 breast breast NOUN fcis-29830 15 23 tumor tumor NOUN fcis-29830 15 24 detection detection NOUN fcis-29830 15 25 and and CCONJ fcis-29830 15 26 diagnosis diagnosis NOUN fcis-29830 15 27 still still ADV fcis-29830 15 28 faces face VERB fcis-29830 15 29 many many ADJ fcis-29830 15 30 challenges challenge NOUN fcis-29830 15 31 . . PUNCT fcis-29830 16 1 first first ADV fcis-29830 16 2 , , PUNCT fcis-29830 16 3 the the DET fcis-29830 16 4 quality quality NOUN fcis-29830 16 5 of of ADP fcis-29830 16 6 ultrasound ultrasound NOUN fcis-29830 16 7 images image NOUN fcis-29830 16 8 and and CCONJ fcis-29830 16 9 the the DET fcis-29830 16 10 complexity complexity NOUN fcis-29830 16 11 of of ADP fcis-29830 16 12 anatomical anatomical ADJ fcis-29830 16 13 structures structure NOUN fcis-29830 16 14 greatly greatly ADV fcis-29830 16 15 affect affect VERB fcis-29830 16 16 the the DET fcis-29830 16 17 accuracy accuracy NOUN fcis-29830 16 18 of of ADP fcis-29830 16 19 detection detection NOUN fcis-29830 16 20 . . PUNCT fcis-29830 17 1 the the DET fcis-29830 17 2 quality quality NOUN fcis-29830 17 3 of of ADP fcis-29830 17 4 ultrasound ultrasound ADJ fcis-29830 17 5 images image NOUN fcis-29830 17 6 is be AUX fcis-29830 17 7 constrained constrain VERB fcis-29830 17 8 by by ADP fcis-29830 17 9 a a DET fcis-29830 17 10 variety variety NOUN fcis-29830 17 11 of of ADP fcis-29830 17 12 factors factor NOUN fcis-29830 17 13 , , PUNCT fcis-29830 17 14 such such ADJ fcis-29830 17 15 as as ADP fcis-29830 17 16 the the DET fcis-29830 17 17 depth depth NOUN fcis-29830 17 18 of of ADP fcis-29830 17 19 acoustic acoustic ADJ fcis-29830 17 20 wave wave NOUN fcis-29830 17 21 penetration penetration NOUN fcis-29830 17 22 , , PUNCT fcis-29830 17 23 the the DET fcis-29830 17 24 density density NOUN fcis-29830 17 25 of of ADP fcis-29830 17 26 the the DET fcis-29830 17 27 breast breast NOUN fcis-29830 17 28 , , PUNCT fcis-29830 17 29 and and CCONJ fcis-29830 17 30 the the DET fcis-29830 17 31 scattering scattering NOUN fcis-29830 17 32 of of ADP fcis-29830 17 33 the the DET fcis-29830 17 34 acoustic acoustic ADJ fcis-29830 17 35 beam beam NOUN fcis-29830 17 36 . . PUNCT fcis-29830 18 1 breast breast NOUN fcis-29830 18 2 tissue tissue NOUN fcis-29830 18 3 consists consist VERB fcis-29830 18 4 of of ADP fcis-29830 18 5 a a DET fcis-29830 18 6 variety variety NOUN fcis-29830 18 7 of of ADP fcis-29830 18 8 components component NOUN fcis-29830 18 9 such such ADJ fcis-29830 18 10 as as ADP fcis-29830 18 11 fat fat ADJ fcis-29830 18 12 , , PUNCT fcis-29830 18 13 glands gland NOUN fcis-29830 18 14 , , PUNCT fcis-29830 18 15 and and CCONJ fcis-29830 18 16 fibers fiber NOUN fcis-29830 18 17 , , PUNCT fcis-29830 18 18 and and CCONJ fcis-29830 18 19 its its PRON fcis-29830 18 20 complex complex ADJ fcis-29830 18 21 anatomical anatomical ADJ fcis-29830 18 22 structure structure NOUN fcis-29830 18 23 is be AUX fcis-29830 18 24 prone prone ADJ fcis-29830 18 25 to to AUX fcis-29830 18 26 form form VERB fcis-29830 18 27 shadows shadow NOUN fcis-29830 18 28 , , PUNCT fcis-29830 18 29 artifacts artifact NOUN fcis-29830 18 30 , , PUNCT fcis-29830 18 31 and and CCONJ fcis-29830 18 32 unclear unclear ADJ fcis-29830 18 33 structures structure NOUN fcis-29830 18 34 in in ADP fcis-29830 18 35 the the DET fcis-29830 18 36 image image NOUN fcis-29830 18 37 , , PUNCT fcis-29830 18 38 which which PRON fcis-29830 18 39 makes make VERB fcis-29830 18 40 accurate accurate ADJ fcis-29830 18 41 detection detection NOUN fcis-29830 18 42 and and CCONJ fcis-29830 18 43 localization localization NOUN fcis-29830 18 44 of of ADP fcis-29830 18 45 tumors tumor NOUN fcis-29830 18 46 extraordinarily extraordinarily ADV fcis-29830 18 47 difficult difficult ADJ fcis-29830 18 48 . . PUNCT fcis-29830 19 1 secondly secondly ADV fcis-29830 19 2 , , PUNCT fcis-29830 19 3 surrounding surround VERB fcis-29830 19 4 tissues tissue NOUN fcis-29830 19 5 , , PUNCT fcis-29830 19 6 shape shape NOUN fcis-29830 19 7 , , PUNCT fcis-29830 19 8 margin margin NOUN fcis-29830 19 9 contour contour NOUN fcis-29830 19 10 , , PUNCT fcis-29830 19 11 lesion lesion NOUN fcis-29830 19 12 boundaries boundary NOUN fcis-29830 19 13 , , PUNCT fcis-29830 19 14 and and CCONJ fcis-29830 19 15 posterior posterior ADJ fcis-29830 19 16 acoustic acoustic ADJ fcis-29830 19 17 features feature NOUN fcis-29830 19 18 are be AUX fcis-29830 19 19 important important ADJ fcis-29830 19 20 factors factor NOUN fcis-29830 19 21 to to PART fcis-29830 19 22 be be AUX fcis-29830 19 23 considered consider VERB fcis-29830 19 24 in in ADP fcis-29830 19 25 classifying classify VERB fcis-29830 19 26 a a DET fcis-29830 19 27 lesion lesion NOUN fcis-29830 19 28 [ [ X fcis-29830 19 29 4 4 NUM fcis-29830 19 30 ] ] PUNCT fcis-29830 19 31 . . PUNCT fcis-29830 20 1 the the DET fcis-29830 20 2 morphology morphology NOUN fcis-29830 20 3 of of ADP fcis-29830 20 4 breast breast NOUN fcis-29830 20 5 tumors tumor NOUN fcis-29830 20 6 may may AUX fcis-29830 20 7 vary vary VERB fcis-29830 20 8 from from ADP fcis-29830 20 9 individual individual ADJ fcis-29830 20 10 to to ADP fcis-29830 20 11 individual individual NOUN fcis-29830 20 12 ; ; PUNCT fcis-29830 20 13 for for ADP fcis-29830 20 14 example example NOUN fcis-29830 20 15 , , PUNCT fcis-29830 20 16 some some DET fcis-29830 20 17 tumors tumor NOUN fcis-29830 20 18 are be AUX fcis-29830 20 19 extremely extremely ADV fcis-29830 20 20 small small ADJ fcis-29830 20 21 and and CCONJ fcis-29830 20 22 irregularly irregularly ADV fcis-29830 20 23 shaped shape VERB fcis-29830 20 24 , , PUNCT fcis-29830 20 25 while while SCONJ fcis-29830 20 26 others other NOUN fcis-29830 20 27 may may AUX fcis-29830 20 28 be be AUX fcis-29830 20 29 highly highly ADV fcis-29830 20 30 similar similar ADJ fcis-29830 20 31 to to ADP fcis-29830 20 32 the the DET fcis-29830 20 33 surrounding surround VERB fcis-29830 20 34 tissue tissue NOUN fcis-29830 20 35 . . PUNCT fcis-29830 21 1 the the DET fcis-29830 21 2 accuracy accuracy NOUN fcis-29830 21 3 for for ADP fcis-29830 21 4 identifying identify VERB fcis-29830 21 5 tumors tumor NOUN fcis-29830 21 6 of of ADP fcis-29830 21 7 different different ADJ fcis-29830 21 8 sizes size NOUN fcis-29830 21 9 also also ADV fcis-29830 21 10 varies vary VERB fcis-29830 21 11 [ [ X fcis-29830 21 12 5 5 NUM fcis-29830 21 13 ] ] PUNCT fcis-29830 21 14 , , PUNCT fcis-29830 21 15 and and CCONJ fcis-29830 21 16 this this DET fcis-29830 21 17 diversity diversity NOUN fcis-29830 21 18 not not PART fcis-29830 21 19 only only ADV fcis-29830 21 20 increases increase VERB fcis-29830 21 21 the the DET fcis-29830 21 22 difficulty difficulty NOUN fcis-29830 21 23 of of ADP fcis-29830 21 24 tumor tumor NOUN fcis-29830 21 25 detection detection NOUN fcis-29830 21 26 in in ADP fcis-29830 21 27 ultrasound ultrasound NOUN fcis-29830 21 28 images image NOUN fcis-29830 21 29 , , PUNCT fcis-29830 21 30 but but CCONJ fcis-29830 21 31 also also ADV fcis-29830 21 32 puts put VERB fcis-29830 21 33 higher high ADJ fcis-29830 21 34 demands demand NOUN fcis-29830 21 35 on on ADP fcis-29830 21 36 the the DET fcis-29830 21 37 robustness robustness NOUN fcis-29830 21 38 of of ADP fcis-29830 21 39 the the DET fcis-29830 21 40 model model NOUN fcis-29830 21 41 . . PUNCT fcis-29830 22 1 in in ADP fcis-29830 22 2 addition addition NOUN fcis-29830 22 3 , , PUNCT fcis-29830 22 4 the the DET fcis-29830 22 5 similarity similarity NOUN fcis-29830 22 6 of of ADP fcis-29830 22 7 benign benign ADJ fcis-29830 22 8 and and CCONJ fcis-29830 22 9 malignant malignant ADJ fcis-29830 22 10 breast breast NOUN fcis-29830 22 11 tumor tumor NOUN fcis-29830 22 12 characteristics characteristic NOUN fcis-29830 22 13 further far ADV fcis-29830 22 14 exacerbates exacerbate VERB fcis-29830 22 15 the the DET fcis-29830 22 16 diagnostic diagnostic ADJ fcis-29830 22 17 complexity complexity NOUN fcis-29830 22 18 . . PUNCT fcis-29830 23 1 certain certain ADJ fcis-29830 23 2 benign benign ADJ fcis-29830 23 3 and and CCONJ fcis-29830 23 4 malignant malignant ADJ fcis-29830 23 5 tumors tumor NOUN fcis-29830 23 6 may may AUX fcis-29830 23 7 present present VERB fcis-29830 23 8 similar similar ADJ fcis-29830 23 9 morphological morphological ADJ fcis-29830 23 10 features feature NOUN fcis-29830 23 11 in in ADP fcis-29830 23 12 ultrasound ultrasound ADJ fcis-29830 23 13 images image NOUN fcis-29830 23 14 , , PUNCT fcis-29830 23 15 such such ADJ fcis-29830 23 16 as as ADP fcis-29830 23 17 clear clear ADJ fcis-29830 23 18 boundaries boundary NOUN fcis-29830 23 19 or or CCONJ fcis-29830 23 20 homogeneous homogeneous ADJ fcis-29830 23 21 internal internal ADJ fcis-29830 23 22 echoes echo NOUN fcis-29830 23 23 , , PUNCT fcis-29830 23 24 etc etc X fcis-29830 23 25 . . X fcis-29830 23 26 , , PUNCT fcis-29830 23 27 which which PRON fcis-29830 23 28 poses pose VERB fcis-29830 23 29 a a DET fcis-29830 23 30 great great ADJ fcis-29830 23 31 challenge challenge NOUN fcis-29830 23 32 to to PART fcis-29830 23 33 accurately accurately ADV fcis-29830 23 34 differentiate differentiate VERB fcis-29830 23 35 benign benign ADJ fcis-29830 23 36 and and CCONJ fcis-29830 23 37 malignant malignant ADJ fcis-29830 23 38 tumors tumor NOUN fcis-29830 23 39 . . PUNCT fcis-29830 24 1 finally finally ADV fcis-29830 24 2 , , PUNCT fcis-29830 24 3 noise noise NOUN fcis-29830 24 4 and and CCONJ fcis-29830 24 5 artifacts artifact NOUN fcis-29830 24 6 in in ADP fcis-29830 24 7 ultrasound ultrasound ADJ fcis-29830 24 8 images image NOUN fcis-29830 24 9 are be AUX fcis-29830 24 10 not not PART fcis-29830 24 11 to to PART fcis-29830 24 12 be be AUX fcis-29830 24 13 ignored ignore VERB fcis-29830 24 14 . . PUNCT fcis-29830 25 1 false false ADJ fcis-29830 25 2 - - PUNCT fcis-29830 25 3 positive positive ADJ fcis-29830 25 4 or or CCONJ fcis-29830 25 5 false false ADJ fcis-29830 25 6 - - PUNCT fcis-29830 25 7 negative negative ADJ fcis-29830 25 8 results result NOUN fcis-29830 25 9 often often ADV fcis-29830 25 10 appear appear VERB fcis-29830 25 11 in in ADP fcis-29830 25 12 ultrasound ultrasound ADJ fcis-29830 25 13 images image NOUN fcis-29830 25 14 due due ADJ fcis-29830 25 15 to to ADP fcis-29830 25 16 interfering interfere VERB fcis-29830 25 17 factors factor NOUN fcis-29830 25 18 such such ADJ fcis-29830 25 19 as as ADP fcis-29830 25 20 instrument instrument NOUN fcis-29830 25 21 noise noise NOUN fcis-29830 25 22 , , PUNCT fcis-29830 25 23 artifacts artifact NOUN fcis-29830 25 24 , , PUNCT fcis-29830 25 25 and and CCONJ fcis-29830 25 26 patient patient ADJ fcis-29830 25 27 motion motion NOUN fcis-29830 25 28 artifacts artifact NOUN fcis-29830 25 29 , , PUNCT fcis-29830 25 30 leading lead VERB fcis-29830 25 31 to to ADP fcis-29830 25 32 misdetections misdetection NOUN fcis-29830 25 33 or or CCONJ fcis-29830 25 34 missed miss VERB fcis-29830 25 35 detections detection NOUN fcis-29830 25 36 . . PUNCT fcis-29830 26 1 deep deep ADJ fcis-29830 26 2 learning learning NOUN fcis-29830 26 3 can can AUX fcis-29830 26 4 learn learn VERB fcis-29830 26 5 more more ADJ fcis-29830 26 6 image image NOUN fcis-29830 26 7 features feature NOUN fcis-29830 26 8 from from ADP fcis-29830 26 9 a a DET fcis-29830 26 10 large large ADJ fcis-29830 26 11 amount amount NOUN fcis-29830 26 12 of of ADP fcis-29830 26 13 image image NOUN fcis-29830 26 14 data datum NOUN fcis-29830 26 15 , , PUNCT fcis-29830 26 16 which which PRON fcis-29830 26 17 is be AUX fcis-29830 26 18 of of ADP fcis-29830 26 19 great great ADJ fcis-29830 26 20 significance significance NOUN fcis-29830 26 21 in in ADP fcis-29830 26 22 the the DET fcis-29830 26 23 field field NOUN fcis-29830 26 24 of of ADP fcis-29830 26 25 image image NOUN fcis-29830 26 26 processing processing NOUN fcis-29830 26 27 [ [ X fcis-29830 26 28 6 6 NUM fcis-29830 26 29 ] ] PUNCT fcis-29830 26 30 . . PUNCT fcis-29830 27 1 it it PRON fcis-29830 27 2 has have AUX fcis-29830 27 3 been be AUX fcis-29830 27 4 widely widely ADV fcis-29830 27 5 used use VERB fcis-29830 27 6 in in ADP fcis-29830 27 7 fields field NOUN fcis-29830 27 8 such such ADJ fcis-29830 27 9 as as ADP fcis-29830 27 10 computer computer NOUN fcis-29830 27 11 vision vision NOUN fcis-29830 27 12 [ [ X fcis-29830 27 13 7 7 NUM fcis-29830 27 14 , , PUNCT fcis-29830 27 15 8 8 NUM fcis-29830 27 16 ] ] PUNCT fcis-29830 27 17 and and CCONJ fcis-29830 27 18 natural natural ADJ fcis-29830 27 19 language language NOUN fcis-29830 27 20 processing processing NOUN fcis-29830 27 21 [ [ X fcis-29830 27 22 9 9 NUM fcis-29830 27 23 , , PUNCT fcis-29830 27 24 10 10 NUM fcis-29830 27 25 ] ] PUNCT fcis-29830 27 26 . . PUNCT fcis-29830 28 1 however however ADV fcis-29830 28 2 , , PUNCT fcis-29830 28 3 traditional traditional ADJ fcis-29830 28 4 ultrasound ultrasound ADJ fcis-29830 28 5 image image NOUN fcis-29830 28 6 diagnosis diagnosis NOUN fcis-29830 28 7 based base VERB fcis-29830 28 8 on on ADP fcis-29830 28 9 texture texture ADJ fcis-29830 28 10 and and CCONJ fcis-29830 28 11 morphological morphological ADJ fcis-29830 28 12 features feature NOUN fcis-29830 28 13 of of ADP fcis-29830 28 14 its its PRON fcis-29830 28 15 images image NOUN fcis-29830 28 16 , , PUNCT fcis-29830 28 17 which which PRON fcis-29830 28 18 are be AUX fcis-29830 28 19 manually manually ADV fcis-29830 28 20 analyzed analyze VERB fcis-29830 28 21 by by ADP fcis-29830 28 22 experienced experienced ADJ fcis-29830 28 23 doctors doctor NOUN fcis-29830 28 24 [ [ X fcis-29830 28 25 11 11 NUM fcis-29830 28 26 ] ] PUNCT fcis-29830 28 27 , , PUNCT fcis-29830 28 28 shows show VERB fcis-29830 28 29 some some DET fcis-29830 28 30 limitations limitation NOUN fcis-29830 28 31 in in ADP fcis-29830 28 32 the the DET fcis-29830 28 33 recognition recognition NOUN fcis-29830 28 34 of of ADP fcis-29830 28 35 breast breast NOUN fcis-29830 28 36 lesions lesion NOUN fcis-29830 28 37 . . PUNCT fcis-29830 29 1 with with ADP fcis-29830 29 2 the the DET fcis-29830 29 3 rapid rapid ADJ fcis-29830 29 4 development development NOUN fcis-29830 29 5 of of ADP fcis-29830 29 6 deep deep ADJ fcis-29830 29 7 learning learning NOUN fcis-29830 29 8 technology technology NOUN fcis-29830 29 9 , , PUNCT fcis-29830 29 10 many many ADJ fcis-29830 29 11 remarkable remarkable ADJ fcis-29830 29 12 results result NOUN fcis-29830 29 13 have have AUX fcis-29830 29 14 been be AUX fcis-29830 29 15 achieved achieve VERB fcis-29830 29 16 in in ADP fcis-29830 29 17 the the DET fcis-29830 29 18 research research NOUN fcis-29830 29 19 of of ADP fcis-29830 29 20 medical medical ADJ fcis-29830 29 21 image image NOUN fcis-29830 29 22 processing processing NOUN fcis-29830 29 23 [ [ X fcis-29830 29 24 12 12 NUM fcis-29830 29 25 , , PUNCT fcis-29830 29 26 13 13 NUM fcis-29830 29 27 ] ] PUNCT fcis-29830 29 28 . . PUNCT fcis-29830 30 1 u u X fcis-29830 30 2 - - PUNCT fcis-29830 30 3 net++ net++ X fcis-29830 30 4 , , PUNCT fcis-29830 30 5 as as ADP fcis-29830 30 6 a a DET fcis-29830 30 7 classical classical ADJ fcis-29830 30 8 convolutional convolutional ADJ fcis-29830 30 9 neural neural ADJ fcis-29830 30 10 network network NOUN fcis-29830 30 11 architecture architecture NOUN fcis-29830 30 12 for for ADP fcis-29830 30 13 image image NOUN fcis-29830 30 14 segmentation segmentation NOUN fcis-29830 30 15 tasks task NOUN fcis-29830 30 16 , , PUNCT fcis-29830 30 17 significantly significantly ADV fcis-29830 30 18 enhances enhance VERB fcis-29830 30 19 the the DET fcis-29830 30 20 network network NOUN fcis-29830 30 21 's 's PART fcis-29830 30 22 ability ability NOUN fcis-29830 30 23 in in ADP fcis-29830 30 24 multi multi ADJ fcis-29830 30 25 - - ADJ fcis-29830 30 26 scale scale ADJ fcis-29830 30 27 feature feature NOUN fcis-29830 30 28 fusion fusion NOUN fcis-29830 30 29 and and CCONJ fcis-29830 30 30 detail detail NOUN fcis-29830 30 31 recovery recovery NOUN fcis-29830 30 32 by by ADP fcis-29830 30 33 introducing introduce VERB fcis-29830 30 34 nested nested ADJ fcis-29830 30 35 bar bar NOUN fcis-29830 30 36 hopping hop VERB fcis-29830 30 37 connections connection NOUN fcis-29830 30 38 and and CCONJ fcis-29830 30 39 dense dense ADJ fcis-29830 30 40 hopping hop VERB fcis-29830 30 41 connection connection NOUN fcis-29830 30 42 structure structure NOUN fcis-29830 30 43 on on ADP fcis-29830 30 44 the the DET fcis-29830 30 45 basis basis NOUN fcis-29830 30 46 of of ADP fcis-29830 30 47 traditional traditional ADJ fcis-29830 30 48 u u NOUN fcis-29830 30 49 - - NOUN fcis-29830 30 50 net net NOUN fcis-29830 30 51 . . PUNCT fcis-29830 31 1 it it PRON fcis-29830 31 2 is be AUX fcis-29830 31 3 widely widely ADV fcis-29830 31 4 used use VERB fcis-29830 31 5 due due ADP fcis-29830 31 6 to to ADP fcis-29830 31 7 its its PRON fcis-29830 31 8 finer fine ADJ fcis-29830 31 9 structure structure NOUN fcis-29830 31 10 and and CCONJ fcis-29830 31 11 excellent excellent ADJ fcis-29830 31 12 segmentation segmentation NOUN fcis-29830 31 13 performance performance NOUN fcis-29830 31 14 . . PUNCT fcis-29830 32 1 in in ADP fcis-29830 32 2 this this DET fcis-29830 32 3 project project NOUN fcis-29830 32 4 , , PUNCT fcis-29830 32 5 we we PRON fcis-29830 32 6 apply apply VERB fcis-29830 32 7 it it PRON fcis-29830 32 8 to to ADP fcis-29830 32 9 an an DET fcis-29830 32 10 image image NOUN fcis-29830 32 11 classification classification NOUN fcis-29830 32 12 task task NOUN fcis-29830 32 13 to to PART fcis-29830 32 14 capture capture VERB fcis-29830 32 15 richer rich ADJ fcis-29830 32 16 and and CCONJ fcis-29830 32 17 more more ADV fcis-29830 32 18 diverse diverse ADJ fcis-29830 32 19 features feature NOUN fcis-29830 32 20 using use VERB fcis-29830 32 21 its its PRON fcis-29830 32 22 dense dense ADJ fcis-29830 32 23 connections connection NOUN fcis-29830 32 24 . . PUNCT fcis-29830 33 1 however however ADV fcis-29830 33 2 , , PUNCT fcis-29830 33 3 u u PROPN fcis-29830 33 4 - - NOUN fcis-29830 33 5 net++ net++ PROPN fcis-29830 33 6 is be AUX fcis-29830 33 7 in in ADP fcis-29830 33 8 the the DET fcis-29830 33 9 encoder encoder NOUN fcis-29830 33 10 part part NOUN fcis-29830 33 11 , , PUNCT fcis-29830 33 12 and and CCONJ fcis-29830 33 13 its its PRON fcis-29830 33 14 feature feature NOUN fcis-29830 33 15 extraction extraction NOUN fcis-29830 33 16 capability capability NOUN fcis-29830 33 17 is be AUX fcis-29830 33 18 insufficient insufficient ADJ fcis-29830 33 19 when when SCONJ fcis-29830 33 20 facing face VERB fcis-29830 33 21 complex complex ADJ fcis-29830 33 22 and and CCONJ fcis-29830 33 23 84 84 NUM fcis-29830 33 24 detail detail NOUN fcis-29830 33 25 - - PUNCT fcis-29830 33 26 rich rich ADJ fcis-29830 33 27 images image NOUN fcis-29830 33 28 . . PUNCT fcis-29830 34 1 we we PRON fcis-29830 34 2 design design VERB fcis-29830 34 3 an an DET fcis-29830 34 4 r r NOUN fcis-29830 34 5 - - PUNCT fcis-29830 34 6 as as NOUN fcis-29830 34 7 encoder encoder NOUN fcis-29830 34 8 module module NOUN fcis-29830 34 9 that that PRON fcis-29830 34 10 employs employ VERB fcis-29830 34 11 resnet50 resnet50 NOUN fcis-29830 34 12 as as ADP fcis-29830 34 13 its its PRON fcis-29830 34 14 backbone backbone NOUN fcis-29830 34 15 network network NOUN fcis-29830 34 16 , , PUNCT fcis-29830 34 17 which which PRON fcis-29830 34 18 utilizes utilize VERB fcis-29830 34 19 its its PRON fcis-29830 34 20 deep deep ADJ fcis-29830 34 21 residual residual ADJ fcis-29830 34 22 structure structure NOUN fcis-29830 34 23 to to PART fcis-29830 34 24 enhance enhance VERB fcis-29830 34 25 the the DET fcis-29830 34 26 feature feature NOUN fcis-29830 34 27 extraction extraction NOUN fcis-29830 34 28 capability capability NOUN fcis-29830 34 29 while while SCONJ fcis-29830 34 30 effectively effectively ADV fcis-29830 34 31 solving solve VERB fcis-29830 34 32 the the DET fcis-29830 34 33 problem problem NOUN fcis-29830 34 34 of of ADP fcis-29830 34 35 gradient gradient ADJ fcis-29830 34 36 vanishing vanishing NOUN fcis-29830 34 37 . . PUNCT fcis-29830 35 1 in in ADP fcis-29830 35 2 addition addition NOUN fcis-29830 35 3 , , PUNCT fcis-29830 35 4 we we PRON fcis-29830 35 5 also also ADV fcis-29830 35 6 introduce introduce VERB fcis-29830 35 7 a a DET fcis-29830 35 8 null null ADJ fcis-29830 35 9 - - PUNCT fcis-29830 35 10 space space NOUN fcis-29830 35 11 pyramid pyramid NOUN fcis-29830 35 12 pooling pool VERB fcis-29830 35 13 module module NOUN fcis-29830 35 14 for for ADP fcis-29830 35 15 enhancing enhance VERB fcis-29830 35 16 the the DET fcis-29830 35 17 model model NOUN fcis-29830 35 18 's 's PART fcis-29830 35 19 ability ability NOUN fcis-29830 35 20 to to PART fcis-29830 35 21 perceive perceive VERB fcis-29830 35 22 features feature NOUN fcis-29830 35 23 at at ADP fcis-29830 35 24 different different ADJ fcis-29830 35 25 scales scale NOUN fcis-29830 35 26 . . PUNCT fcis-29830 36 1 the the DET fcis-29830 36 2 high high ADJ fcis-29830 36 3 computational computational ADJ fcis-29830 36 4 complexity complexity NOUN fcis-29830 36 5 of of ADP fcis-29830 36 6 the the DET fcis-29830 36 7 structure structure NOUN fcis-29830 36 8 of of ADP fcis-29830 36 9 u u PROPN fcis-29830 36 10 - - NOUN fcis-29830 36 11 net++ net++ PROPN fcis-29830 36 12 leads lead VERB fcis-29830 36 13 to to ADP fcis-29830 36 14 slow slow ADJ fcis-29830 36 15 inference inference NOUN fcis-29830 36 16 , , PUNCT fcis-29830 36 17 especially especially ADV fcis-29830 36 18 when when SCONJ fcis-29830 36 19 applied apply VERB fcis-29830 36 20 on on ADP fcis-29830 36 21 resource resource NOUN fcis-29830 36 22 - - PUNCT fcis-29830 36 23 limited limit VERB fcis-29830 36 24 devices device NOUN fcis-29830 36 25 . . PUNCT fcis-29830 37 1 in in ADP fcis-29830 37 2 order order NOUN fcis-29830 37 3 to to PART fcis-29830 37 4 reduce reduce VERB fcis-29830 37 5 the the DET fcis-29830 37 6 computational computational ADJ fcis-29830 37 7 complexity complexity NOUN fcis-29830 37 8 and and CCONJ fcis-29830 37 9 improve improve VERB fcis-29830 37 10 the the DET fcis-29830 37 11 inference inference NOUN fcis-29830 37 12 efficiency efficiency NOUN fcis-29830 37 13 , , PUNCT fcis-29830 37 14 we we PRON fcis-29830 37 15 replace replace VERB fcis-29830 37 16 some some PRON fcis-29830 37 17 of of ADP fcis-29830 37 18 the the DET fcis-29830 37 19 traditional traditional ADJ fcis-29830 37 20 convolutions convolution NOUN fcis-29830 37 21 in in ADP fcis-29830 37 22 u u NOUN fcis-29830 37 23 - - NOUN fcis-29830 37 24 net++ net++ PROPN fcis-29830 37 25 with with ADP fcis-29830 37 26 depth depth NOUN fcis-29830 37 27 - - PUNCT fcis-29830 37 28 separable separable NOUN fcis-29830 37 29 convolutions convolution NOUN fcis-29830 37 30 . . PUNCT fcis-29830 38 1 2 2 X fcis-29830 38 2 . . X fcis-29830 38 3 related relate VERB fcis-29830 38 4 work work NOUN fcis-29830 38 5 : : PUNCT fcis-29830 38 6 2.1 2.1 NUM fcis-29830 38 7 . . PUNCT fcis-29830 39 1 data datum NOUN fcis-29830 39 2 sources source NOUN fcis-29830 39 3 : : PUNCT fcis-29830 39 4 the the DET fcis-29830 39 5 dataset dataset NOUN fcis-29830 39 6 in in ADP fcis-29830 39 7 this this DET fcis-29830 39 8 paper paper NOUN fcis-29830 39 9 is be AUX fcis-29830 39 10 derived derive VERB fcis-29830 39 11 from from ADP fcis-29830 39 12 the the DET fcis-29830 39 13 ai ai ADJ fcis-29830 39 14 algorithm algorithm NOUN fcis-29830 39 15 elite elite ADJ fcis-29830 39 16 challenge challenge NOUN fcis-29830 39 17 competition competition NOUN fcis-29830 39 18 dataset dataset NOUN fcis-29830 39 19 , , PUNCT fcis-29830 39 20 which which PRON fcis-29830 39 21 is be AUX fcis-29830 39 22 labeled label VERB fcis-29830 39 23 with with ADP fcis-29830 39 24 the the DET fcis-29830 39 25 assistance assistance NOUN fcis-29830 39 26 of of ADP fcis-29830 39 27 professional professional ADJ fcis-29830 39 28 physicians physician NOUN fcis-29830 39 29 to to PART fcis-29830 39 30 ensure ensure VERB fcis-29830 39 31 the the DET fcis-29830 39 32 accuracy accuracy NOUN fcis-29830 39 33 of of ADP fcis-29830 39 34 data datum NOUN fcis-29830 39 35 labeling labeling NOUN fcis-29830 39 36 . . PUNCT fcis-29830 40 1 the the DET fcis-29830 40 2 breast breast NOUN fcis-29830 40 3 feature feature NOUN fcis-29830 40 4 dataset dataset NOUN fcis-29830 40 5 contains contain VERB fcis-29830 40 6 four four NUM fcis-29830 40 7 types type NOUN fcis-29830 40 8 of of ADP fcis-29830 40 9 features feature NOUN fcis-29830 40 10 : : PUNCT fcis-29830 40 11 orientation orientation NOUN fcis-29830 40 12 , , PUNCT fcis-29830 40 13 edge edge NOUN fcis-29830 40 14 , , PUNCT fcis-29830 40 15 calcification calcification NOUN fcis-29830 40 16 and and CCONJ fcis-29830 40 17 shape shape NOUN fcis-29830 40 18 , , PUNCT fcis-29830 40 19 and and CCONJ fcis-29830 40 20 each each DET fcis-29830 40 21 type type NOUN fcis-29830 40 22 of of ADP fcis-29830 40 23 feature feature NOUN fcis-29830 40 24 is be AUX fcis-29830 40 25 represented represent VERB fcis-29830 40 26 by by ADP fcis-29830 40 27 a a DET fcis-29830 40 28 binary binary ADJ fcis-29830 40 29 label label NOUN fcis-29830 40 30 ( ( PUNCT fcis-29830 40 31 0 0 NUM fcis-29830 40 32 and and CCONJ fcis-29830 40 33 1 1 NUM fcis-29830 40 34 ) ) PUNCT fcis-29830 40 35 , , PUNCT fcis-29830 40 36 where where SCONJ fcis-29830 40 37 0 0 NUM fcis-29830 40 38 indicates indicate VERB fcis-29830 40 39 benign benign ADJ fcis-29830 40 40 features feature NOUN fcis-29830 40 41 and and CCONJ fcis-29830 40 42 1 1 NUM fcis-29830 40 43 indicates indicate VERB fcis-29830 40 44 malignant malignant ADJ fcis-29830 40 45 features feature NOUN fcis-29830 40 46 . . PUNCT fcis-29830 41 1 as as SCONJ fcis-29830 41 2 shown show VERB fcis-29830 41 3 in in ADP fcis-29830 41 4 table table NOUN fcis-29830 41 5 1 1 NUM fcis-29830 41 6 . . PUNCT fcis-29830 41 7 table table NOUN fcis-29830 41 8 1 1 NUM fcis-29830 41 9 . . PUNCT fcis-29830 41 10 demonstration demonstration NOUN fcis-29830 41 11 of of ADP fcis-29830 41 12 breast breast NOUN fcis-29830 41 13 characterization characterization NOUN fcis-29830 41 14 dataset dataset NOUN fcis-29830 41 15 direction direction NOUN fcis-29830 41 16 boundary boundary ADJ fcis-29830 41 17 calcification calcification NOUN fcis-29830 41 18 shape shape NOUN fcis-29830 41 19 benign benign ADJ fcis-29830 41 20 malignant malignant ADJ fcis-29830 41 21 descriptive descriptive ADJ fcis-29830 41 22 parallel parallel NOUN fcis-29830 41 23 : : PUNCT fcis-29830 41 24 the the DET fcis-29830 41 25 long long ADJ fcis-29830 41 26 axis axis NOUN fcis-29830 41 27 of of ADP fcis-29830 41 28 the the DET fcis-29830 41 29 mass mass NOUN fcis-29830 41 30 is be AUX fcis-29830 41 31 parallel parallel ADJ fcis-29830 41 32 to to ADP fcis-29830 41 33 the the DET fcis-29830 41 34 skin skin NOUN fcis-29830 41 35 , , PUNCT fcis-29830 41 36 i.e. i.e. X fcis-29830 41 37 horizontal horizontal ADJ fcis-29830 41 38 position position NOUN fcis-29830 41 39 . . PUNCT fcis-29830 42 1 non non ADJ fcis-29830 42 2 - - ADJ fcis-29830 42 3 parallel parallel ADJ fcis-29830 42 4 : : PUNCT fcis-29830 42 5 the the DET fcis-29830 42 6 anteriorposterior anteriorposterior ADJ fcis-29830 42 7 diameter diameter NOUN fcis-29830 42 8 of of ADP fcis-29830 42 9 the the DET fcis-29830 42 10 mass mass NOUN fcis-29830 42 11 is be AUX fcis-29830 42 12 larger large ADJ fcis-29830 42 13 than than ADP fcis-29830 42 14 the the DET fcis-29830 42 15 transverse transverse NOUN fcis-29830 42 16 diameter diameter NOUN fcis-29830 42 17 , , PUNCT fcis-29830 42 18 i.e. i.e. X fcis-29830 42 19 , , PUNCT fcis-29830 42 20 vertical vertical ADJ fcis-29830 42 21 position position NOUN fcis-29830 42 22 . . PUNCT fcis-29830 43 1 luminescence luminescence NOUN fcis-29830 43 2 : : PUNCT fcis-29830 43 3 the the DET fcis-29830 43 4 mass mass NOUN fcis-29830 43 5 has have VERB fcis-29830 43 6 well well ADV fcis-29830 43 7 - - PUNCT fcis-29830 43 8 defined define VERB fcis-29830 43 9 borders border NOUN fcis-29830 43 10 and and CCONJ fcis-29830 43 11 there there PRON fcis-29830 43 12 is be VERB fcis-29830 43 13 a a DET fcis-29830 43 14 distinct distinct ADJ fcis-29830 43 15 mutation mutation NOUN fcis-29830 43 16 between between ADP fcis-29830 43 17 the the DET fcis-29830 43 18 lesion lesion NOUN fcis-29830 43 19 and and CCONJ fcis-29830 43 20 the the DET fcis-29830 43 21 surrounding surround VERB fcis-29830 43 22 tissue tissue NOUN fcis-29830 43 23 . . PUNCT fcis-29830 44 1 unglossy unglossy NOUN fcis-29830 44 2 : : PUNCT fcis-29830 44 3 the the DET fcis-29830 44 4 mass mass NOUN fcis-29830 44 5 has have AUX fcis-29830 44 6 blurred blur VERB fcis-29830 44 7 borders border NOUN fcis-29830 44 8 and and CCONJ fcis-29830 44 9 may may AUX fcis-29830 44 10 exhibit exhibit VERB fcis-29830 44 11 a a DET fcis-29830 44 12 hyperechoic hyperechoic ADJ fcis-29830 44 13 halo halo NOUN fcis-29830 44 14 , , PUNCT fcis-29830 44 15 angularity angularity NOUN fcis-29830 44 16 , , PUNCT fcis-29830 44 17 microfoliation microfoliation NOUN fcis-29830 44 18 , , PUNCT fcis-29830 44 19 or or CCONJ fcis-29830 44 20 burr burr NOUN fcis-29830 44 21 - - PUNCT fcis-29830 44 22 like like ADJ fcis-29830 44 23 appearance appearance NOUN fcis-29830 44 24 . . PUNCT fcis-29830 45 1 calcification calcification NOUN fcis-29830 45 2 : : PUNCT fcis-29830 45 3 localized localized ADJ fcis-29830 45 4 areas area NOUN fcis-29830 45 5 of of ADP fcis-29830 45 6 highlighting highlighting NOUN fcis-29830 45 7 ( ( PUNCT fcis-29830 45 8 calcified calcified ADJ fcis-29830 45 9 spots spot NOUN fcis-29830 45 10 ) ) PUNCT fcis-29830 45 11 are be AUX fcis-29830 45 12 visible visible ADJ fcis-29830 45 13 in in ADP fcis-29830 45 14 the the DET fcis-29830 45 15 ultrasound ultrasound NOUN fcis-29830 45 16 image image NOUN fcis-29830 45 17 no no DET fcis-29830 45 18 calcification calcification NOUN fcis-29830 45 19 : : PUNCT fcis-29830 45 20 no no DET fcis-29830 45 21 visible visible ADJ fcis-29830 45 22 areas area NOUN fcis-29830 45 23 of of ADP fcis-29830 45 24 highlighting highlight VERB fcis-29830 45 25 in in ADP fcis-29830 45 26 the the DET fcis-29830 45 27 ultrasound ultrasound NOUN fcis-29830 45 28 image image NOUN fcis-29830 45 29 . . PUNCT fcis-29830 46 1 regular regular ADV fcis-29830 46 2 : : PUNCT fcis-29830 46 3 the the DET fcis-29830 46 4 mass mass NOUN fcis-29830 46 5 is be AUX fcis-29830 46 6 round round ADJ fcis-29830 46 7 , , PUNCT fcis-29830 46 8 oval oval NOUN fcis-29830 46 9 or or CCONJ fcis-29830 46 10 large large ADJ fcis-29830 46 11 lobulated lobulate VERB fcis-29830 46 12 . . PUNCT fcis-29830 47 1 irregular irregular ADJ fcis-29830 47 2 : : PUNCT fcis-29830 47 3 the the DET fcis-29830 47 4 shape shape NOUN fcis-29830 47 5 of of ADP fcis-29830 47 6 the the DET fcis-29830 47 7 mass mass NOUN fcis-29830 47 8 is be AUX fcis-29830 47 9 neither neither CCONJ fcis-29830 47 10 round round ADJ fcis-29830 47 11 nor nor CCONJ fcis-29830 47 12 oval oval NOUN fcis-29830 47 13 . . PUNCT fcis-29830 48 1 2.2 2.2 NUM fcis-29830 48 2 . . PUNCT fcis-29830 48 3 data datum NOUN fcis-29830 48 4 pre pre VERB fcis-29830 48 5 - - VERB fcis-29830 48 6 processing process VERB fcis-29830 48 7 the the DET fcis-29830 48 8 images image NOUN fcis-29830 48 9 of of ADP fcis-29830 48 10 different different ADJ fcis-29830 48 11 sizes size NOUN fcis-29830 48 12 of of ADP fcis-29830 48 13 feature feature NOUN fcis-29830 48 14 training training NOUN fcis-29830 48 15 set set NOUN fcis-29830 48 16 are be AUX fcis-29830 48 17 adjusted adjust VERB fcis-29830 48 18 to to ADP fcis-29830 48 19 224 224 NUM fcis-29830 48 20 * * NUM fcis-29830 48 21 224 224 NUM fcis-29830 48 22 before before SCONJ fcis-29830 48 23 the the DET fcis-29830 48 24 experiment experiment NOUN fcis-29830 48 25 to to PART fcis-29830 48 26 facilitate facilitate VERB fcis-29830 48 27 the the DET fcis-29830 48 28 training training NOUN fcis-29830 48 29 and and CCONJ fcis-29830 48 30 reduce reduce VERB fcis-29830 48 31 the the DET fcis-29830 48 32 amount amount NOUN fcis-29830 48 33 of of ADP fcis-29830 48 34 computation computation NOUN fcis-29830 48 35 at at ADP fcis-29830 48 36 the the DET fcis-29830 48 37 same same ADJ fcis-29830 48 38 time time NOUN fcis-29830 48 39 . . PUNCT fcis-29830 49 1 and and CCONJ fcis-29830 49 2 all all DET fcis-29830 49 3 the the DET fcis-29830 49 4 image image NOUN fcis-29830 49 5 values value NOUN fcis-29830 49 6 are be AUX fcis-29830 49 7 normalized normalize VERB fcis-29830 49 8 to to PART fcis-29830 49 9 improve improve VERB fcis-29830 49 10 the the DET fcis-29830 49 11 algorithm algorithm NOUN fcis-29830 49 12 accuracy accuracy NOUN fcis-29830 49 13 and and CCONJ fcis-29830 49 14 accelerate accelerate VERB fcis-29830 49 15 the the DET fcis-29830 49 16 convergence convergence NOUN fcis-29830 49 17 speed speed NOUN fcis-29830 49 18 of of ADP fcis-29830 49 19 the the DET fcis-29830 49 20 algorithm algorithm NOUN fcis-29830 49 21 . . PUNCT fcis-29830 50 1 the the DET fcis-29830 50 2 images image NOUN fcis-29830 50 3 are be AUX fcis-29830 50 4 randomly randomly ADV fcis-29830 50 5 rotated rotate VERB fcis-29830 50 6 by by ADP fcis-29830 50 7 90 90 NUM fcis-29830 50 8 ° ° NOUN fcis-29830 50 9 , , PUNCT fcis-29830 50 10 and and CCONJ fcis-29830 50 11 then then ADV fcis-29830 50 12 some some PRON fcis-29830 50 13 of of ADP fcis-29830 50 14 the the DET fcis-29830 50 15 images image NOUN fcis-29830 50 16 are be AUX fcis-29830 50 17 vertically vertically ADV fcis-29830 50 18 flipped flip VERB fcis-29830 50 19 and and CCONJ fcis-29830 50 20 horizontally horizontally ADV fcis-29830 50 21 flipped flip VERB fcis-29830 50 22 . . PUNCT fcis-29830 51 1 it it PRON fcis-29830 51 2 helps help VERB fcis-29830 51 3 to to PART fcis-29830 51 4 enhance enhance VERB fcis-29830 51 5 the the DET fcis-29830 51 6 diversity diversity NOUN fcis-29830 51 7 of of ADP fcis-29830 51 8 the the DET fcis-29830 51 9 data datum NOUN fcis-29830 51 10 and and CCONJ fcis-29830 51 11 reduce reduce VERB fcis-29830 51 12 the the DET fcis-29830 51 13 overfitting overfitting NOUN fcis-29830 51 14 caused cause VERB fcis-29830 51 15 by by ADP fcis-29830 51 16 different different ADJ fcis-29830 51 17 orientations orientation NOUN fcis-29830 51 18 . . PUNCT fcis-29830 52 1 2.3 2.3 NUM fcis-29830 52 2 . . PUNCT fcis-29830 53 1 network network NOUN fcis-29830 53 2 model model NOUN fcis-29830 53 3 dsa dsa PROPN fcis-29830 53 4 - - PUNCT fcis-29830 53 5 u++ u++ NOUN fcis-29830 53 6 we we PRON fcis-29830 53 7 propose propose VERB fcis-29830 53 8 an an DET fcis-29830 53 9 improved improved ADJ fcis-29830 53 10 deep deep ADJ fcis-29830 53 11 learning learning NOUN fcis-29830 53 12 model model NOUN fcis-29830 53 13 based base VERB fcis-29830 53 14 on on ADP fcis-29830 53 15 resnet resnet NOUN fcis-29830 53 16 [ [ X fcis-29830 53 17 1 1 NUM fcis-29830 53 18 ] ] PUNCT fcis-29830 53 19 and and CCONJ fcis-29830 53 20 unet++ unet++ PRON fcis-29830 54 1 [ [ X fcis-29830 54 2 2 2 NUM fcis-29830 54 3 ] ] PUNCT fcis-29830 54 4 for for ADP fcis-29830 54 5 feature feature NOUN fcis-29830 54 6 classification classification NOUN fcis-29830 54 7 of of ADP fcis-29830 54 8 ultrasound ultrasound ADJ fcis-29830 54 9 breast breast NOUN fcis-29830 54 10 images image NOUN fcis-29830 54 11 . . PUNCT fcis-29830 55 1 specifically specifically ADV fcis-29830 55 2 , , PUNCT fcis-29830 55 3 the the DET fcis-29830 55 4 dsa dsa NOUN fcis-29830 55 5 - - PUNCT fcis-29830 55 6 u++ u++ PROPN fcis-29830 55 7 model model NOUN fcis-29830 55 8 first first ADV fcis-29830 55 9 uses use VERB fcis-29830 55 10 the the DET fcis-29830 55 11 r r NOUN fcis-29830 55 12 - - PUNCT fcis-29830 55 13 as as NOUN fcis-29830 55 14 module module NOUN fcis-29830 55 15 instead instead ADV fcis-29830 55 16 of of ADP fcis-29830 55 17 the the DET fcis-29830 55 18 original original ADJ fcis-29830 55 19 encoder encoder NOUN fcis-29830 55 20 of of ADP fcis-29830 55 21 u u PROPN fcis-29830 55 22 - - NOUN fcis-29830 55 23 net++ net++ PROPN fcis-29830 55 24 for for ADP fcis-29830 55 25 feature feature NOUN fcis-29830 55 26 extraction extraction NOUN fcis-29830 55 27 , , PUNCT fcis-29830 55 28 and and CCONJ fcis-29830 55 29 the the DET fcis-29830 55 30 r r NOUN fcis-29830 55 31 - - PUNCT fcis-29830 55 32 as as NOUN fcis-29830 55 33 module module NOUN fcis-29830 55 34 employs employ VERB fcis-29830 55 35 resnet50 resnet50 NOUN fcis-29830 55 36 as as ADV fcis-29830 55 37 well well ADV fcis-29830 55 38 as as ADP fcis-29830 55 39 the the DET fcis-29830 55 40 aspp aspp NOUN fcis-29830 55 41 module module NOUN fcis-29830 55 42 to to PART fcis-29830 55 43 adequately adequately ADV fcis-29830 55 44 extract extract VERB fcis-29830 55 45 the the DET fcis-29830 55 46 multi multi ADJ fcis-29830 55 47 - - ADJ fcis-29830 55 48 scale scale ADJ fcis-29830 55 49 features feature NOUN fcis-29830 55 50 of of ADP fcis-29830 55 51 the the DET fcis-29830 55 52 image image NOUN fcis-29830 55 53 . . PUNCT fcis-29830 56 1 resnet resnet NOUN fcis-29830 56 2 is be AUX fcis-29830 56 3 a a DET fcis-29830 56 4 network network NOUN fcis-29830 56 5 model model NOUN fcis-29830 56 6 that that PRON fcis-29830 56 7 has have AUX fcis-29830 56 8 been be AUX fcis-29830 56 9 applied apply VERB fcis-29830 56 10 to to ADP fcis-29830 56 11 visual visual ADJ fcis-29830 56 12 tasks task NOUN fcis-29830 56 13 such such ADJ fcis-29830 56 14 as as ADP fcis-29830 56 15 image image NOUN fcis-29830 56 16 classification classification NOUN fcis-29830 56 17 and and CCONJ fcis-29830 56 18 semantic semantic ADJ fcis-29830 56 19 segmentation segmentation NOUN fcis-29830 56 20 . . PUNCT fcis-29830 57 1 its its PRON fcis-29830 57 2 multiple multiple ADJ fcis-29830 57 3 stacking stacking NOUN fcis-29830 57 4 by by ADP fcis-29830 57 5 a a DET fcis-29830 57 6 special special ADJ fcis-29830 57 7 residual residual ADJ fcis-29830 57 8 module module NOUN fcis-29830 57 9 can can AUX fcis-29830 57 10 avoid avoid VERB fcis-29830 57 11 the the DET fcis-29830 57 12 situation situation NOUN fcis-29830 57 13 of of ADP fcis-29830 57 14 gradient gradient ADJ fcis-29830 57 15 vanishing vanishing NOUN fcis-29830 57 16 or or CCONJ fcis-29830 57 17 gradient gradient NOUN fcis-29830 57 18 explosion explosion NOUN fcis-29830 57 19 , , PUNCT fcis-29830 57 20 but but CCONJ fcis-29830 57 21 also also ADV fcis-29830 57 22 can can AUX fcis-29830 57 23 learn learn VERB fcis-29830 57 24 deep deep ADJ fcis-29830 57 25 features feature NOUN fcis-29830 57 26 in in ADP fcis-29830 57 27 a a DET fcis-29830 57 28 short short ADJ fcis-29830 57 29 time time NOUN fcis-29830 57 30 . . PUNCT fcis-29830 58 1 the the DET fcis-29830 58 2 r r NOUN fcis-29830 58 3 - - PUNCT fcis-29830 58 4 as as NOUN fcis-29830 58 5 module module NOUN fcis-29830 58 6 is be AUX fcis-29830 58 7 shown show VERB fcis-29830 58 8 in in ADP fcis-29830 58 9 fig fig NOUN fcis-29830 58 10 1 1 NUM fcis-29830 58 11 . . PUNCT fcis-29830 59 1 the the DET fcis-29830 59 2 r r NOUN fcis-29830 59 3 - - PUNCT fcis-29830 59 4 as as ADP fcis-29830 59 5 module module NOUN fcis-29830 59 6 contains contain VERB fcis-29830 59 7 1 1 NUM fcis-29830 59 8 convolution convolution NOUN fcis-29830 59 9 block block NOUN fcis-29830 59 10 , , PUNCT fcis-29830 59 11 4 4 NUM fcis-29830 59 12 residual residual ADJ fcis-29830 59 13 blocks block NOUN fcis-29830 59 14 , , PUNCT fcis-29830 59 15 and and CCONJ fcis-29830 59 16 5 5 NUM fcis-29830 59 17 outputs output NOUN fcis-29830 59 18 . . PUNCT fcis-29830 60 1 the the DET fcis-29830 60 2 convolution convolution NOUN fcis-29830 60 3 block block NOUN fcis-29830 60 4 is be AUX fcis-29830 60 5 composed compose VERB fcis-29830 60 6 of of ADP fcis-29830 60 7 1 1 NUM fcis-29830 60 8 convolutional convolutional ADJ fcis-29830 60 9 layer layer NOUN fcis-29830 60 10 of of ADP fcis-29830 60 11 size size NOUN fcis-29830 60 12 7×7 7×7 NOUN fcis-29830 60 13 , , PUNCT fcis-29830 60 14 batch batch VERB fcis-29830 60 15 normalization normalization NOUN fcis-29830 60 16 layer layer NOUN fcis-29830 60 17 , , PUNCT fcis-29830 60 18 activation activation NOUN fcis-29830 60 19 layer layer NOUN fcis-29830 60 20 , , PUNCT fcis-29830 60 21 and and CCONJ fcis-29830 60 22 3×3 3×3 NUM fcis-29830 60 23 maximum maximum ADJ fcis-29830 60 24 pooling pool VERB fcis-29830 60 25 layer layer NOUN fcis-29830 60 26 . . PUNCT fcis-29830 61 1 the the DET fcis-29830 61 2 convolutional convolutional ADJ fcis-29830 61 3 residual residual ADJ fcis-29830 61 4 blocks block NOUN fcis-29830 61 5 are be AUX fcis-29830 61 6 all all PRON fcis-29830 61 7 composed compose VERB fcis-29830 61 8 of of ADP fcis-29830 61 9 convolutional convolutional ADJ fcis-29830 61 10 kernels kernel NOUN fcis-29830 61 11 of of ADP fcis-29830 61 12 size size NOUN fcis-29830 61 13 1 1 NUM fcis-29830 61 14 × × NOUN fcis-29830 61 15 1 1 NUM fcis-29830 61 16 , , PUNCT fcis-29830 61 17 3 3 NUM fcis-29830 61 18 × × NOUN fcis-29830 61 19 3 3 NUM fcis-29830 61 20 , , PUNCT fcis-29830 61 21 and and CCONJ fcis-29830 61 22 1 1 NUM fcis-29830 61 23 × × NOUN fcis-29830 61 24 1 1 NUM fcis-29830 61 25 . . PUNCT fcis-29830 62 1 the the DET fcis-29830 62 2 residual residual ADJ fcis-29830 62 3 formula formula NOUN fcis-29830 62 4 is be AUX fcis-29830 62 5 shown show VERB fcis-29830 62 6 in in ADP fcis-29830 62 7 ( ( PUNCT fcis-29830 62 8 1 1 NUM fcis-29830 62 9 ) ) PUNCT fcis-29830 62 10 . . PUNCT fcis-29830 63 1 ( ( PUNCT fcis-29830 63 2 ) ) PUNCT fcis-29830 63 3 ( ( PUNCT fcis-29830 63 4 ) ) PUNCT fcis-29830 63 5 h h NOUN fcis-29830 64 1 x x PUNCT fcis-29830 64 2 f f PROPN fcis-29830 64 3 x x SYM fcis-29830 64 4 x x PROPN fcis-29830 64 5   X fcis-29830 64 6 ( ( PUNCT fcis-29830 64 7 1 1 X fcis-29830 64 8 ) ) PUNCT fcis-29830 64 9 in in ADP fcis-29830 64 10 the the DET fcis-29830 64 11 formula formula NOUN fcis-29830 64 12 : : PUNCT fcis-29830 65 1 f f X fcis-29830 65 2 ( ( PUNCT fcis-29830 65 3 ) ) PUNCT fcis-29830 65 4 x x PRON fcis-29830 65 5 denotes denote VERB fcis-29830 65 6 the the DET fcis-29830 65 7 output output NOUN fcis-29830 65 8 of of ADP fcis-29830 65 9 the the DET fcis-29830 65 10 convolutional convolutional ADJ fcis-29830 65 11 layer layer NOUN fcis-29830 65 12 and and CCONJ fcis-29830 65 13 represents represent VERB fcis-29830 65 14 the the DET fcis-29830 65 15 output output NOUN fcis-29830 65 16 obtained obtain VERB fcis-29830 65 17 after after ADP fcis-29830 65 18 convolution convolution NOUN fcis-29830 65 19 ; ; PUNCT fcis-29830 65 20 x x X fcis-29830 65 21 is be AUX fcis-29830 65 22 the the DET fcis-29830 65 23 input input NOUN fcis-29830 65 24 to to ADP fcis-29830 65 25 the the DET fcis-29830 65 26 residual residual ADJ fcis-29830 65 27 block block NOUN fcis-29830 65 28 , , PUNCT fcis-29830 65 29 from from ADP fcis-29830 65 30 the the DET fcis-29830 65 31 feature feature NOUN fcis-29830 65 32 map map NOUN fcis-29830 65 33 of of ADP fcis-29830 65 34 the the DET fcis-29830 65 35 previous previous ADJ fcis-29830 65 36 layer layer NOUN fcis-29830 65 37 ; ; PUNCT fcis-29830 65 38 ( ( PUNCT fcis-29830 65 39 ) ) PUNCT fcis-29830 65 40 h h NOUN fcis-29830 65 41 x x NOUN fcis-29830 65 42 are be AUX fcis-29830 65 43 the the DET fcis-29830 65 44 outputs output NOUN fcis-29830 65 45 of of ADP fcis-29830 65 46 the the DET fcis-29830 65 47 residual residual ADJ fcis-29830 65 48 blocks.r blocks.r NOUN fcis-29830 65 49 - - PUNCT fcis-29830 65 50 as2 as2 NOUN fcis-29830 65 51 , , PUNCT fcis-29830 65 52 r r PROPN fcis-29830 65 53 - - PUNCT fcis-29830 65 54 as3 as3 PROPN fcis-29830 65 55 , , PUNCT fcis-29830 65 56 r r NOUN fcis-29830 65 57 - - PUNCT fcis-29830 65 58 as4 as4 PROPN fcis-29830 65 59 , , PUNCT fcis-29830 65 60 and and CCONJ fcis-29830 65 61 r r X fcis-29830 65 62 - - PUNCT fcis-29830 65 63 as5 as5 NOUN fcis-29830 65 64 are be AUX fcis-29830 65 65 obtained obtain VERB fcis-29830 65 66 from from ADP fcis-29830 65 67 the the DET fcis-29830 65 68 outputs output NOUN fcis-29830 65 69 of of ADP fcis-29830 65 70 3 3 NUM fcis-29830 65 71 , , PUNCT fcis-29830 65 72 4 4 NUM fcis-29830 65 73 , , PUNCT fcis-29830 65 74 6 6 NUM fcis-29830 65 75 , , PUNCT fcis-29830 65 76 and and CCONJ fcis-29830 65 77 3 3 NUM fcis-29830 65 78 residual residual ADJ fcis-29830 65 79 blocks block NOUN fcis-29830 65 80 , , PUNCT fcis-29830 65 81 respectively respectively ADV fcis-29830 65 82 . . PUNCT fcis-29830 65 83 85 85 NUM fcis-29830 65 84 1×1 1×1 NUM fcis-29830 65 85     SPACE fcis-29830 65 86 conv conv ADJ fcis-29830 65 87 3×3 3×3 NUM fcis-29830 65 88     SPACE fcis-29830 65 89 conv conv ADJ fcis-29830 65 90 1×1 1×1 ADJ fcis-29830 65 91     SPACE fcis-29830 65 92 conv conv ADJ fcis-29830 65 93 w w PROPN fcis-29830 65 94 h h NOUN fcis-29830 65 95 input input NOUN fcis-29830 65 96 maxpool maxpool NOUN fcis-29830 65 97     SPACE fcis-29830 65 98 1×1 1×1 NUM fcis-29830 65 99     SPACE fcis-29830 65 100 conv conv ADJ fcis-29830 65 101 3×3 3×3 NUM fcis-29830 65 102     SPACE fcis-29830 65 103 conv conv ADJ fcis-29830 65 104 1×1 1×1 ADJ fcis-29830 65 105     SPACE fcis-29830 65 106 conv conv ADJ fcis-29830 65 107 ×3 ×3 NOUN fcis-29830 65 108 ×4 ×4 NOUN fcis-29830 65 109 1×1 1×1 NUM fcis-29830 65 110     SPACE fcis-29830 65 111 conv conv ADJ fcis-29830 65 112 3×3 3×3 NUM fcis-29830 65 113     SPACE fcis-29830 65 114 conv conv ADJ fcis-29830 65 115 1×1 1×1 ADJ fcis-29830 65 116     SPACE fcis-29830 65 117 conv conv ADJ fcis-29830 65 118 1×1 1×1 ADJ fcis-29830 65 119     SPACE fcis-29830 65 120 conv conv ADJ fcis-29830 65 121 3×3 3×3 NUM fcis-29830 65 122     SPACE fcis-29830 65 123 conv conv ADJ fcis-29830 65 124 1×1 1×1 ADJ fcis-29830 65 125     SPACE fcis-29830 65 126 conv conv ADJ fcis-29830 65 127 ×3 ×3 NOUN fcis-29830 65 128 ×6 ×6 ADJ fcis-29830 65 129         SPACE fcis-29830 65 130 7×7 7×7 NUM fcis-29830 65 131     SPACE fcis-29830 65 132 conv conv ADJ fcis-29830 65 133 batchnorm batchnorm NOUN fcis-29830 65 134 relu relu NOUN fcis-29830 65 135     SPACE fcis-29830 65 136 r‐as1 r‐as1 NOUN fcis-29830 65 137 r‐as2 r‐as2 ADJ fcis-29830 65 138 r‐as3 r‐as3 NOUN fcis-29830 65 139 r‐as4 r‐as4 PROPN fcis-29830 65 140 r‐as5 r‐as5 PROPN fcis-29830 65 141 aspp aspp PROPN fcis-29830 65 142 1c 1c NUM fcis-29830 65 143 2c 2c NOUN fcis-29830 65 144 3c 3c NUM fcis-29830 65 145 4c 4c NOUN fcis-29830 65 146 5c 5c NUM fcis-29830 65 147 fig fig NOUN fcis-29830 65 148 1 1 NUM fcis-29830 65 149 . . PUNCT fcis-29830 66 1 r r X fcis-29830 66 2 - - PUNCT fcis-29830 66 3 as as NOUN fcis-29830 66 4 module module NOUN fcis-29830 66 5 in in ADP fcis-29830 66 6 our our PRON fcis-29830 66 7 model model NOUN fcis-29830 66 8 dsa dsa NOUN fcis-29830 66 9 - - PUNCT fcis-29830 66 10 u++ u++ NOUN fcis-29830 66 11 , , PUNCT fcis-29830 66 12 r r NOUN fcis-29830 66 13 - - PUNCT fcis-29830 66 14 as1 as1 NOUN fcis-29830 66 15 is be AUX fcis-29830 66 16 rich rich ADJ fcis-29830 66 17 in in ADP fcis-29830 66 18 spatial spatial ADJ fcis-29830 66 19 detail detail NOUN fcis-29830 66 20 information information NOUN fcis-29830 66 21 due due ADP fcis-29830 66 22 to to ADP fcis-29830 66 23 its its PRON fcis-29830 66 24 inclusion inclusion NOUN fcis-29830 66 25 . . PUNCT fcis-29830 67 1 so so ADV fcis-29830 67 2 , , PUNCT fcis-29830 67 3 the the DET fcis-29830 67 4 output output NOUN fcis-29830 67 5 feature feature NOUN fcis-29830 67 6 map map NOUN fcis-29830 67 7 ras1 ras1 NOUN fcis-29830 67 8 is be AUX fcis-29830 67 9 obtained obtain VERB fcis-29830 67 10 after after ADP fcis-29830 67 11 multi multi ADJ fcis-29830 67 12 - - ADJ fcis-29830 67 13 scale scale ADJ fcis-29830 67 14 feature feature NOUN fcis-29830 67 15 extraction extraction NOUN fcis-29830 67 16 by by ADP fcis-29830 67 17 applying apply VERB fcis-29830 67 18 the the DET fcis-29830 67 19 null null ADJ fcis-29830 67 20 - - PUNCT fcis-29830 67 21 space space NOUN fcis-29830 67 22 convolutional convolutional ADJ fcis-29830 67 23 pooling pool VERB fcis-29830 67 24 pyramid pyramid NOUN fcis-29830 67 25 , , PUNCT fcis-29830 67 26 together together ADV fcis-29830 67 27 with with ADP fcis-29830 67 28 the the DET fcis-29830 67 29 output output NOUN fcis-29830 67 30 of of ADP fcis-29830 67 31 the the DET fcis-29830 67 32 four four NUM fcis-29830 67 33 feature feature NOUN fcis-29830 67 34 maps map NOUN fcis-29830 67 35 r r NOUN fcis-29830 67 36 - - PUNCT fcis-29830 67 37 as as ADP fcis-29830 67 38 via via ADP fcis-29830 67 39 four four NUM fcis-29830 67 40 residual residual ADJ fcis-29830 67 41 blocks block NOUN fcis-29830 67 42 , , PUNCT fcis-29830 67 43 respectively respectively ADV fcis-29830 67 44 , , PUNCT fcis-29830 67 45 constitutes constitute VERB fcis-29830 67 46 the the DET fcis-29830 67 47 encoder encoder NOUN fcis-29830 67 48 part part NOUN fcis-29830 67 49 of of ADP fcis-29830 67 50 u u PROPN fcis-29830 67 51 - - PROPN fcis-29830 67 52 net++ net++ PROPN fcis-29830 67 53 . . PUNCT fcis-29830 68 1 subsequently subsequently ADV fcis-29830 68 2 , , PUNCT fcis-29830 68 3 in in ADP fcis-29830 68 4 the the DET fcis-29830 68 5 decoder decoder NOUN fcis-29830 68 6 part part NOUN fcis-29830 68 7 , , PUNCT fcis-29830 68 8 we we PRON fcis-29830 68 9 improve improve VERB fcis-29830 68 10 the the DET fcis-29830 68 11 network network NOUN fcis-29830 68 12 based base VERB fcis-29830 68 13 on on ADP fcis-29830 68 14 the the DET fcis-29830 68 15 u u NOUN fcis-29830 68 16 - - NOUN fcis-29830 68 17 net++ net++ ADJ fcis-29830 68 18 architecture architecture NOUN fcis-29830 68 19 by by ADP fcis-29830 68 20 replacing replace VERB fcis-29830 68 21 some some PRON fcis-29830 68 22 of of ADP fcis-29830 68 23 the the DET fcis-29830 68 24 traditional traditional ADJ fcis-29830 68 25 convolution convolution NOUN fcis-29830 68 26 operations operation NOUN fcis-29830 68 27 with with ADP fcis-29830 68 28 depthseparable depthseparable ADJ fcis-29830 68 29 convolutions convolution NOUN fcis-29830 68 30 , , PUNCT fcis-29830 68 31 thus thus ADV fcis-29830 68 32 effectively effectively ADV fcis-29830 68 33 reducing reduce VERB fcis-29830 68 34 the the DET fcis-29830 68 35 number number NOUN fcis-29830 68 36 of of ADP fcis-29830 68 37 model model NOUN fcis-29830 68 38 parameters parameter NOUN fcis-29830 68 39 and and CCONJ fcis-29830 68 40 computational computational ADJ fcis-29830 68 41 complexity complexity NOUN fcis-29830 68 42 while while SCONJ fcis-29830 68 43 maintaining maintain VERB fcis-29830 68 44 the the DET fcis-29830 68 45 feature feature NOUN fcis-29830 68 46 expression expression NOUN fcis-29830 68 47 capability capability NOUN fcis-29830 68 48 . . PUNCT fcis-29830 69 1 five five NUM fcis-29830 69 2 feature feature NOUN fcis-29830 69 3 maps map NOUN fcis-29830 69 4 of of ADP fcis-29830 69 5 different different ADJ fcis-29830 69 6 sizes size NOUN fcis-29830 69 7 r r NOUN fcis-29830 69 8 - - PUNCT fcis-29830 69 9 as as ADV fcis-29830 69 10 replace replace VERB fcis-29830 69 11 the the DET fcis-29830 69 12 original original ADJ fcis-29830 69 13 encoder encoder NOUN fcis-29830 69 14 output output NOUN fcis-29830 69 15 of of ADP fcis-29830 69 16 u u PROPN fcis-29830 69 17 - - PROPN fcis-29830 69 18 net++ net++ PROPN fcis-29830 69 19 , , PUNCT fcis-29830 69 20 which which PRON fcis-29830 69 21 is be AUX fcis-29830 69 22 then then ADV fcis-29830 69 23 decoded decode VERB fcis-29830 69 24 . . PUNCT fcis-29830 70 1 , , PUNCT fcis-29830 70 2 i i PRON fcis-29830 70 3 jx jx PROPN fcis-29830 70 4 is be AUX fcis-29830 70 5 a a DET fcis-29830 70 6 convolutional convolutional ADJ fcis-29830 70 7 unit unit NOUN fcis-29830 70 8 . . PUNCT fcis-29830 71 1 i i PRON fcis-29830 71 2 denotes denote VERB fcis-29830 71 3 the the DET fcis-29830 71 4 i i PROPN fcis-29830 71 5 - - PUNCT fcis-29830 71 6 th th VERB fcis-29830 71 7 downsampling downsample VERB fcis-29830 71 8 layer layer NOUN fcis-29830 71 9 , , PUNCT fcis-29830 71 10 j j PROPN fcis-29830 71 11 represents represent VERB fcis-29830 71 12 the the DET fcis-29830 71 13 jth jth PROPN fcis-29830 71 14 convolutional convolutional ADJ fcis-29830 71 15 layer layer NOUN fcis-29830 71 16 in in ADP fcis-29830 71 17 the the DET fcis-29830 71 18 current current ADJ fcis-29830 71 19 jumpconnected jumpconnecte VERB fcis-29830 71 20 layer layer NOUN fcis-29830 71 21 , , PUNCT fcis-29830 71 22 suppose suppose VERB fcis-29830 71 23 , , PUNCT fcis-29830 71 24 i i PRON fcis-29830 71 25 jx jx PROPN fcis-29830 71 26 is be AUX fcis-29830 71 27 the the DET fcis-29830 71 28 output output NOUN fcis-29830 71 29 of of ADP fcis-29830 71 30 the the DET fcis-29830 71 31 convolution convolution NOUN fcis-29830 71 32 , , PUNCT fcis-29830 72 1 i i PRON fcis-29830 72 2 jx jx PROPN fcis-29830 73 1 .the .the PROPN fcis-29830 73 2 output output NOUN fcis-29830 73 3 , , PUNCT fcis-29830 73 4 i i PRON fcis-29830 73 5 jx jx VERB fcis-29830 73 6 of of ADP fcis-29830 73 7 each each DET fcis-29830 73 8 convolutional convolutional ADJ fcis-29830 73 9 unit unit NOUN fcis-29830 73 10 in in ADP fcis-29830 73 11 fig fig NOUN fcis-29830 73 12 . . PUNCT fcis-29830 74 1 2 2 NUM fcis-29830 74 2 can can AUX fcis-29830 74 3 be be AUX fcis-29830 74 4 expressed express VERB fcis-29830 74 5 by by ADP fcis-29830 74 6 equation equation NOUN fcis-29830 74 7 ( ( PUNCT fcis-29830 74 8 2 2 NUM fcis-29830 74 9 ): ): PUNCT fcis-29830 74 10 1 1 NUM fcis-29830 74 11 , , PUNCT fcis-29830 74 12 , , PUNCT fcis-29830 74 13 , , PUNCT fcis-29830 74 14 1 1 NUM fcis-29830 74 15 1 1 NUM fcis-29830 74 16 , , PUNCT fcis-29830 74 17 1 1 NUM fcis-29830 74 18 0 0 NUM fcis-29830 74 19 ( ( PUNCT fcis-29830 74 20 ( ( PUNCT fcis-29830 74 21 ) ) PUNCT fcis-29830 74 22 ) ) PUNCT fcis-29830 74 23 , , PUNCT fcis-29830 74 24 0 0 NUM fcis-29830 74 25 0 0 NUM fcis-29830 74 26 ( ( PUNCT fcis-29830 74 27 [ [ X fcis-29830 74 28 ( ( PUNCT fcis-29830 74 29 ) ) PUNCT fcis-29830 74 30 , , PUNCT fcis-29830 74 31 ( ( PUNCT fcis-29830 74 32 ) ) PUNCT fcis-29830 74 33 ] ] X fcis-29830 74 34 ) ) PUNCT fcis-29830 74 35 , , PUNCT fcis-29830 75 1 i i PRON fcis-29830 75 2 j j VERB fcis-29830 76 1 i i PRON fcis-29830 76 2 j j VERB fcis-29830 77 1 i i PRON fcis-29830 77 2 k k PROPN fcis-29830 78 1 j j PROPN fcis-29830 79 1 i i PRON fcis-29830 79 2 j j PROPN fcis-29830 80 1 k k NOUN fcis-29830 80 2 h h PROPN fcis-29830 81 1 d d NOUN fcis-29830 81 2 x x X fcis-29830 81 3 j j X fcis-29830 81 4 x x SYM fcis-29830 81 5 jh jh PROPN fcis-29830 81 6 x x SYM fcis-29830 81 7 u u PROPN fcis-29830 81 8 x x X fcis-29830 81 9   PROPN fcis-29830 81 10   PROPN fcis-29830 81 11   VERB fcis-29830 81 12   VERB fcis-29830 81 13   PRON fcis-29830 81 14   ADP fcis-29830 81 15   NUM fcis-29830 81 16   NUM fcis-29830 81 17   SCONJ fcis-29830 81 18 ( ( PUNCT fcis-29830 81 19 2 2 NUM fcis-29830 81 20 ) ) PUNCT fcis-29830 81 21 in in ADP fcis-29830 81 22 the the DET fcis-29830 81 23 formula formula NOUN fcis-29830 81 24 : : PUNCT fcis-29830 81 25 ( ( PUNCT fcis-29830 81 26 ) ) PUNCT fcis-29830 81 27 d d NOUN fcis-29830 81 28 denotes denote VERB fcis-29830 81 29 the the DET fcis-29830 81 30 current current ADJ fcis-29830 81 31 node node NOUN fcis-29830 81 32 , , PUNCT fcis-29830 81 33 ( ( PUNCT fcis-29830 81 34 ) ) PUNCT fcis-29830 81 35 h h NOUN fcis-29830 81 36 denotes denote VERB fcis-29830 81 37 the the DET fcis-29830 81 38 convolution convolution NOUN fcis-29830 81 39 operation operation NOUN fcis-29830 81 40 and and CCONJ fcis-29830 81 41 the the DET fcis-29830 81 42 activation activation NOUN fcis-29830 81 43 function function NOUN fcis-29830 81 44 ; ; PUNCT fcis-29830 81 45 denotes denote VERB fcis-29830 81 46 the the DET fcis-29830 81 47 upsampling upsampling NOUN fcis-29830 81 48 operation operation NOUN fcis-29830 81 49 ; ; PUNCT fcis-29830 82 1 [ [ PUNCT fcis-29830 82 2 ] ] X fcis-29830 82 3 denotes denote VERB fcis-29830 82 4 the the DET fcis-29830 82 5 feature feature NOUN fcis-29830 82 6 map map VERB fcis-29830 82 7 splicing splicing NOUN fcis-29830 82 8 operation operation NOUN fcis-29830 82 9 . . PUNCT fcis-29830 83 1 when when SCONJ fcis-29830 83 2 0j 0j NOUN fcis-29830 83 3   PROPN fcis-29830 83 4 , , PUNCT fcis-29830 83 5 the the DET fcis-29830 83 6 node node NOUN fcis-29830 83 7 receives receive VERB fcis-29830 83 8 only only ADV fcis-29830 83 9 inputs input NOUN fcis-29830 83 10 from from ADP fcis-29830 83 11 the the DET fcis-29830 83 12 previous previous ADJ fcis-29830 83 13 downsampled downsample VERB fcis-29830 83 14 layer layer NOUN fcis-29830 83 15 ; ; PUNCT fcis-29830 83 16 when when SCONJ fcis-29830 83 17 0j 0j NOUN fcis-29830 83 18   X fcis-29830 83 19 , , PUNCT fcis-29830 83 20 the the DET fcis-29830 83 21 node node NOUN fcis-29830 83 22 receives receive VERB fcis-29830 83 23 1j 1j NUM fcis-29830 83 24   ADJ fcis-29830 83 25 inputs input NOUN fcis-29830 83 26 , , PUNCT fcis-29830 83 27 including include VERB fcis-29830 83 28 jump jump NOUN fcis-29830 83 29 connections connection NOUN fcis-29830 83 30 and and CCONJ fcis-29830 83 31 the the DET fcis-29830 83 32 upsampled upsample VERB fcis-29830 83 33 layer layer NOUN fcis-29830 83 34 as as ADP fcis-29830 83 35 inputs input NOUN fcis-29830 83 36 . . PUNCT fcis-29830 84 1 r‐as2 r‐as2 ADJ fcis-29830 84 2 r‐as3 r‐as3 PROPN fcis-29830 84 3 r‐as4 r‐as4 PROPN fcis-29830 84 4 r‐as5 r‐as5 PROPN fcis-29830 84 5 dwconv dwconv ADJ fcis-29830 84 6 dwconv dwconv ADJ fcis-29830 84 7 dwconv dwconv ADJ fcis-29830 84 8 dwconv dwconv ADJ fcis-29830 84 9 dwconv dwconv ADJ fcis-29830 84 10 dwconv dwconv ADJ fcis-29830 84 11 dwconv dwconv ADJ fcis-29830 84 12 dwconv dwconv ADJ fcis-29830 84 13 l l NOUN fcis-29830 84 14 r‐as1 r‐as1 PROPN fcis-29830 84 15 ( ( PUNCT fcis-29830 84 16 aspp aspp NOUN fcis-29830 84 17 ) ) PUNCT fcis-29830 84 18 0 0 NUM fcis-29830 85 1 2x 2x NUM fcis-29830 85 2 ， ， PROPN fcis-29830 85 3 0 0 PROPN fcis-29830 85 4 1x 1x NUM fcis-29830 86 1 ， ， PROPN fcis-29830 86 2 0 0 PUNCT fcis-29830 87 1 0x 0x X fcis-29830 87 2 ， ， PROPN fcis-29830 87 3 1 1 NUM fcis-29830 87 4 3x 3x NUM fcis-29830 87 5 ， ， PROPN fcis-29830 87 6 1 1 NUM fcis-29830 87 7 2x 2x NUM fcis-29830 87 8 ， ， PROPN fcis-29830 87 9 1 1 NUM fcis-29830 87 10 0x 0x NOUN fcis-29830 87 11 ， ， PROPN fcis-29830 87 12 1 1 NUM fcis-29830 87 13 1x 1x NUM fcis-29830 87 14 ， ， PUNCT fcis-29830 87 15 2 2 NUM fcis-29830 87 16 0x 0x NOUN fcis-29830 87 17 ， ， PUNCT fcis-29830 87 18 3 3 NUM fcis-29830 87 19 0x 0x NOUN fcis-29830 87 20 ， ， PROPN fcis-29830 87 21 2 2 NUM fcis-29830 87 22 1x 1x NUM fcis-29830 87 23 ， ， PUNCT fcis-29830 87 24 3 3 NUM fcis-29830 87 25 1x 1x NOUN fcis-29830 87 26 ， ， PUNCT fcis-29830 87 27 2 2 NUM fcis-29830 87 28 2x 2x NOUN fcis-29830 87 29 ， ， PUNCT fcis-29830 87 30 4 4 NUM fcis-29830 87 31 0x 0x NOUN fcis-29830 87 32 ， ， PUNCT fcis-29830 87 33 down‐sampling down‐sample VERB fcis-29830 87 34 up‐sampling up‐sample VERB fcis-29830 87 35 skip‐connection skip‐connection NOUN fcis-29830 87 36 convdwconv convdwconv NOUN fcis-29830 87 37 output0 output0 ADP fcis-29830 87 38 3x 3x PRON fcis-29830 87 39 ， ， PROPN fcis-29830 87 40 0 0 NUM fcis-29830 87 41 4x 4x NUM fcis-29830 87 42 ， ， PUNCT fcis-29830 87 43 fig fig NOUN fcis-29830 87 44 2 2 NUM fcis-29830 87 45 . . PUNCT fcis-29830 87 46 sa sa PROPN fcis-29830 87 47 - - PUNCT fcis-29830 87 48 u++ u++ NOUN fcis-29830 87 49 network network NOUN fcis-29830 87 50 model model NOUN fcis-29830 87 51 we we PRON fcis-29830 87 52 also also ADV fcis-29830 87 53 replace replace VERB fcis-29830 87 54 some some PRON fcis-29830 87 55 of of ADP fcis-29830 87 56 the the DET fcis-29830 87 57 traditional traditional ADJ fcis-29830 87 58 convolution convolution NOUN fcis-29830 87 59 blocks block NOUN fcis-29830 87 60 with with ADP fcis-29830 87 61 depth depth NOUN fcis-29830 87 62 - - PUNCT fcis-29830 87 63 separable separable NOUN fcis-29830 87 64 convolutions convolution NOUN fcis-29830 87 65 in in ADP fcis-29830 87 66 the the DET fcis-29830 87 67 dsa dsa NOUN fcis-29830 87 68 - - PUNCT fcis-29830 87 69 u++ u++ PROPN fcis-29830 87 70 model model NOUN fcis-29830 87 71 . . PUNCT fcis-29830 88 1 86 86 NUM fcis-29830 89 1 its its PRON fcis-29830 89 2 effectively effectively ADV fcis-29830 89 3 reduces reduce VERB fcis-29830 89 4 the the DET fcis-29830 89 5 computational computational ADJ fcis-29830 89 6 and and CCONJ fcis-29830 89 7 parametric parametric ADJ fcis-29830 89 8 quantities quantity NOUN fcis-29830 89 9 of of ADP fcis-29830 89 10 the the DET fcis-29830 89 11 model model NOUN fcis-29830 89 12 by by ADP fcis-29830 89 13 splitting split VERB fcis-29830 89 14 the the DET fcis-29830 89 15 standard standard ADJ fcis-29830 89 16 convolution convolution NOUN fcis-29830 89 17 into into ADP fcis-29830 89 18 depth depth NOUN fcis-29830 89 19 convolution convolution NOUN fcis-29830 89 20 and and CCONJ fcis-29830 89 21 point point NOUN fcis-29830 89 22 - - PUNCT fcis-29830 89 23 by by ADP fcis-29830 89 24 - - PUNCT fcis-29830 89 25 point point NOUN fcis-29830 89 26 convolution convolution NOUN fcis-29830 89 27 , , PUNCT fcis-29830 89 28 while while SCONJ fcis-29830 89 29 retaining retain VERB fcis-29830 89 30 the the DET fcis-29830 89 31 important important ADJ fcis-29830 89 32 feature feature NOUN fcis-29830 89 33 extraction extraction NOUN fcis-29830 89 34 capabilities capability NOUN fcis-29830 89 35 . . PUNCT fcis-29830 90 1 the the DET fcis-29830 90 2 output output NOUN fcis-29830 90 3 image image NOUN fcis-29830 90 4 of of ADP fcis-29830 90 5 the the DET fcis-29830 90 6 decoder decoder NOUN fcis-29830 90 7 part part NOUN fcis-29830 90 8 goes go VERB fcis-29830 90 9 through through ADP fcis-29830 90 10 a a DET fcis-29830 90 11 fully fully ADV fcis-29830 90 12 connected connect VERB fcis-29830 90 13 layer layer NOUN fcis-29830 90 14 for for ADP fcis-29830 90 15 the the DET fcis-29830 90 16 classification classification NOUN fcis-29830 90 17 task task NOUN fcis-29830 90 18 . . PUNCT fcis-29830 91 1 with with ADP fcis-29830 91 2 this this DET fcis-29830 91 3 improved improve VERB fcis-29830 91 4 design design NOUN fcis-29830 91 5 , , PUNCT fcis-29830 91 6 especially especially ADV fcis-29830 91 7 in in ADP fcis-29830 91 8 the the DET fcis-29830 91 9 case case NOUN fcis-29830 91 10 of of ADP fcis-29830 91 11 ultrasound ultrasound ADJ fcis-29830 91 12 images image NOUN fcis-29830 91 13 where where SCONJ fcis-29830 91 14 lesions lesion NOUN fcis-29830 91 15 are be AUX fcis-29830 91 16 usually usually ADV fcis-29830 91 17 small small ADJ fcis-29830 91 18 and and CCONJ fcis-29830 91 19 of of ADP fcis-29830 91 20 low low ADJ fcis-29830 91 21 contrast contrast NOUN fcis-29830 91 22 , , PUNCT fcis-29830 91 23 this this DET fcis-29830 91 24 optimization optimization NOUN fcis-29830 91 25 not not PART fcis-29830 91 26 only only ADV fcis-29830 91 27 enhances enhance VERB fcis-29830 91 28 the the DET fcis-29830 91 29 computational computational ADJ fcis-29830 91 30 efficiency efficiency NOUN fcis-29830 91 31 of of ADP fcis-29830 91 32 the the DET fcis-29830 91 33 model model NOUN fcis-29830 91 34 , , PUNCT fcis-29830 91 35 but but CCONJ fcis-29830 91 36 also also ADV fcis-29830 91 37 improves improve VERB fcis-29830 91 38 the the DET fcis-29830 91 39 classification classification NOUN fcis-29830 91 40 accuracy accuracy NOUN fcis-29830 91 41 to to ADP fcis-29830 91 42 a a DET fcis-29830 91 43 small small ADJ fcis-29830 91 44 extent extent NOUN fcis-29830 91 45 . . PUNCT fcis-29830 92 1 2.4 2.4 NUM fcis-29830 92 2 . . PUNCT fcis-29830 93 1 depthwise depthwise NOUN fcis-29830 93 2 separable separable ADJ fcis-29830 93 3 convolution convolution NOUN fcis-29830 93 4 depthwise depthwise PROPN fcis-29830 93 5 separable separable ADJ fcis-29830 93 6 convolution convolution NOUN fcis-29830 93 7 was be AUX fcis-29830 93 8 first first ADV fcis-29830 93 9 proposed propose VERB fcis-29830 93 10 by by ADP fcis-29830 93 11 howarda howarda PROPN fcis-29830 93 12 [ [ X fcis-29830 93 13 1 1 NUM fcis-29830 93 14 ] ] PUNCT fcis-29830 93 15 and and CCONJ fcis-29830 93 16 others other NOUN fcis-29830 93 17 . . PUNCT fcis-29830 94 1 it it PRON fcis-29830 94 2 is be AUX fcis-29830 94 3 shown show VERB fcis-29830 94 4 to to PART fcis-29830 94 5 be be AUX fcis-29830 94 6 divided divide VERB fcis-29830 94 7 into into ADP fcis-29830 94 8 two two NUM fcis-29830 94 9 main main ADJ fcis-29830 94 10 processes process NOUN fcis-29830 94 11 , , PUNCT fcis-29830 94 12 channel channel NOUN fcis-29830 94 13 - - PUNCT fcis-29830 94 14 by by ADP fcis-29830 94 15 - - PUNCT fcis-29830 94 16 channel channel NOUN fcis-29830 94 17 convolution convolution NOUN fcis-29830 94 18 and and CCONJ fcis-29830 94 19 pointby pointby NOUN fcis-29830 94 20 - - PUNCT fcis-29830 94 21 point point NOUN fcis-29830 94 22 convolution convolution NOUN fcis-29830 94 23 . . PUNCT fcis-29830 95 1 channel channel PROPN fcis-29830 95 2 - - PUNCT fcis-29830 95 3 by by ADP fcis-29830 95 4 - - PUNCT fcis-29830 95 5 channel channel NOUN fcis-29830 95 6 convolution convolution NOUN fcis-29830 95 7 differs differ VERB fcis-29830 95 8 from from ADP fcis-29830 95 9 ordinary ordinary ADJ fcis-29830 95 10 convolution convolution NOUN fcis-29830 95 11 in in ADP fcis-29830 95 12 that that SCONJ fcis-29830 95 13 its its PRON fcis-29830 95 14 convolution convolution NOUN fcis-29830 95 15 kernel kernel PROPN fcis-29830 95 16 adopts adopt VERB fcis-29830 95 17 a a DET fcis-29830 95 18 single single ADJ fcis-29830 95 19 - - PUNCT fcis-29830 95 20 channel channel NOUN fcis-29830 95 21 mode mode NOUN fcis-29830 95 22 , , PUNCT fcis-29830 95 23 where where SCONJ fcis-29830 95 24 each each DET fcis-29830 95 25 channel channel NOUN fcis-29830 95 26 of of ADP fcis-29830 95 27 the the DET fcis-29830 95 28 input input NOUN fcis-29830 95 29 image image NOUN fcis-29830 95 30 is be AUX fcis-29830 95 31 convolved convolve VERB fcis-29830 95 32 to to PART fcis-29830 95 33 obtain obtain VERB fcis-29830 95 34 an an DET fcis-29830 95 35 output output NOUN fcis-29830 95 36 feature feature NOUN fcis-29830 95 37 map map NOUN fcis-29830 95 38 with with ADP fcis-29830 95 39 the the DET fcis-29830 95 40 same same ADJ fcis-29830 95 41 number number NOUN fcis-29830 95 42 of of ADP fcis-29830 95 43 channels channel NOUN fcis-29830 95 44 as as ADP fcis-29830 95 45 that that PRON fcis-29830 95 46 of of ADP fcis-29830 95 47 the the DET fcis-29830 95 48 input input NOUN fcis-29830 95 49 , , PUNCT fcis-29830 95 50 which which PRON fcis-29830 95 51 extracts extract VERB fcis-29830 95 52 the the DET fcis-29830 95 53 spatial spatial ADJ fcis-29830 95 54 features feature NOUN fcis-29830 95 55 of of ADP fcis-29830 95 56 the the DET fcis-29830 95 57 input input NOUN fcis-29830 95 58 image image NOUN fcis-29830 95 59 . . PUNCT fcis-29830 96 1 point point NOUN fcis-29830 96 2 - - PUNCT fcis-29830 96 3 by by ADP fcis-29830 96 4 - - PUNCT fcis-29830 96 5 point point NOUN fcis-29830 96 6 convolution convolution NOUN fcis-29830 96 7 makes make VERB fcis-29830 96 8 up up ADP fcis-29830 96 9 for for ADP fcis-29830 96 10 the the DET fcis-29830 96 11 shortcomings shortcoming NOUN fcis-29830 96 12 of of ADP fcis-29830 96 13 channel channel NOUN fcis-29830 96 14 - - PUNCT fcis-29830 96 15 bychannel bychannel NOUN fcis-29830 96 16 convolution convolution NOUN fcis-29830 96 17 that that PRON fcis-29830 96 18 does do AUX fcis-29830 96 19 not not PART fcis-29830 96 20 effectively effectively ADV fcis-29830 96 21 utilize utilize VERB fcis-29830 96 22 the the DET fcis-29830 96 23 feature feature NOUN fcis-29830 96 24 information information NOUN fcis-29830 96 25 of of ADP fcis-29830 96 26 different different ADJ fcis-29830 96 27 channels channel NOUN fcis-29830 96 28 , , PUNCT fcis-29830 96 29 using use VERB fcis-29830 96 30 a a DET fcis-29830 96 31 1 1 NUM fcis-29830 96 32 * * SYM fcis-29830 96 33 1 1 NUM fcis-29830 96 34 convolution convolution NOUN fcis-29830 96 35 kernel kernel NOUN fcis-29830 96 36 , , PUNCT fcis-29830 96 37 and and CCONJ fcis-29830 96 38 outputs output VERB fcis-29830 96 39 a a DET fcis-29830 96 40 feature feature NOUN fcis-29830 96 41 map map NOUN fcis-29830 96 42 after after SCONJ fcis-29830 96 43 all all DET fcis-29830 96 44 channels channel NOUN fcis-29830 96 45 are be AUX fcis-29830 96 46 aggregated aggregate VERB fcis-29830 96 47 . . PUNCT fcis-29830 97 1 1c 1c NOUN fcis-29830 97 2 w w ADP fcis-29830 97 3 h h NOUN fcis-29830 97 4 1 1 NUM fcis-29830 97 5 w w NOUN fcis-29830 97 6 hc hc PROPN fcis-29830 97 7 k k PROPN fcis-29830 97 8 k k PROPN fcis-29830 97 9   PROPN fcis-29830 97 10 11 11 NUM fcis-29830 97 11 1 1 NUM fcis-29830 97 12 c c PROPN fcis-29830 97 13   PROPN fcis-29830 97 14 2c 2c NOUN fcis-29830 97 15 2c 2c NUM fcis-29830 97 16 fig fig NOUN fcis-29830 97 17 3 3 NUM fcis-29830 97 18 . . PUNCT fcis-29830 98 1 depthwise depthwise VERB fcis-29830 98 2 separable separable ADJ fcis-29830 98 3 convolution convolution NOUN fcis-29830 98 4 first first ADV fcis-29830 98 5 , , PUNCT fcis-29830 98 6 channel channel NOUN fcis-29830 98 7 - - PUNCT fcis-29830 98 8 by by ADP fcis-29830 98 9 - - PUNCT fcis-29830 98 10 channel channel NOUN fcis-29830 98 11 convolution convolution NOUN fcis-29830 98 12 convolves convolve VERB fcis-29830 98 13 its its PRON fcis-29830 98 14 input input NOUN fcis-29830 98 15 image image NOUN fcis-29830 98 16 using use VERB fcis-29830 98 17 convolution convolution NOUN fcis-29830 98 18 kernels kernel NOUN fcis-29830 98 19 with with ADP fcis-29830 98 20 the the DET fcis-29830 98 21 same same ADJ fcis-29830 98 22 number number NOUN fcis-29830 98 23 of of ADP fcis-29830 98 24 channels channel NOUN fcis-29830 98 25 1c 1c NOUN fcis-29830 98 26 respectively respectively ADV fcis-29830 98 27 , , PUNCT fcis-29830 98 28 and and CCONJ fcis-29830 98 29 1c 1c NUM fcis-29830 98 30 feature feature NOUN fcis-29830 98 31 maps map NOUN fcis-29830 98 32 are be AUX fcis-29830 98 33 obtained obtain VERB fcis-29830 98 34 after after ADP fcis-29830 98 35 convolution convolution NOUN fcis-29830 98 36 : : PUNCT fcis-29830 98 37 1w 1w NUM fcis-29830 98 38 w w PROPN fcis-29830 98 39 hhc hhc PROPN fcis-29830 98 40 k k PROPN fcis-29830 98 41 k k PROPN fcis-29830 98 42 , , PUNCT fcis-29830 98 43 however however ADV fcis-29830 98 44 , , PUNCT fcis-29830 98 45 the the DET fcis-29830 98 46 feature feature NOUN fcis-29830 98 47 information information NOUN fcis-29830 98 48 between between ADP fcis-29830 98 49 the the DET fcis-29830 98 50 channels channel NOUN fcis-29830 98 51 was be AUX fcis-29830 98 52 not not PART fcis-29830 98 53 utilized utilize VERB fcis-29830 98 54 in in ADP fcis-29830 98 55 the the DET fcis-29830 98 56 meantime meantime NOUN fcis-29830 98 57 . . PUNCT fcis-29830 99 1 thus thus ADV fcis-29830 99 2 , , PUNCT fcis-29830 99 3 in in ADP fcis-29830 99 4 point point NOUN fcis-29830 99 5 - - PUNCT fcis-29830 99 6 by by ADP fcis-29830 99 7 - - PUNCT fcis-29830 99 8 point point NOUN fcis-29830 99 9 convolution convolution NOUN fcis-29830 99 10 , , PUNCT fcis-29830 99 11 using use VERB fcis-29830 99 12 11 11 NUM fcis-29830 99 13 1 1 NUM fcis-29830 99 14 c c ADJ fcis-29830 99 15   NOUN fcis-29830 99 16 convolution convolution NOUN fcis-29830 99 17 kernel kernel PROPN fcis-29830 99 18 for for ADP fcis-29830 99 19 channel channel NOUN fcis-29830 99 20 - - PUNCT fcis-29830 99 21 by by ADP fcis-29830 99 22 - - PUNCT fcis-29830 99 23 channel channel NOUN fcis-29830 99 24 convolution convolution NOUN fcis-29830 99 25 to to PART fcis-29830 99 26 get get VERB fcis-29830 99 27 1c 1c NUM fcis-29830 99 28 feature feature NOUN fcis-29830 99 29 maps map NOUN fcis-29830 99 30 for for ADP fcis-29830 99 31 inter inter ADJ fcis-29830 99 32 - - ADJ fcis-29830 99 33 channel channel ADJ fcis-29830 99 34 feature feature NOUN fcis-29830 99 35 information information NOUN fcis-29830 99 36 fusion fusion NOUN fcis-29830 99 37 , , PUNCT fcis-29830 99 38 to to PART fcis-29830 99 39 get get VERB fcis-29830 99 40 2c 2c NUM fcis-29830 99 41 number number NOUN fcis-29830 99 42 of of ADP fcis-29830 99 43 channels channel NOUN fcis-29830 99 44 of of ADP fcis-29830 99 45 feature feature NOUN fcis-29830 99 46 maps map NOUN fcis-29830 99 47 as as ADP fcis-29830 99 48 the the DET fcis-29830 99 49 output output NOUN fcis-29830 99 50 , , PUNCT fcis-29830 99 51 the the DET fcis-29830 99 52 computational computational ADJ fcis-29830 99 53 volume volume NOUN fcis-29830 99 54 of of ADP fcis-29830 99 55 its its PRON fcis-29830 99 56 convolution convolution NOUN fcis-29830 99 57 is be AUX fcis-29830 99 58 : : PUNCT fcis-29830 99 59 1 1 NUM fcis-29830 99 60 2whc 2whc NUM fcis-29830 99 61 c c NOUN fcis-29830 99 62 . . PUNCT fcis-29830 100 1 the the DET fcis-29830 100 2 computational computational ADJ fcis-29830 100 3 effort effort NOUN fcis-29830 100 4 of of ADP fcis-29830 100 5 using use VERB fcis-29830 100 6 depthwise depthwise NOUN fcis-29830 100 7 separable separable ADJ fcis-29830 100 8 convolution convolution NOUN fcis-29830 100 9 is be AUX fcis-29830 100 10 : : PUNCT fcis-29830 100 11 1 1 NUM fcis-29830 100 12 1 1 NUM fcis-29830 100 13 2w 2w NUM fcis-29830 100 14 w w PROPN fcis-29830 100 15 hhc hhc PROPN fcis-29830 100 16 k k PROPN fcis-29830 100 17 k k PROPN fcis-29830 100 18 whc whc PROPN fcis-29830 100 19 c c PROPN fcis-29830 100 20 . . PUNCT fcis-29830 101 1 the the DET fcis-29830 101 2 computational computational ADJ fcis-29830 101 3 effort effort NOUN fcis-29830 101 4 to to PART fcis-29830 101 5 obtain obtain VERB fcis-29830 101 6 2c 2c NUM fcis-29830 101 7 output output NOUN fcis-29830 101 8 feature feature NOUN fcis-29830 101 9 maps map NOUN fcis-29830 101 10 using use VERB fcis-29830 101 11 conventional conventional ADJ fcis-29830 101 12 convolution convolution NOUN fcis-29830 101 13 of of ADP fcis-29830 101 14 input input NOUN fcis-29830 101 15 convolution convolution NOUN fcis-29830 101 16 with with ADP fcis-29830 101 17 channel channel NOUN fcis-29830 101 18 number number NOUN fcis-29830 101 19 1c 1c NOUN fcis-29830 101 20 and and CCONJ fcis-29830 101 21 size size NOUN fcis-29830 101 22 hwk hwk PROPN fcis-29830 101 23 k k PROPN fcis-29830 101 24 is be AUX fcis-29830 101 25 : : PUNCT fcis-29830 101 26 1w 1w NUM fcis-29830 101 27 w w PROPN fcis-29830 101 28 hhc hhc PROPN fcis-29830 101 29 k k PROPN fcis-29830 101 30 k k PROPN fcis-29830 101 31 . . PUNCT fcis-29830 102 1 the the DET fcis-29830 102 2 ratio ratio NOUN fcis-29830 102 3 of of ADP fcis-29830 102 4 the the DET fcis-29830 102 5 computational computational ADJ fcis-29830 102 6 effort effort NOUN fcis-29830 102 7 of of ADP fcis-29830 102 8 the the DET fcis-29830 102 9 two two NUM fcis-29830 102 10 is be AUX fcis-29830 102 11 : : PUNCT fcis-29830 102 12 2 2 NUM fcis-29830 102 13 1 1 NUM fcis-29830 102 14 1 1 NUM fcis-29830 102 15 w w NOUN fcis-29830 102 16 hc hc PROPN fcis-29830 102 17 k k PROPN fcis-29830 102 18 k k PROPN fcis-29830 102 19   PROPN fcis-29830 102 20 . . PUNCT fcis-29830 103 1 the the DET fcis-29830 103 2 convolution convolution NOUN fcis-29830 103 3 kernel kernel PROPN fcis-29830 103 4 size size PROPN fcis-29830 103 5 w w PROPN fcis-29830 103 6 hk hk PROPN fcis-29830 103 7 k k PROPN fcis-29830 103 8 used use VERB fcis-29830 103 9 in in ADP fcis-29830 103 10 this this DET fcis-29830 103 11 experiment experiment NOUN fcis-29830 103 12 is be AUX fcis-29830 103 13 3x3 3x3 NUM fcis-29830 103 14 . . PUNCT fcis-29830 104 1 it it PRON fcis-29830 104 2 can can AUX fcis-29830 104 3 be be AUX fcis-29830 104 4 seen see VERB fcis-29830 104 5 that that SCONJ fcis-29830 104 6 the the DET fcis-29830 104 7 computational computational ADJ fcis-29830 104 8 effort effort NOUN fcis-29830 104 9 of of ADP fcis-29830 104 10 depthseparable depthseparable ADJ fcis-29830 104 11 convolution convolution NOUN fcis-29830 104 12 is be AUX fcis-29830 104 13 about about ADP fcis-29830 104 14 1/9th 1/9th NUM fcis-29830 104 15 of of ADP fcis-29830 104 16 that that PRON fcis-29830 104 17 of of ADP fcis-29830 104 18 conventional conventional ADJ fcis-29830 104 19 convolution convolution NOUN fcis-29830 104 20 . . PUNCT fcis-29830 105 1 2.5 2.5 NUM fcis-29830 105 2 . . PUNCT fcis-29830 106 1 atrous atrous ADJ fcis-29830 106 2 spatial spatial ADJ fcis-29830 106 3 pyramid pyramid NOUN fcis-29830 106 4 pooling pool VERB fcis-29830 106 5 the the DET fcis-29830 106 6 aspp aspp NOUN fcis-29830 106 7 module module NOUN fcis-29830 106 8 is be AUX fcis-29830 106 9 used use VERB fcis-29830 106 10 to to PART fcis-29830 106 11 enhance enhance VERB fcis-29830 106 12 the the DET fcis-29830 106 13 capability capability NOUN fcis-29830 106 14 of of ADP fcis-29830 106 15 convolutional convolutional ADJ fcis-29830 106 16 neural neural ADJ fcis-29830 106 17 networks network NOUN fcis-29830 106 18 in in ADP fcis-29830 106 19 processing process VERB fcis-29830 106 20 multi multi ADJ fcis-29830 106 21 - - ADJ fcis-29830 106 22 scale scale ADJ fcis-29830 106 23 information information NOUN fcis-29830 106 24 , , PUNCT fcis-29830 106 25 especially especially ADV fcis-29830 106 26 in in ADP fcis-29830 106 27 image image NOUN fcis-29830 106 28 segmentation segmentation NOUN fcis-29830 106 29 tasks task NOUN fcis-29830 106 30 . . PUNCT fcis-29830 107 1 aspp aspp NOUN fcis-29830 107 2 was be AUX fcis-29830 107 3 firstly firstly ADV fcis-29830 107 4 applied apply VERB fcis-29830 107 5 in in ADP fcis-29830 107 6 the the DET fcis-29830 107 7 deeplab deeplab NOUN fcis-29830 107 8 [ [ X fcis-29830 107 9 1 1 NUM fcis-29830 107 10 ] ] X fcis-29830 107 11 family family NOUN fcis-29830 107 12 of of ADP fcis-29830 107 13 models model NOUN fcis-29830 107 14 , , PUNCT fcis-29830 107 15 aiming aim VERB fcis-29830 107 16 to to PART fcis-29830 107 17 enhance enhance VERB fcis-29830 107 18 the the DET fcis-29830 107 19 model model NOUN fcis-29830 107 20 's 's PART fcis-29830 107 21 ability ability NOUN fcis-29830 107 22 to to PART fcis-29830 107 23 perceive perceive VERB fcis-29830 107 24 features feature NOUN fcis-29830 107 25 at at ADP fcis-29830 107 26 different different ADJ fcis-29830 107 27 spatial spatial ADJ fcis-29830 107 28 scales scale NOUN fcis-29830 107 29 through through ADP fcis-29830 107 30 the the DET fcis-29830 107 31 introduction introduction NOUN fcis-29830 107 32 of of ADP fcis-29830 107 33 null null ADJ fcis-29830 107 34 convolution convolution NOUN fcis-29830 107 35 at at ADP fcis-29830 107 36 different different ADJ fcis-29830 107 37 scales scale NOUN fcis-29830 107 38 and and CCONJ fcis-29830 107 39 global global ADJ fcis-29830 107 40 average average ADJ fcis-29830 107 41 pooling pooling NOUN fcis-29830 107 42 . . PUNCT fcis-29830 108 1 by by ADP fcis-29830 108 2 capturing capture VERB fcis-29830 108 3 the the DET fcis-29830 108 4 contextual contextual ADJ fcis-29830 108 5 information information NOUN fcis-29830 108 6 of of ADP fcis-29830 108 7 an an DET fcis-29830 108 8 image image NOUN fcis-29830 108 9 at at ADP fcis-29830 108 10 different different ADJ fcis-29830 108 11 scales scale NOUN fcis-29830 108 12 , , PUNCT fcis-29830 108 13 the the DET fcis-29830 108 14 model model NOUN fcis-29830 108 15 's 's PART fcis-29830 108 16 ability ability NOUN fcis-29830 108 17 to to PART fcis-29830 108 18 distinguish distinguish VERB fcis-29830 108 19 between between ADP fcis-29830 108 20 objects object NOUN fcis-29830 108 21 and and CCONJ fcis-29830 108 22 backgrounds background NOUN fcis-29830 108 23 is be AUX fcis-29830 108 24 improved improve VERB fcis-29830 108 25 , , PUNCT fcis-29830 108 26 especially especially ADV fcis-29830 108 27 in in ADP fcis-29830 108 28 images image NOUN fcis-29830 108 29 with with ADP fcis-29830 108 30 complex complex ADJ fcis-29830 108 31 structures structure NOUN fcis-29830 108 32 and and CCONJ fcis-29830 108 33 fuzzy fuzzy ADJ fcis-29830 108 34 boundaries boundary NOUN fcis-29830 108 35 . . PUNCT fcis-29830 109 1 null null ADJ fcis-29830 109 2 convolution convolution NOUN fcis-29830 109 3 , , PUNCT fcis-29830 109 4 i.e. i.e. X fcis-29830 109 5 , , PUNCT fcis-29830 109 6 inserting insert VERB fcis-29830 109 7 nulls null NOUN fcis-29830 109 8 in in ADP fcis-29830 109 9 the the DET fcis-29830 109 10 convolution convolution NOUN fcis-29830 109 11 kernel kernel NOUN fcis-29830 109 12 to to PART fcis-29830 109 13 increase increase VERB fcis-29830 109 14 the the DET fcis-29830 109 15 size size NOUN fcis-29830 109 16 of of ADP fcis-29830 109 17 the the DET fcis-29830 109 18 perceptual perceptual ADJ fcis-29830 109 19 field field NOUN fcis-29830 109 20 without without ADP fcis-29830 109 21 increasing increase VERB fcis-29830 109 22 the the DET fcis-29830 109 23 amount amount NOUN fcis-29830 109 24 of of ADP fcis-29830 109 25 computation computation NOUN fcis-29830 109 26 or or CCONJ fcis-29830 109 27 pooling pool VERB fcis-29830 109 28 operations operation NOUN fcis-29830 109 29 . . PUNCT fcis-29830 110 1 in in ADP fcis-29830 110 2 the the DET fcis-29830 110 3 aspp aspp NOUN fcis-29830 110 4 module module NOUN fcis-29830 110 5 of of ADP fcis-29830 110 6 this this DET fcis-29830 110 7 experiment experiment NOUN fcis-29830 110 8 , , PUNCT fcis-29830 110 9 cavity cavity NOUN fcis-29830 110 10 convolution convolution NOUN fcis-29830 110 11 with with ADP fcis-29830 110 12 expansion expansion NOUN fcis-29830 110 13 rates rate NOUN fcis-29830 110 14 of of ADP fcis-29830 110 15 6 6 NUM fcis-29830 110 16 , , PUNCT fcis-29830 110 17 12 12 NUM fcis-29830 110 18 , , PUNCT fcis-29830 110 19 and and CCONJ fcis-29830 110 20 18 18 NUM fcis-29830 110 21 is be AUX fcis-29830 110 22 used use VERB fcis-29830 110 23 , , PUNCT fcis-29830 110 24 which which PRON fcis-29830 110 25 enables enable VERB fcis-29830 110 26 the the DET fcis-29830 110 27 model model NOUN fcis-29830 110 28 to to PART fcis-29830 110 29 perceive perceive VERB fcis-29830 110 30 the the DET fcis-29830 110 31 features feature NOUN fcis-29830 110 32 of of ADP fcis-29830 110 33 the the DET fcis-29830 110 34 image image NOUN fcis-29830 110 35 at at ADP fcis-29830 110 36 different different ADJ fcis-29830 110 37 scales scale NOUN fcis-29830 110 38 and and CCONJ fcis-29830 110 39 thus thus ADV fcis-29830 110 40 capture capture VERB fcis-29830 110 41 both both CCONJ fcis-29830 110 42 local local ADJ fcis-29830 110 43 and and CCONJ fcis-29830 110 44 global global ADJ fcis-29830 110 45 information information NOUN fcis-29830 110 46 . . PUNCT fcis-29830 111 1 it it PRON fcis-29830 111 2 is be AUX fcis-29830 111 3 also also ADV fcis-29830 111 4 capable capable ADJ fcis-29830 111 5 of of ADP fcis-29830 111 6 expanding expand VERB fcis-29830 111 7 the the DET fcis-29830 111 8 receptive receptive ADJ fcis-29830 111 9 field field NOUN fcis-29830 111 10 , , PUNCT fcis-29830 111 11 thus thus ADV fcis-29830 111 12 increasing increase VERB fcis-29830 111 13 the the DET fcis-29830 111 14 model model NOUN fcis-29830 111 15 's 's PART fcis-29830 111 16 ability ability NOUN fcis-29830 111 17 to to PART fcis-29830 111 18 perceive perceive VERB fcis-29830 111 19 larger large ADJ fcis-29830 111 20 regions region NOUN fcis-29830 111 21 . . PUNCT fcis-29830 112 1 input input NOUN fcis-29830 112 2 pooling pool VERB fcis-29830 112 3 3x3 3x3 NUM fcis-29830 112 4     SPACE fcis-29830 112 5 conv conv ADJ fcis-29830 112 6 rate rate NOUN fcis-29830 112 7     SPACE fcis-29830 112 8 18 18 NUM fcis-29830 112 9 3x3 3x3 NUM fcis-29830 112 10     SPACE fcis-29830 112 11 conv conv ADJ fcis-29830 112 12 rate rate NOUN fcis-29830 112 13     SPACE fcis-29830 112 14 12 12 NUM fcis-29830 112 15 3x3 3x3 NUM fcis-29830 112 16     SPACE fcis-29830 112 17 conv conv ADJ fcis-29830 112 18 rate rate NOUN fcis-29830 112 19     SPACE fcis-29830 112 20 6 6 NUM fcis-29830 112 21 1x1 1x1 NUM fcis-29830 112 22     SPACE fcis-29830 112 23 conv conv ADJ fcis-29830 112 24 upsample upsample PROPN fcis-29830 112 25 output output NOUN fcis-29830 112 26 1x1 1x1 NUM fcis-29830 112 27     SPACE fcis-29830 112 28 conv conv ADJ fcis-29830 112 29 fig fig NOUN fcis-29830 112 30 4 4 NUM fcis-29830 112 31 . . PUNCT fcis-29830 112 32 aspp aspp NOUN fcis-29830 112 33 the the DET fcis-29830 112 34 module module NOUN fcis-29830 112 35 consists consist VERB fcis-29830 112 36 of of ADP fcis-29830 112 37 multiple multiple ADJ fcis-29830 112 38 parallel parallel ADJ fcis-29830 112 39 branches branch NOUN fcis-29830 112 40 that that PRON fcis-29830 112 41 capture capture VERB fcis-29830 112 42 features feature VERB fcis-29830 112 43 at at ADP fcis-29830 112 44 different different ADJ fcis-29830 112 45 scales scale NOUN fcis-29830 112 46 in in ADP fcis-29830 112 47 the the DET fcis-29830 112 48 image image NOUN fcis-29830 112 49 by by ADP fcis-29830 112 50 using use VERB fcis-29830 112 51 3x3 3x3 NUM fcis-29830 112 52 null null ADJ fcis-29830 112 53 convolution convolution NOUN fcis-29830 112 54 with with ADP fcis-29830 112 55 different different ADJ fcis-29830 112 56 expansion expansion NOUN fcis-29830 112 57 rates rate NOUN fcis-29830 112 58 . . PUNCT fcis-29830 113 1 the the DET fcis-29830 113 2 1x1 1x1 NUM fcis-29830 113 3 ordinary ordinary ADJ fcis-29830 113 4 convolution convolution NOUN fcis-29830 113 5 , , PUNCT fcis-29830 113 6 null null ADJ fcis-29830 113 7 convolution convolution NOUN fcis-29830 113 8 is be AUX fcis-29830 113 9 spliced splice VERB fcis-29830 113 10 with with ADP fcis-29830 113 11 the the DET fcis-29830 113 12 image image NOUN fcis-29830 113 13 obtained obtain VERB fcis-29830 113 14 after after ADP fcis-29830 113 15 global global ADJ fcis-29830 113 16 average average ADJ fcis-29830 113 17 pooling pooling NOUN fcis-29830 113 18 via via ADP fcis-29830 113 19 1x1 1x1 NUM fcis-29830 113 20 ordinary ordinary ADJ fcis-29830 113 21 convolution convolution NOUN fcis-29830 113 22 and and CCONJ fcis-29830 113 23 up up ADP fcis-29830 113 24 adoption adoption NOUN fcis-29830 113 25 again again ADV fcis-29830 113 26 . . PUNCT fcis-29830 114 1 and and CCONJ fcis-29830 114 2 further further ADJ fcis-29830 114 3 feature feature NOUN fcis-29830 114 4 fusion fusion NOUN fcis-29830 114 5 87 87 NUM fcis-29830 114 6 is be AUX fcis-29830 114 7 performed perform VERB fcis-29830 114 8 by by ADP fcis-29830 114 9 a a DET fcis-29830 114 10 1x1 1x1 NUM fcis-29830 114 11 ordinary ordinary ADJ fcis-29830 114 12 convolution convolution NOUN fcis-29830 114 13 to to ADP fcis-29830 114 14 output output NOUN fcis-29830 114 15 multiscale multiscale NOUN fcis-29830 114 16 features feature NOUN fcis-29830 114 17 . . PUNCT fcis-29830 115 1 this this PRON fcis-29830 115 2 enables enable VERB fcis-29830 115 3 the the DET fcis-29830 115 4 model model NOUN fcis-29830 115 5 to to PART fcis-29830 115 6 capture capture VERB fcis-29830 115 7 both both DET fcis-29830 115 8 local local ADJ fcis-29830 115 9 features feature NOUN fcis-29830 115 10 and and CCONJ fcis-29830 115 11 global global ADJ fcis-29830 115 12 information information NOUN fcis-29830 115 13 in in ADP fcis-29830 115 14 the the DET fcis-29830 115 15 image image NOUN fcis-29830 115 16 , , PUNCT fcis-29830 115 17 thus thus ADV fcis-29830 115 18 improving improve VERB fcis-29830 115 19 the the DET fcis-29830 115 20 robustness robustness NOUN fcis-29830 115 21 and and CCONJ fcis-29830 115 22 accuracy accuracy NOUN fcis-29830 115 23 of of ADP fcis-29830 115 24 the the DET fcis-29830 115 25 model model NOUN fcis-29830 115 26 . . PUNCT fcis-29830 116 1 as as SCONJ fcis-29830 116 2 shown show VERB fcis-29830 116 3 in in ADP fcis-29830 116 4 fig fig NOUN fcis-29830 116 5 . . PUNCT fcis-29830 117 1 4 4 NUM fcis-29830 117 2 . . NOUN fcis-29830 117 3 3 3 NUM fcis-29830 117 4 . . NOUN fcis-29830 117 5 experiments experiment NOUN fcis-29830 117 6 and and CCONJ fcis-29830 117 7 results result NOUN fcis-29830 117 8 3.1 3.1 NUM fcis-29830 117 9 . . PUNCT fcis-29830 118 1 experimental experimental ADJ fcis-29830 118 2 environment environment NOUN fcis-29830 118 3 this this DET fcis-29830 118 4 experiment experiment NOUN fcis-29830 118 5 experiments experiment NOUN fcis-29830 118 6 pycharm pycharm VERB fcis-29830 118 7 as as ADP fcis-29830 118 8 compiler compiler NOUN fcis-29830 118 9 , , PUNCT fcis-29830 118 10 using use VERB fcis-29830 118 11 python-3.10.15 python-3.10.15 NOUN fcis-29830 118 12 as as ADP fcis-29830 118 13 compiler compiler NOUN fcis-29830 118 14 , , PUNCT fcis-29830 118 15 the the DET fcis-29830 118 16 experimental experimental ADJ fcis-29830 118 17 framework framework NOUN fcis-29830 118 18 is be AUX fcis-29830 118 19 pytorch pytorch NOUN fcis-29830 118 20 . . PUNCT fcis-29830 119 1 the the DET fcis-29830 119 2 hardware hardware NOUN fcis-29830 119 3 environment environment NOUN fcis-29830 119 4 is be AUX fcis-29830 119 5 12th 12th ADJ fcis-29830 119 6 gen gen PROPN fcis-29830 119 7 intel(r intel(r PROPN fcis-29830 119 8 ) ) PUNCT fcis-29830 119 9 core core NOUN fcis-29830 119 10 ( ( PUNCT fcis-29830 119 11 tm tm NOUN fcis-29830 119 12 ) ) PUNCT fcis-29830 119 13 i7 i7 NOUN fcis-29830 119 14 - - PUNCT fcis-29830 119 15 12700h 12700h NUM fcis-29830 119 16 2.30 2.30 NUM fcis-29830 119 17 ghz ghz NOUN fcis-29830 119 18 , , PUNCT fcis-29830 119 19 graphics graphic NOUN fcis-29830 119 20 card card NOUN fcis-29830 119 21 is be AUX fcis-29830 119 22 nvidia nvidia PROPN fcis-29830 119 23 geforce geforce PROPN fcis-29830 119 24 rtx3060 rtx3060 VERB fcis-29830 119 25 laptop laptop PROPN fcis-29830 119 26 gpu gpu PROPN fcis-29830 119 27 , , PUNCT fcis-29830 119 28 64 64 NUM fcis-29830 119 29 - - PUNCT fcis-29830 119 30 bit bit NOUN fcis-29830 119 31 windows window NOUN fcis-29830 119 32 operating operating NOUN fcis-29830 119 33 system system NOUN fcis-29830 119 34 . . PUNCT fcis-29830 120 1 3.2 3.2 NUM fcis-29830 120 2 . . PUNCT fcis-29830 120 3 evaluation evaluation NOUN fcis-29830 120 4 indicators indicator NOUN fcis-29830 120 5 in in ADP fcis-29830 120 6 this this DET fcis-29830 120 7 experiment experiment NOUN fcis-29830 120 8 , , PUNCT fcis-29830 120 9 the the DET fcis-29830 120 10 accuracy accuracy NOUN fcis-29830 120 11 accuracy accuracy NOUN fcis-29830 120 12 , , PUNCT fcis-29830 120 13 f1 f1 NOUN fcis-29830 120 14 value value NOUN fcis-29830 120 15 was be AUX fcis-29830 120 16 used use VERB fcis-29830 120 17 to to PART fcis-29830 120 18 evaluate evaluate VERB fcis-29830 120 19 the the DET fcis-29830 120 20 result result NOUN fcis-29830 120 21 of of ADP fcis-29830 120 22 ultrasound ultrasound ADJ fcis-29830 120 23 breast breast NOUN fcis-29830 120 24 image image NOUN fcis-29830 120 25 feature feature NOUN fcis-29830 120 26 recognition recognition NOUN fcis-29830 120 27 and and CCONJ fcis-29830 120 28 the the DET fcis-29830 120 29 formula formula NOUN fcis-29830 120 30 was be AUX fcis-29830 120 31 calculated calculate VERB fcis-29830 120 32 as as SCONJ fcis-29830 120 33 follows follow VERB fcis-29830 120 34 : : PUNCT fcis-29830 120 35 n n PROPN fcis-29830 121 1 + + CCONJ fcis-29830 121 2 f f PROPN fcis-29830 122 1 p p NOUN fcis-29830 122 2 n n PROPN fcis-29830 122 3 p p NOUN fcis-29830 122 4 n n PROPN fcis-29830 122 5 p p NOUN fcis-29830 122 6 t t PROPN fcis-29830 122 7 t t NOUN fcis-29830 122 8 accuracy accuracy NOUN fcis-29830 122 9 t t PROPN fcis-29830 122 10 t t PROPN fcis-29830 122 11 f f PROPN fcis-29830 123 1   PROPN fcis-29830 123 2   PROPN fcis-29830 123 3   PUNCT fcis-29830 123 4   X fcis-29830 123 5 ( ( PUNCT fcis-29830 123 6 3 3 X fcis-29830 123 7 ) ) PUNCT fcis-29830 123 8 p p NOUN fcis-29830 123 9 p p PROPN fcis-29830 123 10 n n PROPN fcis-29830 123 11 t t PROPN fcis-29830 123 12 recall recall NOUN fcis-29830 123 13 t t PROPN fcis-29830 123 14 f f PROPN fcis-29830 124 1   PROPN fcis-29830 124 2   PUNCT fcis-29830 124 3 ( ( PUNCT fcis-29830 124 4 4 4 NUM fcis-29830 124 5 ) ) PUNCT fcis-29830 124 6 p p NOUN fcis-29830 124 7 p p X fcis-29830 124 8 p p X fcis-29830 124 9 t t NOUN fcis-29830 124 10 pr pr NOUN fcis-29830 124 11 t t PROPN fcis-29830 124 12 f f PROPN fcis-29830 124 13 ecesion ecesion PROPN fcis-29830 124 14   PROPN fcis-29830 124 15   PUNCT fcis-29830 124 16 ( ( PUNCT fcis-29830 124 17 5 5 NUM fcis-29830 124 18 ) ) SYM fcis-29830 124 19 1 1 NUM fcis-29830 124 20 2 2 NUM fcis-29830 124 21 pr pr NOUN fcis-29830 124 22 re re VERB fcis-29830 124 23 pr pr X fcis-29830 124 24 re re VERB fcis-29830 124 25 ecision ecision NOUN fcis-29830 124 26 call call VERB fcis-29830 124 27 f f PROPN fcis-29830 124 28 ecision ecision NOUN fcis-29830 124 29 call call VERB fcis-29830 124 30   PROPN fcis-29830 124 31   PROPN fcis-29830 124 32   PROPN fcis-29830 124 33   X fcis-29830 124 34 ( ( PUNCT fcis-29830 124 35 6 6 NUM fcis-29830 124 36 ) ) PUNCT fcis-29830 124 37 in in ADP fcis-29830 124 38 the the DET fcis-29830 124 39 formula formula NOUN fcis-29830 124 40 : : PUNCT fcis-29830 124 41 pt pt PROPN fcis-29830 124 42 refers refer VERB fcis-29830 124 43 to to ADP fcis-29830 124 44 the the DET fcis-29830 124 45 number number NOUN fcis-29830 124 46 of of ADP fcis-29830 124 47 correctly correctly ADV fcis-29830 124 48 categorized categorize VERB fcis-29830 124 49 positive positive ADJ fcis-29830 124 50 samples sample NOUN fcis-29830 124 51 , , PUNCT fcis-29830 124 52 which which PRON fcis-29830 124 53 are be AUX fcis-29830 124 54 predicted predict VERB fcis-29830 124 55 to to PART fcis-29830 124 56 be be AUX fcis-29830 124 57 positive positive ADJ fcis-29830 124 58 and and CCONJ fcis-29830 124 59 are be AUX fcis-29830 124 60 actually actually ADV fcis-29830 124 61 positive positive ADJ fcis-29830 124 62 ; ; PUNCT fcis-29830 124 63 pf pf PROPN fcis-29830 124 64 refers refer VERB fcis-29830 124 65 to to ADP fcis-29830 124 66 the the DET fcis-29830 124 67 number number NOUN fcis-29830 124 68 of of ADP fcis-29830 124 69 negative negative ADJ fcis-29830 124 70 samples sample NOUN fcis-29830 124 71 incorrectly incorrectly ADV fcis-29830 124 72 labeled label VERB fcis-29830 124 73 as as ADP fcis-29830 124 74 positive positive ADJ fcis-29830 124 75 samples sample NOUN fcis-29830 124 76 , , PUNCT fcis-29830 124 77 which which PRON fcis-29830 124 78 are be AUX fcis-29830 124 79 actually actually ADV fcis-29830 124 80 negative negative ADJ fcis-29830 124 81 samples sample NOUN fcis-29830 124 82 predicted predict VERB fcis-29830 124 83 to to PART fcis-29830 124 84 be be AUX fcis-29830 124 85 positive positive ADJ fcis-29830 124 86 ; ; PUNCT fcis-29830 124 87 nt not PART fcis-29830 124 88 refers refer VERB fcis-29830 124 89 to to ADP fcis-29830 124 90 the the DET fcis-29830 124 91 number number NOUN fcis-29830 124 92 of of ADP fcis-29830 124 93 correctly correctly ADV fcis-29830 124 94 categorized categorize VERB fcis-29830 124 95 negative negative ADJ fcis-29830 124 96 samples sample NOUN fcis-29830 124 97 , , PUNCT fcis-29830 124 98 which which PRON fcis-29830 124 99 are be AUX fcis-29830 124 100 predicted predict VERB fcis-29830 124 101 to to PART fcis-29830 124 102 be be AUX fcis-29830 124 103 negative negative ADJ fcis-29830 124 104 and and CCONJ fcis-29830 124 105 are be AUX fcis-29830 124 106 actually actually ADV fcis-29830 124 107 negative negative ADJ fcis-29830 124 108 ; ; PUNCT fcis-29830 124 109 and and CCONJ fcis-29830 124 110 nf nf NOUN fcis-29830 124 111 refers refer VERB fcis-29830 124 112 to to ADP fcis-29830 124 113 the the DET fcis-29830 124 114 number number NOUN fcis-29830 124 115 of of ADP fcis-29830 124 116 positive positive ADJ fcis-29830 124 117 samples sample NOUN fcis-29830 124 118 incorrectly incorrectly ADV fcis-29830 124 119 labeled label VERB fcis-29830 124 120 as as ADP fcis-29830 124 121 negative negative ADJ fcis-29830 124 122 samples sample NOUN fcis-29830 124 123 , , PUNCT fcis-29830 124 124 which which PRON fcis-29830 124 125 are be AUX fcis-29830 124 126 actually actually ADV fcis-29830 124 127 positive positive ADJ fcis-29830 124 128 samples sample NOUN fcis-29830 124 129 predicted predict VERB fcis-29830 124 130 to to PART fcis-29830 124 131 be be AUX fcis-29830 124 132 negative negative ADJ fcis-29830 124 133 . . PUNCT fcis-29830 125 1 3.3 3.3 NUM fcis-29830 125 2 . . PUNCT fcis-29830 126 1 experimental experimental ADJ fcis-29830 126 2 results result NOUN fcis-29830 126 3 in in ADP fcis-29830 126 4 this this DET fcis-29830 126 5 experiment experiment NOUN fcis-29830 126 6 the the DET fcis-29830 126 7 dsa dsa NOUN fcis-29830 126 8 - - PUNCT fcis-29830 126 9 u++ u++ ADJ fcis-29830 126 10 model model NOUN fcis-29830 126 11 is be AUX fcis-29830 126 12 compared compare VERB fcis-29830 126 13 with with ADP fcis-29830 126 14 known know VERB fcis-29830 126 15 image image NOUN fcis-29830 126 16 classification classification NOUN fcis-29830 126 17 models model NOUN fcis-29830 126 18 and and CCONJ fcis-29830 126 19 modified modify VERB fcis-29830 126 20 image image NOUN fcis-29830 126 21 segmentation segmentation NOUN fcis-29830 126 22 models model NOUN fcis-29830 126 23 . . PUNCT fcis-29830 127 1 these these DET fcis-29830 127 2 models model NOUN fcis-29830 127 3 include include VERB fcis-29830 127 4 resnet34 resnet34 NOUN fcis-29830 127 5 , , PUNCT fcis-29830 127 6 u u NOUN fcis-29830 127 7 - - NOUN fcis-29830 127 8 net net ADJ fcis-29830 127 9 , , PUNCT fcis-29830 127 10 u u NOUN fcis-29830 127 11 - - ADJ fcis-29830 127 12 net++ net++ PROPN fcis-29830 127 13 , , PUNCT fcis-29830 127 14 vgg16 vgg16 VERB fcis-29830 127 15 [ [ NOUN fcis-29830 127 16 1 1 NUM fcis-29830 127 17 ] ] PUNCT fcis-29830 127 18 , , PUNCT fcis-29830 127 19 and and CCONJ fcis-29830 127 20 msa2net msa2net ADJ fcis-29830 127 21 [ [ X fcis-29830 127 22 2 2 NUM fcis-29830 127 23 ] ] PUNCT fcis-29830 127 24 . . PUNCT fcis-29830 128 1 the the DET fcis-29830 128 2 experimental experimental ADJ fcis-29830 128 3 results result NOUN fcis-29830 128 4 are be AUX fcis-29830 128 5 shown show VERB fcis-29830 128 6 in in ADP fcis-29830 128 7 the the DET fcis-29830 128 8 table table NOUN fcis-29830 128 9 . . PUNCT fcis-29830 129 1 table table NOUN fcis-29830 129 2 2 2 NUM fcis-29830 129 3 . . PUNCT fcis-29830 129 4 accuracy accuracy NOUN fcis-29830 129 5 , , PUNCT fcis-29830 129 6 f1 f1 NOUN fcis-29830 129 7 value value NOUN fcis-29830 129 8 results result NOUN fcis-29830 129 9 for for ADP fcis-29830 129 10 each each DET fcis-29830 129 11 model model NOUN fcis-29830 129 12 model model NOUN fcis-29830 129 13 accuracy accuracy NOUN fcis-29830 129 14 ( ( PUNCT fcis-29830 129 15 % % INTJ fcis-29830 129 16 ) ) PUNCT fcis-29830 129 17 macc(% macc(% NOUN fcis-29830 129 18 ) ) PUNCT fcis-29830 129 19 mf1(% mf1(% PROPN fcis-29830 129 20 ) ) PUNCT fcis-29830 129 21 direction direction NOUN fcis-29830 129 22 boundary boundary ADJ fcis-29830 129 23 calcification calcification NOUN fcis-29830 129 24 shape shape NOUN fcis-29830 129 25 u u NOUN fcis-29830 129 26 - - NOUN fcis-29830 129 27 net net ADJ fcis-29830 129 28 84.5 84.5 NUM fcis-29830 129 29 78.9 78.9 NUM fcis-29830 129 30 87.7 87.7 NUM fcis-29830 129 31 88.7 88.7 NUM fcis-29830 129 32 84.9 84.9 NUM fcis-29830 129 33 68.2 68.2 NUM fcis-29830 129 34 u u NOUN fcis-29830 129 35 - - PROPN fcis-29830 129 36 net++ net++ PROPN fcis-29830 129 37 84.8 84.8 NUM fcis-29830 129 38 81.6 81.6 NUM fcis-29830 129 39 84.2 84.2 NUM fcis-29830 129 40 89.6 89.6 NUM fcis-29830 129 41 85.0 85.0 NUM fcis-29830 129 42 67.4 67.4 NUM fcis-29830 129 43 vgg16 vgg16 NOUN fcis-29830 129 44 84.1 84.1 NUM fcis-29830 129 45 80.6 80.6 NUM fcis-29830 129 46 86.7 86.7 NUM fcis-29830 129 47 89.7 89.7 NUM fcis-29830 129 48 85.2 85.2 NUM fcis-29830 129 49 68.8 68.8 NUM fcis-29830 129 50 msa2net msa2net NOUN fcis-29830 129 51 84.1 84.1 NUM fcis-29830 129 52 80.4 80.4 NUM fcis-29830 129 53 83.3 83.3 NUM fcis-29830 129 54 89.1 89.1 NUM fcis-29830 129 55 84.2 84.2 NUM fcis-29830 129 56 68.8 68.8 NUM fcis-29830 129 57 dsa dsa NOUN fcis-29830 129 58 - - PUNCT fcis-29830 129 59 u++ u++ ADJ fcis-29830 129 60 85.4↑ 85.4↑ NUM fcis-29830 129 61 79.8↑ 79.8↑ NUM fcis-29830 129 62 87.1 87.1 NUM fcis-29830 129 63 89.8↑ 89.8↑ NUM fcis-29830 129 64 85.5↑ 85.5↑ NUM fcis-29830 129 65 67.5 67.5 NUM fcis-29830 129 66 as as SCONJ fcis-29830 129 67 shown show VERB fcis-29830 129 68 in in ADP fcis-29830 129 69 table table NOUN fcis-29830 129 70 2 2 NUM fcis-29830 129 71 , , PUNCT fcis-29830 129 72 the the DET fcis-29830 129 73 improved improved ADJ fcis-29830 129 74 algorithm algorithm NOUN fcis-29830 129 75 based base VERB fcis-29830 129 76 on on ADP fcis-29830 129 77 resnet resnet NOUN fcis-29830 129 78 - - PUNCT fcis-29830 129 79 u u NOUN fcis-29830 129 80 - - NOUN fcis-29830 129 81 net++ net++ NOUN fcis-29830 129 82 for for ADP fcis-29830 129 83 image image NOUN fcis-29830 129 84 classification classification NOUN fcis-29830 129 85 is be AUX fcis-29830 129 86 also also ADV fcis-29830 129 87 effective effective ADJ fcis-29830 129 88 and and CCONJ fcis-29830 129 89 accurate accurate ADJ fcis-29830 129 90 . . PUNCT fcis-29830 130 1 compared compare VERB fcis-29830 130 2 to to ADP fcis-29830 130 3 the the DET fcis-29830 130 4 common common ADJ fcis-29830 130 5 image image NOUN fcis-29830 130 6 classification classification NOUN fcis-29830 130 7 algorithm algorithm NOUN fcis-29830 130 8 vgg16 vgg16 NOUN fcis-29830 130 9 , , PUNCT fcis-29830 130 10 there there PRON fcis-29830 130 11 is be VERB fcis-29830 130 12 also also ADV fcis-29830 130 13 some some DET fcis-29830 130 14 improvement improvement NOUN fcis-29830 130 15 . . PUNCT fcis-29830 131 1 3.4 3.4 NUM fcis-29830 131 2 . . PUNCT fcis-29830 131 3 conclusion conclusion NOUN fcis-29830 131 4 in in ADP fcis-29830 131 5 this this DET fcis-29830 131 6 paper paper NOUN fcis-29830 131 7 , , PUNCT fcis-29830 131 8 we we PRON fcis-29830 131 9 propose propose VERB fcis-29830 131 10 a a DET fcis-29830 131 11 deep deep ADJ fcis-29830 131 12 learning learning NOUN fcis-29830 131 13 model model NOUN fcis-29830 131 14 , , PUNCT fcis-29830 131 15 dsa dsa NOUN fcis-29830 131 16 - - PUNCT fcis-29830 131 17 u++ u++ NOUN fcis-29830 131 18 , , PUNCT fcis-29830 131 19 based base VERB fcis-29830 131 20 on on ADP fcis-29830 131 21 resnet34 resnet34 NOUN fcis-29830 131 22 and and CCONJ fcis-29830 131 23 u u NOUN fcis-29830 131 24 - - NOUN fcis-29830 131 25 net++ net++ ADJ fcis-29830 131 26 improvement improvement NOUN fcis-29830 131 27 for for ADP fcis-29830 131 28 feature feature NOUN fcis-29830 131 29 classification classification NOUN fcis-29830 131 30 of of ADP fcis-29830 131 31 ultrasound ultrasound ADJ fcis-29830 131 32 breast breast NOUN fcis-29830 131 33 images image NOUN fcis-29830 131 34 . . PUNCT fcis-29830 132 1 the the DET fcis-29830 132 2 model model NOUN fcis-29830 132 3 employs employ VERB fcis-29830 132 4 resnet34 resnet34 NOUN fcis-29830 132 5 as as ADP fcis-29830 132 6 an an DET fcis-29830 132 7 encoder encoder NOUN fcis-29830 132 8 to to PART fcis-29830 132 9 efficiently efficiently ADV fcis-29830 132 10 extract extract VERB fcis-29830 132 11 multiscale multiscale NOUN fcis-29830 132 12 features feature NOUN fcis-29830 132 13 in in ADP fcis-29830 132 14 the the DET fcis-29830 132 15 image image NOUN fcis-29830 132 16 , , PUNCT fcis-29830 132 17 and and CCONJ fcis-29830 132 18 also also ADV fcis-29830 132 19 introduces introduce VERB fcis-29830 132 20 the the DET fcis-29830 132 21 ras ras NOUN fcis-29830 132 22 module module NOUN fcis-29830 132 23 designed design VERB fcis-29830 132 24 by by ADP fcis-29830 132 25 the the DET fcis-29830 132 26 convolutional convolutional ADJ fcis-29830 132 27 pooling pool VERB fcis-29830 132 28 pyramid pyramid NOUN fcis-29830 132 29 in in ADP fcis-29830 132 30 the the DET fcis-29830 132 31 null null ADJ fcis-29830 132 32 space space NOUN fcis-29830 132 33 to to PART fcis-29830 132 34 enhance enhance VERB fcis-29830 132 35 the the DET fcis-29830 132 36 representation representation NOUN fcis-29830 132 37 of of ADP fcis-29830 132 38 multiscale multiscale ADJ fcis-29830 132 39 contextual contextual ADJ fcis-29830 132 40 information information NOUN fcis-29830 132 41 . . PUNCT fcis-29830 133 1 subsequently subsequently ADV fcis-29830 133 2 , , PUNCT fcis-29830 133 3 in in ADP fcis-29830 133 4 the the DET fcis-29830 133 5 decoder decoder NOUN fcis-29830 133 6 part part NOUN fcis-29830 133 7 , , PUNCT fcis-29830 133 8 based base VERB fcis-29830 133 9 on on ADP fcis-29830 133 10 the the DET fcis-29830 133 11 improved improve VERB fcis-29830 133 12 das das PROPN fcis-29830 133 13 - - PUNCT fcis-29830 133 14 u++ u++ ADJ fcis-29830 133 15 model model NOUN fcis-29830 133 16 , , PUNCT fcis-29830 133 17 we we PRON fcis-29830 133 18 replace replace VERB fcis-29830 133 19 part part NOUN fcis-29830 133 20 of of ADP fcis-29830 133 21 the the DET fcis-29830 133 22 traditional traditional ADJ fcis-29830 133 23 convolutional convolutional ADJ fcis-29830 133 24 blocks block NOUN fcis-29830 133 25 with with ADP fcis-29830 133 26 depth depth NOUN fcis-29830 133 27 - - PUNCT fcis-29830 133 28 separable separable NOUN fcis-29830 133 29 convolutions convolution NOUN fcis-29830 133 30 , , PUNCT fcis-29830 133 31 which which PRON fcis-29830 133 32 not not PART fcis-29830 133 33 only only ADV fcis-29830 133 34 reduces reduce VERB fcis-29830 133 35 the the DET fcis-29830 133 36 number number NOUN fcis-29830 133 37 of of ADP fcis-29830 133 38 model model NOUN fcis-29830 133 39 parameters parameter NOUN fcis-29830 133 40 and and CCONJ fcis-29830 133 41 the the DET fcis-29830 133 42 computational computational ADJ fcis-29830 133 43 complexity complexity NOUN fcis-29830 133 44 while while SCONJ fcis-29830 133 45 maintaining maintain VERB fcis-29830 133 46 the the DET fcis-29830 133 47 feature feature NOUN fcis-29830 133 48 expressiveness expressiveness NOUN fcis-29830 133 49 , , PUNCT fcis-29830 133 50 but but CCONJ fcis-29830 133 51 also also ADV fcis-29830 133 52 preserves preserve VERB fcis-29830 133 53 the the DET fcis-29830 133 54 detailed detailed ADJ fcis-29830 133 55 information information NOUN fcis-29830 133 56 of of ADP fcis-29830 133 57 the the DET fcis-29830 133 58 image image NOUN fcis-29830 133 59 more more ADV fcis-29830 133 60 adequately adequately ADV fcis-29830 133 61 . . PUNCT fcis-29830 134 1 the the DET fcis-29830 134 2 dense dense ADJ fcis-29830 134 3 hopping hop VERB fcis-29830 134 4 connections connection NOUN fcis-29830 134 5 of of ADP fcis-29830 134 6 u u PROPN fcis-29830 134 7 - - PROPN fcis-29830 134 8 net++ net++ PROPN fcis-29830 134 9 also also ADV fcis-29830 134 10 help help VERB fcis-29830 134 11 to to PART fcis-29830 134 12 fuse fuse VERB fcis-29830 134 13 the the DET fcis-29830 134 14 different different ADJ fcis-29830 134 15 levels level NOUN fcis-29830 134 16 of of ADP fcis-29830 134 17 features feature NOUN fcis-29830 134 18 , , PUNCT fcis-29830 134 19 enabling enable VERB fcis-29830 134 20 the the DET fcis-29830 134 21 model model NOUN fcis-29830 134 22 to to PART fcis-29830 134 23 can can AUX fcis-29830 134 24 utilize utilize VERB fcis-29830 134 25 more more ADJ fcis-29830 134 26 contextual contextual ADJ fcis-29830 134 27 information information NOUN fcis-29830 134 28 . . PUNCT fcis-29830 135 1 with with ADP fcis-29830 135 2 the the DET fcis-29830 135 3 above above ADJ fcis-29830 135 4 improved improved ADJ fcis-29830 135 5 design design NOUN fcis-29830 135 6 , , PUNCT fcis-29830 135 7 the the DET fcis-29830 135 8 dsa dsa NOUN fcis-29830 135 9 - - PUNCT fcis-29830 135 10 u++ u++ ADJ fcis-29830 135 11 model model NOUN fcis-29830 135 12 is be AUX fcis-29830 135 13 able able ADJ fcis-29830 135 14 to to PART fcis-29830 135 15 capture capture VERB fcis-29830 135 16 subtle subtle ADJ fcis-29830 135 17 lesion lesion NOUN fcis-29830 135 18 features feature NOUN fcis-29830 135 19 in in ADP fcis-29830 135 20 ultrasound ultrasound ADJ fcis-29830 135 21 breast breast NOUN fcis-29830 135 22 images image NOUN fcis-29830 135 23 more more ADV fcis-29830 135 24 comprehensively comprehensively ADV fcis-29830 135 25 and and CCONJ fcis-29830 135 26 achieve achieve VERB fcis-29830 135 27 accurate accurate ADJ fcis-29830 135 28 classification classification NOUN fcis-29830 135 29 based base VERB fcis-29830 135 30 on on ADP fcis-29830 135 31 multi multi ADJ fcis-29830 135 32 - - ADJ fcis-29830 135 33 level level ADJ fcis-29830 135 34 feature feature NOUN fcis-29830 135 35 fusion fusion NOUN fcis-29830 135 36 . . PUNCT fcis-29830 136 1 the the DET fcis-29830 136 2 experimental experimental ADJ fcis-29830 136 3 results result NOUN fcis-29830 136 4 show show VERB fcis-29830 136 5 that that SCONJ fcis-29830 136 6 the the DET fcis-29830 136 7 model model NOUN fcis-29830 136 8 also also ADV fcis-29830 136 9 improves improve VERB fcis-29830 136 10 in in ADP fcis-29830 136 11 classification classification NOUN fcis-29830 136 12 accuracy accuracy NOUN fcis-29830 136 13 compared compare VERB fcis-29830 136 14 to to ADP fcis-29830 136 15 the the DET fcis-29830 136 16 traditional traditional ADJ fcis-29830 136 17 model model NOUN fcis-29830 136 18 , , PUNCT fcis-29830 136 19 which which PRON fcis-29830 136 20 verifies verify VERB fcis-29830 136 21 the the DET fcis-29830 136 22 effectiveness effectiveness NOUN fcis-29830 136 23 of of ADP fcis-29830 136 24 the the DET fcis-29830 136 25 proposed propose VERB fcis-29830 136 26 method method NOUN fcis-29830 136 27 in in ADP fcis-29830 136 28 medical medical ADJ fcis-29830 136 29 image image NOUN fcis-29830 136 30 feature feature NOUN fcis-29830 136 31 extraction extraction NOUN fcis-29830 136 32 and and CCONJ fcis-29830 136 33 classification classification NOUN fcis-29830 136 34 tasks task NOUN fcis-29830 136 35 . . PUNCT fcis-29830 137 1 acknowledgments acknowledgment NOUN fcis-29830 137 2 funded fund VERB fcis-29830 137 3 project project NOUN fcis-29830 137 4 : : PUNCT fcis-29830 137 5 2025 2025 NUM fcis-29830 137 6 student student NOUN fcis-29830 137 7 innovation innovation NOUN fcis-29830 137 8 and and CCONJ fcis-29830 137 9 entrepreneurship entrepreneurship NOUN fcis-29830 137 10 training training NOUN fcis-29830 137 11 program program NOUN fcis-29830 137 12 project project NOUN fcis-29830 137 13 ( ( PUNCT fcis-29830 137 14 feature feature NOUN fcis-29830 137 15 recognition recognition NOUN fcis-29830 137 16 of of ADP fcis-29830 137 17 ultrasound ultrasound ADJ fcis-29830 137 18 breast breast NOUN fcis-29830 137 19 images image NOUN fcis-29830 137 20 based base VERB fcis-29830 137 21 on on ADP fcis-29830 137 22 unet++ unet++ PROPN fcis-29830 137 23 ) ) PUNCT fcis-29830 137 24 . . PUNCT fcis-29830 138 1 references reference NOUN fcis-29830 138 2 [ [ X fcis-29830 138 3 1 1 X fcis-29830 138 4 ] ] PUNCT fcis-29830 138 5 j. j. PROPN fcis-29830 138 6 ferlay ferlay PROPN fcis-29830 138 7 , , PUNCT fcis-29830 138 8 m. m. NOUN fcis-29830 138 9 colombet colombet PROPN fcis-29830 138 10 , , PUNCT fcis-29830 138 11 i. i. PROPN fcis-29830 138 12 soerjomataram soerjomataram PROPN fcis-29830 138 13 , , PUNCT fcis-29830 138 14 d.m d.m PROPN fcis-29830 138 15 . . PUNCT fcis-29830 138 16 parkin parkin PROPN fcis-29830 138 17 , , PUNCT fcis-29830 138 18 m. m. NOUN fcis-29830 138 19 piñeros piñero NOUN fcis-29830 138 20 , , PUNCT fcis-29830 138 21 a. a. NOUN fcis-29830 138 22 znaor znaor PROPN fcis-29830 138 23 , , PUNCT fcis-29830 138 24 f. f. PROPN fcis-29830 138 25 bray bray PROPN fcis-29830 138 26 , , PUNCT fcis-29830 138 27 cancer cancer NOUN fcis-29830 138 28 statistics statistic NOUN fcis-29830 138 29 for for ADP fcis-29830 138 30 the the DET fcis-29830 138 31 year year NOUN fcis-29830 138 32 2020 2020 NUM fcis-29830 138 33 : : PUNCT fcis-29830 138 34 an an DET fcis-29830 138 35 overview overview NOUN fcis-29830 138 36 , , PUNCT fcis-29830 138 37 international international ADJ fcis-29830 138 38 journal journal NOUN fcis-29830 138 39 of of ADP fcis-29830 138 40 cancer cancer NOUN fcis-29830 138 41 ( ( PUNCT fcis-29830 138 42 2021 2021 NUM fcis-29830 138 43 ) ) PUNCT fcis-29830 138 44 . . PUNCT fcis-29830 139 1 [ [ X fcis-29830 139 2 2 2 NUM fcis-29830 139 3 ] ] PUNCT fcis-29830 139 4 i. i. PROPN fcis-29830 139 5 soerjomataram soerjomataram PROPN fcis-29830 139 6 , , PUNCT fcis-29830 139 7 f.j.n.r.c.o f.j.n.r.c.o PROPN fcis-29830 139 8 . . PROPN fcis-29830 139 9 bray bray PROPN fcis-29830 139 10 , , PUNCT fcis-29830 139 11 planning plan VERB fcis-29830 139 12 for for ADP fcis-29830 139 13 tomorrow tomorrow NOUN fcis-29830 139 14 : : PUNCT fcis-29830 139 15 global global ADJ fcis-29830 139 16 cancer cancer NOUN fcis-29830 139 17 incidence incidence NOUN fcis-29830 139 18 and and CCONJ fcis-29830 139 19 the the DET fcis-29830 139 20 role role NOUN fcis-29830 139 21 of of ADP fcis-29830 139 22 prevention prevention NOUN fcis-29830 139 23 2020–2070 2020–2070 NUM fcis-29830 139 24 , , PUNCT fcis-29830 139 25 18(10 18(10 NUM fcis-29830 139 26 ) ) PUNCT fcis-29830 139 27 ( ( PUNCT fcis-29830 139 28 2021 2021 NUM fcis-29830 139 29 ) ) PUNCT fcis-29830 139 30 663 663 NUM fcis-29830 139 31 - - SYM fcis-29830 139 32 672 672 NUM fcis-29830 139 33 . . PUNCT fcis-29830 140 1 [ [ X fcis-29830 140 2 3 3 X fcis-29830 140 3 ] ] X fcis-29830 140 4 h. h. PROPN fcis-29830 140 5 sung sung PROPN fcis-29830 140 6 , , PUNCT fcis-29830 140 7 j. j. PROPN fcis-29830 140 8 ferlay ferlay PROPN fcis-29830 140 9 , , PUNCT fcis-29830 140 10 r.l r.l PROPN fcis-29830 140 11 . . PROPN fcis-29830 140 12 siegel siegel PROPN fcis-29830 140 13 , , PUNCT fcis-29830 140 14 m. m. NOUN fcis-29830 140 15 laversanne laversanne PROPN fcis-29830 140 16 , , PUNCT fcis-29830 140 17 i. i. PROPN fcis-29830 140 18 soerjomataram soerjomataram PROPN fcis-29830 140 19 , , PUNCT fcis-29830 140 20 a. a. NOUN fcis-29830 140 21 jemal jemal PROPN fcis-29830 140 22 , , PUNCT fcis-29830 140 23 f. f. PROPN fcis-29830 140 24 bray bray PROPN fcis-29830 140 25 , , PUNCT fcis-29830 140 26 global global ADJ fcis-29830 140 27 cancer cancer NOUN fcis-29830 140 28 statistics statistic NOUN fcis-29830 140 29 2020 2020 NUM fcis-29830 140 30 : : PUNCT fcis-29830 140 31 globocan globocan PROPN fcis-29830 140 32 estimates estimate NOUN fcis-29830 140 33 of of ADP fcis-29830 140 34 incidence incidence NOUN fcis-29830 140 35 and and CCONJ fcis-29830 140 36 mortality mortality NOUN fcis-29830 140 37 worldwide worldwide ADV fcis-29830 140 38 for for ADP fcis-29830 140 39 36 36 NUM fcis-29830 140 40 cancers cancer NOUN fcis-29830 140 41 in in ADP fcis-29830 140 42 185 185 NUM fcis-29830 140 43 countries country NOUN fcis-29830 140 44 , , PUNCT fcis-29830 140 45 ca ca NOUN fcis-29830 140 46 : : PUNCT fcis-29830 140 47 a a DET fcis-29830 140 48 cancer cancer NOUN fcis-29830 140 49 journal journal NOUN fcis-29830 140 50 for for ADP fcis-29830 140 51 clinicians clinician NOUN fcis-29830 140 52 71(3 71(3 NUM fcis-29830 140 53 ) ) PUNCT fcis-29830 140 54 ( ( PUNCT fcis-29830 140 55 2021 2021 NUM fcis-29830 140 56 ) ) PUNCT fcis-29830 140 57 209 209 NUM fcis-29830 140 58 - - SYM fcis-29830 140 59 249 249 NUM fcis-29830 140 60 . . PUNCT fcis-29830 141 1 [ [ X fcis-29830 141 2 4 4 NUM fcis-29830 141 3 ] ] X fcis-29830 141 4 r. r. PROPN fcis-29830 141 5 guo guo PROPN fcis-29830 141 6 , , PUNCT fcis-29830 141 7 g. g. PROPN fcis-29830 141 8 lu lu PROPN fcis-29830 141 9 , , PUNCT fcis-29830 141 10 b. b. PROPN fcis-29830 141 11 qin qin PROPN fcis-29830 141 12 , , PUNCT fcis-29830 141 13 b. b. PROPN fcis-29830 141 14 fei fei PROPN fcis-29830 141 15 , , PUNCT fcis-29830 141 16 ultrasound ultrasound ADJ fcis-29830 141 17 imaging imaging NOUN fcis-29830 141 18 technologies technology NOUN fcis-29830 141 19 for for ADP fcis-29830 141 20 breast breast NOUN fcis-29830 141 21 cancer cancer NOUN fcis-29830 141 22 detection detection NOUN fcis-29830 141 23 and and CCONJ fcis-29830 141 24 management management NOUN fcis-29830 141 25 : : PUNCT fcis-29830 141 26 a a DET fcis-29830 141 27 review review NOUN fcis-29830 141 28 , , PUNCT fcis-29830 141 29 ultrasound ultrasound NOUN fcis-29830 141 30 in in ADP fcis-29830 141 31 medicine medicine NOUN fcis-29830 141 32 & & CCONJ fcis-29830 141 33 biology biology NOUN fcis-29830 141 34 44(1 44(1 PROPN fcis-29830 141 35 ) ) PUNCT fcis-29830 141 36 ( ( PUNCT fcis-29830 141 37 2018 2018 NUM fcis-29830 141 38 ) ) PUNCT fcis-29830 141 39 3770 3770 NUM fcis-29830 141 40 . . PUNCT fcis-29830 142 1 [ [ X fcis-29830 142 2 5 5 NUM fcis-29830 142 3 ] ] X fcis-29830 142 4 s.c s.c PROPN fcis-29830 142 5 . . PROPN fcis-29830 142 6 chen chen PROPN fcis-29830 142 7 , , PUNCT fcis-29830 142 8 y.c y.c PROPN fcis-29830 142 9 . . PROPN fcis-29830 142 10 cheung cheung PROPN fcis-29830 142 11 , , PUNCT fcis-29830 142 12 c.h c.h PROPN fcis-29830 142 13 . . PROPN fcis-29830 142 14 su su PROPN fcis-29830 142 15 , , PUNCT fcis-29830 142 16 m.f m.f PROPN fcis-29830 142 17 . . PUNCT fcis-29830 143 1 chen chen PROPN fcis-29830 143 2 , , PUNCT fcis-29830 143 3 t.l t.l PROPN fcis-29830 143 4 . . PUNCT fcis-29830 143 5 hwang hwang PROPN fcis-29830 143 6 , , PUNCT fcis-29830 143 7 s.j.u.i.o s.j.u.i.o PROPN fcis-29830 143 8 . . PUNCT fcis-29830 144 1 hsueh hsueh PROPN fcis-29830 144 2 , , PUNCT fcis-29830 144 3 g.t.o.j.o.t.i.s.o.u.i g.t.o.j.o.t.i.s.o.u.i PROPN fcis-29830 144 4 . . PUNCT fcis-29830 145 1 obstetrics obstetric NOUN fcis-29830 145 2 , , PUNCT fcis-29830 145 3 gynecology gynecology NOUN fcis-29830 145 4 , , PUNCT fcis-29830 145 5 analysis analysis NOUN fcis-29830 145 6 of of ADP fcis-29830 145 7 sonographic sonographic ADJ fcis-29830 145 8 features feature NOUN fcis-29830 145 9 for for ADP fcis-29830 145 10 the the DET fcis-29830 145 11 differentiation differentiation NOUN fcis-29830 145 12 of of ADP fcis-29830 145 13 88 88 NUM fcis-29830 145 14 benign benign ADJ fcis-29830 145 15 and and CCONJ fcis-29830 145 16 malignant malignant ADJ fcis-29830 145 17 breast breast NOUN fcis-29830 145 18 tumors tumor NOUN fcis-29830 145 19 of of ADP fcis-29830 145 20 different different ADJ fcis-29830 145 21 sizes size NOUN fcis-29830 145 22 , , PUNCT fcis-29830 145 23 23(2 23(2 NUM fcis-29830 145 24 ) ) PUNCT fcis-29830 145 25 ( ( PUNCT fcis-29830 145 26 2004 2004 NUM fcis-29830 145 27 ) ) PUNCT fcis-29830 145 28 188 188 NUM fcis-29830 145 29 - - SYM fcis-29830 145 30 193 193 NUM fcis-29830 145 31 . . PUNCT fcis-29830 146 1 [ [ X fcis-29830 146 2 6 6 NUM fcis-29830 146 3 ] ] X fcis-29830 146 4 s.t s.t PROPN fcis-29830 146 5 . . PROPN fcis-29830 146 6 chen chen PROPN fcis-29830 146 7 , , PUNCT fcis-29830 146 8 y.h y.h PROPN fcis-29830 146 9 . . PROPN fcis-29830 146 10 hsiao hsiao PROPN fcis-29830 146 11 , , PUNCT fcis-29830 146 12 y.l y.l PROPN fcis-29830 146 13 . . PROPN fcis-29830 146 14 huang huang PROPN fcis-29830 146 15 , , PUNCT fcis-29830 146 16 s.j s.j PROPN fcis-29830 146 17 . . PROPN fcis-29830 146 18 kuo kuo PROPN fcis-29830 146 19 , , PUNCT fcis-29830 146 20 h.s h.s PROPN fcis-29830 146 21 . . PROPN fcis-29830 146 22 tseng tseng PROPN fcis-29830 146 23 , , PUNCT fcis-29830 146 24 h.k h.k PROPN fcis-29830 146 25 . . PROPN fcis-29830 146 26 wu wu PROPN fcis-29830 146 27 , , PUNCT fcis-29830 146 28 d.r d.r PROPN fcis-29830 146 29 . . PROPN fcis-29830 146 30 chen chen PROPN fcis-29830 146 31 , , PUNCT fcis-29830 146 32 comparative comparative ADJ fcis-29830 146 33 analysis analysis NOUN fcis-29830 146 34 of of ADP fcis-29830 146 35 logistic logistic ADJ fcis-29830 146 36 regression regression NOUN fcis-29830 146 37 , , PUNCT fcis-29830 146 38 support support NOUN fcis-29830 146 39 vector vector NOUN fcis-29830 146 40 machine machine NOUN fcis-29830 146 41 and and CCONJ fcis-29830 146 42 artificial artificial ADJ fcis-29830 146 43 neural neural ADJ fcis-29830 146 44 network network NOUN fcis-29830 146 45 for for ADP fcis-29830 146 46 the the DET fcis-29830 146 47 differential differential ADJ fcis-29830 146 48 diagnosis diagnosis NOUN fcis-29830 146 49 of of ADP fcis-29830 146 50 benign benign ADJ fcis-29830 146 51 and and CCONJ fcis-29830 146 52 malignant malignant ADJ fcis-29830 146 53 solid solid ADJ fcis-29830 146 54 breast breast NOUN fcis-29830 146 55 tumors tumor NOUN fcis-29830 146 56 by by ADP fcis-29830 146 57 the the DET fcis-29830 146 58 use use NOUN fcis-29830 146 59 of of ADP fcis-29830 146 60 three three NUM fcis-29830 146 61 - - PUNCT fcis-29830 146 62 dimensional dimensional ADJ fcis-29830 146 63 power power NOUN fcis-29830 146 64 doppler doppler NOUN fcis-29830 146 65 imaging imaging NOUN fcis-29830 146 66 , , PUNCT fcis-29830 146 67 korean korean ADJ fcis-29830 146 68 journal journal NOUN fcis-29830 146 69 of of ADP fcis-29830 146 70 radiology radiology NOUN fcis-29830 146 71 10(5 10(5 NOUN fcis-29830 146 72 ) ) PUNCT fcis-29830 146 73 ( ( PUNCT fcis-29830 146 74 2009 2009 NUM fcis-29830 146 75 ) ) PUNCT fcis-29830 146 76 464 464 NUM fcis-29830 146 77 - - SYM fcis-29830 146 78 71 71 NUM fcis-29830 146 79 . . PUNCT fcis-29830 147 1 [ [ X fcis-29830 147 2 7 7 X fcis-29830 147 3 ] ] X fcis-29830 147 4 j. j. PROPN fcis-29830 147 5 redmon redmon PROPN fcis-29830 147 6 , , PUNCT fcis-29830 147 7 s.k s.k PROPN fcis-29830 147 8 . . PROPN fcis-29830 147 9 divvala divvala PROPN fcis-29830 147 10 , , PUNCT fcis-29830 147 11 r.b r.b PROPN fcis-29830 147 12 . . PROPN fcis-29830 147 13 girshick girshick PROPN fcis-29830 147 14 , , PUNCT fcis-29830 147 15 a.j.i.c.o.c.v a.j.i.c.o.c.v PUNCT fcis-29830 147 16 . . PUNCT fcis-29830 148 1 farhadi farhadi NOUN fcis-29830 148 2 , , PUNCT fcis-29830 148 3 p. p. NOUN fcis-29830 148 4 recognition recognition NOUN fcis-29830 148 5 , , PUNCT fcis-29830 148 6 you you PRON fcis-29830 148 7 only only ADV fcis-29830 148 8 look look VERB fcis-29830 148 9 once once ADV fcis-29830 148 10 : : PUNCT fcis-29830 148 11 unified unified ADJ fcis-29830 148 12 , , PUNCT fcis-29830 148 13 realtime realtime ADJ fcis-29830 148 14 object object NOUN fcis-29830 148 15 detection detection NOUN fcis-29830 148 16 , , PUNCT fcis-29830 148 17 ( ( PUNCT fcis-29830 148 18 2015 2015 NUM fcis-29830 148 19 ) ) PUNCT fcis-29830 148 20 779 779 NUM fcis-29830 148 21 - - SYM fcis-29830 148 22 788 788 NUM fcis-29830 148 23 . . PUNCT fcis-29830 149 1 [ [ X fcis-29830 149 2 8 8 NUM fcis-29830 149 3 ] ] X fcis-29830 149 4 i.j i.j PROPN fcis-29830 149 5 . . PROPN fcis-29830 149 6 goodfellow goodfellow PROPN fcis-29830 149 7 , , PUNCT fcis-29830 149 8 j. j. PROPN fcis-29830 149 9 pouget pouget PROPN fcis-29830 149 10 - - PUNCT fcis-29830 149 11 abadie abadie ADJ fcis-29830 149 12 , , PUNCT fcis-29830 149 13 m. m. NOUN fcis-29830 149 14 mirza mirza PROPN fcis-29830 149 15 , , PUNCT fcis-29830 149 16 b. b. PROPN fcis-29830 149 17 xu xu PROPN fcis-29830 149 18 , , PUNCT fcis-29830 149 19 d. d. PROPN fcis-29830 149 20 wardefarley wardefarley PROPN fcis-29830 149 21 , , PUNCT fcis-29830 149 22 s. s. PROPN fcis-29830 149 23 ozair ozair PROPN fcis-29830 149 24 , , PUNCT fcis-29830 149 25 a.c a.c PROPN fcis-29830 149 26 . . PROPN fcis-29830 149 27 courville courville NOUN fcis-29830 149 28 , , PUNCT fcis-29830 149 29 y. y. PROPN fcis-29830 149 30 bengio bengio PROPN fcis-29830 149 31 , , PUNCT fcis-29830 149 32 generative generative ADJ fcis-29830 149 33 adversarial adversarial ADJ fcis-29830 149 34 nets net NOUN fcis-29830 149 35 , , PUNCT fcis-29830 149 36 neural neural ADJ fcis-29830 149 37 information information NOUN fcis-29830 149 38 processing processing NOUN fcis-29830 149 39 systems system NOUN fcis-29830 149 40 , , PUNCT fcis-29830 149 41 2014 2014 NUM fcis-29830 149 42 . . PUNCT fcis-29830 150 1 [ [ X fcis-29830 150 2 9 9 NUM fcis-29830 150 3 ] ] PUNCT fcis-29830 150 4 a. a. NOUN fcis-29830 150 5 vaswani vaswani NOUN fcis-29830 150 6 , , PUNCT fcis-29830 150 7 n.m n.m PROPN fcis-29830 150 8 . . PROPN fcis-29830 150 9 shazeer shazeer PROPN fcis-29830 150 10 , , PUNCT fcis-29830 150 11 n. n. PROPN fcis-29830 150 12 parmar parmar PROPN fcis-29830 150 13 , , PUNCT fcis-29830 150 14 j. j. PROPN fcis-29830 150 15 uszkoreit uszkoreit PROPN fcis-29830 150 16 , , PUNCT fcis-29830 150 17 l. l. PROPN fcis-29830 150 18 jones jones PROPN fcis-29830 150 19 , , PUNCT fcis-29830 150 20 a.n a.n PROPN fcis-29830 150 21 . . PROPN fcis-29830 150 22 gomez gomez PROPN fcis-29830 150 23 , , PUNCT fcis-29830 150 24 l. l. PROPN fcis-29830 150 25 kaiser kaiser PROPN fcis-29830 150 26 , , PUNCT fcis-29830 150 27 i. i. PROPN fcis-29830 150 28 polosukhin polosukhin PROPN fcis-29830 150 29 , , PUNCT fcis-29830 150 30 attention attention NOUN fcis-29830 150 31 is be AUX fcis-29830 150 32 all all PRON fcis-29830 150 33 you you PRON fcis-29830 150 34 need need VERB fcis-29830 150 35 , , PUNCT fcis-29830 150 36 neural neural ADJ fcis-29830 150 37 information information NOUN fcis-29830 150 38 processing processing NOUN fcis-29830 150 39 systems system NOUN fcis-29830 150 40 , , PUNCT fcis-29830 150 41 2017 2017 NUM fcis-29830 150 42 . . PUNCT fcis-29830 151 1 [ [ X fcis-29830 151 2 10 10 NUM fcis-29830 151 3 ] ] PUNCT fcis-29830 151 4 j. j. PROPN fcis-29830 151 5 devlin devlin PROPN fcis-29830 151 6 , , PUNCT fcis-29830 151 7 m.-w m.-w PROPN fcis-29830 151 8 . . PUNCT fcis-29830 152 1 chang chang PROPN fcis-29830 152 2 , , PUNCT fcis-29830 152 3 k. k. PROPN fcis-29830 152 4 lee lee PROPN fcis-29830 152 5 , , PUNCT fcis-29830 152 6 k. k. PROPN fcis-29830 152 7 toutanova toutanova PROPN fcis-29830 152 8 , , PUNCT fcis-29830 152 9 bert bert PROPN fcis-29830 152 10 : : PUNCT fcis-29830 152 11 pretraining pretraine VERB fcis-29830 152 12 of of ADP fcis-29830 152 13 deep deep ADJ fcis-29830 152 14 bidirectional bidirectional ADJ fcis-29830 152 15 transformers transformer NOUN fcis-29830 152 16 for for ADP fcis-29830 152 17 language language NOUN fcis-29830 152 18 understanding understanding NOUN fcis-29830 152 19 , , PUNCT fcis-29830 152 20 north north ADJ fcis-29830 152 21 american american ADJ fcis-29830 152 22 chapter chapter NOUN fcis-29830 152 23 of of ADP fcis-29830 152 24 the the DET fcis-29830 152 25 association association NOUN fcis-29830 152 26 for for ADP fcis-29830 152 27 computational computational ADJ fcis-29830 152 28 linguistics linguistic NOUN fcis-29830 152 29 , , PUNCT fcis-29830 152 30 2019 2019 NUM fcis-29830 152 31 . . PUNCT fcis-29830 153 1 [ [ X fcis-29830 153 2 11 11 NUM fcis-29830 153 3 ] ] PUNCT fcis-29830 153 4 t. t. NOUN fcis-29830 153 5 steifer steifer NOUN fcis-29830 153 6 , , PUNCT fcis-29830 153 7 m. m. NOUN fcis-29830 153 8 lewandowski lewandowski PROPN fcis-29830 153 9 , , PUNCT fcis-29830 153 10 ultrasound ultrasound ADJ fcis-29830 153 11 tissue tissue NOUN fcis-29830 153 12 characterization characterization NOUN fcis-29830 153 13 based base VERB fcis-29830 153 14 on on ADP fcis-29830 153 15 the the DET fcis-29830 153 16 lempel lempel PROPN fcis-29830 153 17 – – PUNCT fcis-29830 153 18 ziv ziv PROPN fcis-29830 153 19 complexity complexity NOUN fcis-29830 153 20 with with ADP fcis-29830 153 21 application application NOUN fcis-29830 153 22 to to ADP fcis-29830 153 23 breast breast NOUN fcis-29830 153 24 lesion lesion NOUN fcis-29830 153 25 classification classification NOUN fcis-29830 153 26 , , PUNCT fcis-29830 153 27 biomedical biomedical ADJ fcis-29830 153 28 signal signal NOUN fcis-29830 153 29 processing processing NOUN fcis-29830 153 30 and and CCONJ fcis-29830 153 31 control control NOUN fcis-29830 153 32 51 51 NUM fcis-29830 153 33 ( ( PUNCT fcis-29830 153 34 2019 2019 NUM fcis-29830 153 35 ) ) PUNCT fcis-29830 153 36 235 235 NUM fcis-29830 153 37 - - SYM fcis-29830 153 38 242 242 NUM fcis-29830 153 39 . . PUNCT fcis-29830 154 1 [ [ X fcis-29830 154 2 12 12 NUM fcis-29830 154 3 ] ] X fcis-29830 154 4 o. o. NOUN fcis-29830 154 5 ronneberger ronneberger NOUN fcis-29830 154 6 , , PUNCT fcis-29830 154 7 p. p. NOUN fcis-29830 154 8 fischer fischer PROPN fcis-29830 154 9 , , PUNCT fcis-29830 154 10 t.j.a t.j.a PROPN fcis-29830 154 11 . . PUNCT fcis-29830 154 12 brox brox PROPN fcis-29830 154 13 , , PUNCT fcis-29830 154 14 u u NOUN fcis-29830 154 15 - - NOUN fcis-29830 154 16 net net ADJ fcis-29830 154 17 : : PUNCT fcis-29830 154 18 convolutional convolutional ADJ fcis-29830 154 19 networks network NOUN fcis-29830 154 20 for for ADP fcis-29830 154 21 biomedical biomedical ADJ fcis-29830 154 22 image image NOUN fcis-29830 154 23 segmentation segmentation NOUN fcis-29830 154 24 , , PUNCT fcis-29830 154 25 abs/1505 abs/1505 NOUN fcis-29830 154 26 . . PUNCT fcis-29830 155 1 04597 04597 NUM fcis-29830 155 2 ( ( PUNCT fcis-29830 155 3 2015 2015 NUM fcis-29830 155 4 ) ) PUNCT fcis-29830 155 5 . . PUNCT fcis-29830 156 1 [ [ X fcis-29830 156 2 13 13 NUM fcis-29830 156 3 ] ] PUNCT fcis-29830 156 4 ö. ö. PROPN fcis-29830 156 5 çiçek çiçek PROPN fcis-29830 156 6 , , PUNCT fcis-29830 156 7 a. a. NOUN fcis-29830 156 8 abdulkadir abdulkadir PROPN fcis-29830 156 9 , , PUNCT fcis-29830 156 10 s.s s.s PROPN fcis-29830 156 11 . . PROPN fcis-29830 156 12 lienkamp lienkamp PROPN fcis-29830 156 13 , , PUNCT fcis-29830 156 14 t. t. PROPN fcis-29830 156 15 brox brox PROPN fcis-29830 156 16 , , PUNCT fcis-29830 156 17 o. o. NOUN fcis-29830 156 18 ronneberger ronneberger NOUN fcis-29830 156 19 , , PUNCT fcis-29830 156 20 3d 3d PROPN fcis-29830 156 21 u u NOUN fcis-29830 156 22 - - NOUN fcis-29830 156 23 net net ADJ fcis-29830 156 24 : : PUNCT fcis-29830 156 25 learning learn VERB fcis-29830 156 26 dense dense ADJ fcis-29830 156 27 volumetric volumetric NOUN fcis-29830 156 28 segmentation segmentation NOUN fcis-29830 156 29 from from ADP fcis-29830 156 30 sparse sparse ADJ fcis-29830 156 31 annotation annotation NOUN fcis-29830 156 32 , , PUNCT fcis-29830 156 33 international international ADJ fcis-29830 156 34 conference conference NOUN fcis-29830 156 35 on on ADP fcis-29830 156 36 medical medical ADJ fcis-29830 156 37 image image NOUN fcis-29830 156 38 computing computing NOUN fcis-29830 156 39 and and CCONJ fcis-29830 156 40 computerassisted computerassiste VERB fcis-29830 156 41 intervention intervention NOUN fcis-29830 156 42 , , PUNCT fcis-29830 156 43 2016 2016 NUM fcis-29830 156 44 . . PUNCT fcis-29830 157 1 [ [ X fcis-29830 157 2 14 14 NUM fcis-29830 157 3 ] ] PUNCT fcis-29830 157 4 k. k. NOUN fcis-29830 158 1 he he PROPN fcis-29830 158 2 , , PUNCT fcis-29830 158 3 x. x. PROPN fcis-29830 158 4 zhang zhang PROPN fcis-29830 158 5 , , PUNCT fcis-29830 158 6 s. s. PROPN fcis-29830 158 7 ren ren PROPN fcis-29830 158 8 , , PUNCT fcis-29830 158 9 j. j. PROPN fcis-29830 158 10 sun sun PROPN fcis-29830 158 11 , , PUNCT fcis-29830 158 12 deep deep ADJ fcis-29830 158 13 residual residual ADJ fcis-29830 158 14 learning learning NOUN fcis-29830 158 15 for for ADP fcis-29830 158 16 image image NOUN fcis-29830 158 17 recognition recognition NOUN fcis-29830 158 18 , , PUNCT fcis-29830 158 19 2016 2016 NUM fcis-29830 158 20 ieee ieee NOUN fcis-29830 158 21 conference conference NOUN fcis-29830 158 22 on on ADP fcis-29830 158 23 computer computer NOUN fcis-29830 158 24 vision vision NOUN fcis-29830 158 25 and and CCONJ fcis-29830 158 26 pattern pattern NOUN fcis-29830 158 27 recognition recognition NOUN fcis-29830 158 28 ( ( PUNCT fcis-29830 158 29 cvpr cvpr NOUN fcis-29830 158 30 ) ) PUNCT fcis-29830 158 31 , , PUNCT fcis-29830 158 32 2016 2016 NUM fcis-29830 158 33 , , PUNCT fcis-29830 158 34 pp pp ADJ fcis-29830 158 35 . . PUNCT fcis-29830 159 1 770 770 NUM fcis-29830 159 2 - - SYM fcis-29830 159 3 778 778 NUM fcis-29830 159 4 . . PUNCT fcis-29830 160 1 [ [ X fcis-29830 160 2 15 15 NUM fcis-29830 160 3 ] ] PUNCT fcis-29830 160 4 z. z. PROPN fcis-29830 160 5 zhou zhou PROPN fcis-29830 160 6 , , PUNCT fcis-29830 160 7 m.m.r m.m.r ADJ fcis-29830 160 8 . . PUNCT fcis-29830 160 9 siddiquee siddiquee PROPN fcis-29830 160 10 , , PUNCT fcis-29830 160 11 n. n. PROPN fcis-29830 160 12 tajbakhsh tajbakhsh PROPN fcis-29830 160 13 , , PUNCT fcis-29830 160 14 j.j.d.l.i.m.i.a j.j.d.l.i.m.i.a PROPN fcis-29830 160 15 . . PUNCT fcis-29830 161 1 liang liang PROPN fcis-29830 161 2 , , PUNCT fcis-29830 161 3 d. d. PROPN fcis-29830 161 4 multimodal multimodal NOUN fcis-29830 161 5 learning learn VERB fcis-29830 161 6 for for ADP fcis-29830 161 7 clinical clinical ADJ fcis-29830 161 8 decision decision NOUN fcis-29830 161 9 support support NOUN fcis-29830 161 10 : : PUNCT fcis-29830 161 11 4th 4th ADJ fcis-29830 161 12 international international ADJ fcis-29830 161 13 workshop workshop NOUN fcis-29830 161 14 , , PUNCT fcis-29830 161 15 m.-c m.-c NOUN fcis-29830 161 16 . . PUNCT fcis-29830 162 1 8th 8th ADJ fcis-29830 162 2 international international ADJ fcis-29830 162 3 workshop workshop NOUN fcis-29830 162 4 , , PUNCT fcis-29830 162 5 held hold VERB fcis-29830 162 6 in in ADP fcis-29830 162 7 conjunction conjunction NOUN fcis-29830 162 8 with with ADP fcis-29830 162 9 miccai miccai NOUN fcis-29830 162 10 , , PUNCT fcis-29830 162 11 granada granada PROPN fcis-29830 162 12 , , PUNCT fcis-29830 162 13 spain spain PROPN fcis-29830 162 14 , , PUNCT fcis-29830 162 15 s s X fcis-29830 162 16 ... ... PUNCT fcis-29830 162 17 unet++ unet++ ADJ fcis-29830 162 18 : : PUNCT fcis-29830 162 19 a a DET fcis-29830 162 20 nested nested ADJ fcis-29830 162 21 u u ADJ fcis-29830 162 22 - - ADJ fcis-29830 162 23 net net ADJ fcis-29830 162 24 architecture architecture NOUN fcis-29830 162 25 for for ADP fcis-29830 162 26 medical medical ADJ fcis-29830 162 27 image image NOUN fcis-29830 162 28 segmentation segmentation NOUN fcis-29830 162 29 , , PUNCT fcis-29830 162 30 11045 11045 NUM fcis-29830 162 31 ( ( PUNCT fcis-29830 162 32 2018 2018 NUM fcis-29830 162 33 ) ) PUNCT fcis-29830 162 34 3 3 NUM fcis-29830 162 35 - - SYM fcis-29830 162 36 11 11 NUM fcis-29830 162 37 . . PUNCT fcis-29830 163 1 [ [ X fcis-29830 163 2 16 16 NUM fcis-29830 163 3 ] ] X fcis-29830 163 4 a.g a.g PROPN fcis-29830 163 5 . . PROPN fcis-29830 163 6 howard howard PROPN fcis-29830 163 7 , , PUNCT fcis-29830 163 8 m. m. PROPN fcis-29830 163 9 zhu zhu PROPN fcis-29830 163 10 , , PUNCT fcis-29830 163 11 b. b. PROPN fcis-29830 163 12 chen chen PROPN fcis-29830 163 13 , , PUNCT fcis-29830 163 14 d. d. PROPN fcis-29830 163 15 kalenichenko kalenichenko PROPN fcis-29830 163 16 , , PUNCT fcis-29830 163 17 w. w. PROPN fcis-29830 163 18 wang wang PROPN fcis-29830 163 19 , , PUNCT fcis-29830 163 20 t. t. PROPN fcis-29830 163 21 weyand weyand PROPN fcis-29830 163 22 , , PUNCT fcis-29830 163 23 m. m. NOUN fcis-29830 163 24 andreetto andreetto PROPN fcis-29830 163 25 , , PUNCT fcis-29830 163 26 h.j.a h.j.a PROPN fcis-29830 163 27 . . PUNCT fcis-29830 163 28 adam adam PROPN fcis-29830 163 29 , , PUNCT fcis-29830 163 30 mobilenets mobilenet NOUN fcis-29830 163 31 : : PUNCT fcis-29830 163 32 efficient efficient ADJ fcis-29830 163 33 convolutional convolutional ADJ fcis-29830 163 34 neural neural ADJ fcis-29830 163 35 networks network NOUN fcis-29830 163 36 for for ADP fcis-29830 163 37 mobile mobile ADJ fcis-29830 163 38 vision vision NOUN fcis-29830 163 39 applications application NOUN fcis-29830 163 40 , , PUNCT fcis-29830 163 41 abs/1704.04861 abs/1704.04861 NUM fcis-29830 163 42 ( ( PUNCT fcis-29830 163 43 2017 2017 NUM fcis-29830 163 44 ) ) PUNCT fcis-29830 163 45 . . PUNCT fcis-29830 164 1 [ [ X fcis-29830 164 2 17 17 NUM fcis-29830 164 3 ] ] X fcis-29830 164 4 l.-c l.-c NOUN fcis-29830 164 5 . . PUNCT fcis-29830 165 1 chen chen PROPN fcis-29830 165 2 , , PUNCT fcis-29830 165 3 g. g. PROPN fcis-29830 165 4 papandreou papandreou PROPN fcis-29830 165 5 , , PUNCT fcis-29830 165 6 i. i. PROPN fcis-29830 165 7 kokkinos kokkinos PROPN fcis-29830 165 8 , , PUNCT fcis-29830 165 9 k.p k.p PROPN fcis-29830 165 10 . . PROPN fcis-29830 165 11 murphy murphy PROPN fcis-29830 165 12 , , PUNCT fcis-29830 165 13 a.l.j.i.t.o.p.a a.l.j.i.t.o.p.a PROPN fcis-29830 165 14 . . PUNCT fcis-29830 165 15 yuille yuille PROPN fcis-29830 165 16 , , PUNCT fcis-29830 165 17 m. m. NOUN fcis-29830 165 18 intelligence intelligence NOUN fcis-29830 165 19 , , PUNCT fcis-29830 165 20 deeplab deeplab PROPN fcis-29830 165 21 : : PUNCT fcis-29830 165 22 semantic semantic ADJ fcis-29830 165 23 image image NOUN fcis-29830 165 24 segmentation segmentation NOUN fcis-29830 165 25 with with ADP fcis-29830 165 26 deep deep ADJ fcis-29830 165 27 convolutional convolutional ADJ fcis-29830 165 28 nets net NOUN fcis-29830 165 29 , , PUNCT fcis-29830 165 30 atrous atrous ADJ fcis-29830 165 31 convolution convolution NOUN fcis-29830 165 32 , , PUNCT fcis-29830 165 33 and and CCONJ fcis-29830 165 34 fully fully ADV fcis-29830 165 35 connected connect VERB fcis-29830 165 36 crfs crf NOUN fcis-29830 165 37 , , PUNCT fcis-29830 165 38 40 40 NUM fcis-29830 165 39 ( ( PUNCT fcis-29830 165 40 2016 2016 NUM fcis-29830 165 41 ) ) PUNCT fcis-29830 165 42 834 834 NUM fcis-29830 165 43 - - SYM fcis-29830 165 44 848 848 NUM fcis-29830 165 45 . . PUNCT fcis-29830 166 1 [ [ X fcis-29830 166 2 18 18 NUM fcis-29830 166 3 ] ] PUNCT fcis-29830 166 4 k. k. PROPN fcis-29830 166 5 simonyan simonyan PROPN fcis-29830 166 6 , , PUNCT fcis-29830 166 7 a.j.c a.j.c NOUN fcis-29830 166 8 . . PUNCT fcis-29830 166 9 zisserman zisserman PROPN fcis-29830 166 10 , , PUNCT fcis-29830 166 11 very very ADV fcis-29830 166 12 deep deep ADJ fcis-29830 166 13 convolutional convolutional ADJ fcis-29830 166 14 networks network NOUN fcis-29830 166 15 for for ADP fcis-29830 166 16 large large ADJ fcis-29830 166 17 - - PUNCT fcis-29830 166 18 scale scale NOUN fcis-29830 166 19 image image NOUN fcis-29830 166 20 recognition recognition NOUN fcis-29830 166 21 , , PUNCT fcis-29830 166 22 abs/1409.1556 abs/1409.1556 PROPN fcis-29830 166 23 ( ( PUNCT fcis-29830 166 24 2014 2014 NUM fcis-29830 166 25 ) ) PUNCT fcis-29830 166 26 . . PUNCT fcis-29830 167 1 [ [ X fcis-29830 167 2 19 19 NUM fcis-29830 167 3 ] ] X fcis-29830 167 4 s.g s.g PROPN fcis-29830 167 5 . . PROPN fcis-29830 167 6 kolahi kolahi PROPN fcis-29830 167 7 , , PUNCT fcis-29830 167 8 s.k s.k PROPN fcis-29830 167 9 . . PROPN fcis-29830 167 10 chaharsooghi chaharsooghi PROPN fcis-29830 167 11 , , PUNCT fcis-29830 167 12 t. t. NOUN fcis-29830 167 13 khatibi khatibi NOUN fcis-29830 167 14 , , PUNCT fcis-29830 167 15 a. a. NOUN fcis-29830 167 16 bozorgpour bozorgpour PROPN fcis-29830 167 17 , , PUNCT fcis-29830 167 18 r. r. PROPN fcis-29830 167 19 azad azad PROPN fcis-29830 167 20 , , PUNCT fcis-29830 167 21 m. m. NOUN fcis-29830 167 22 heidari heidari PROPN fcis-29830 167 23 , , PUNCT fcis-29830 167 24 i. i. PROPN fcis-29830 167 25 hacihaliloglu hacihaliloglu PROPN fcis-29830 167 26 , , PUNCT fcis-29830 167 27 d. d. PROPN fcis-29830 167 28 merhof merhof PROPN fcis-29830 167 29 , , PUNCT fcis-29830 167 30 msa$^2$net msa$^2$net PROPN fcis-29830 167 31 : : PUNCT fcis-29830 167 32 multi multi ADJ fcis-29830 167 33 - - ADJ fcis-29830 167 34 scale scale ADJ fcis-29830 167 35 adaptive adaptive ADJ fcis-29830 167 36 attention attention NOUN fcis-29830 167 37 - - PUNCT fcis-29830 167 38 guided guide VERB fcis-29830 167 39 network network NOUN fcis-29830 167 40 for for ADP fcis-29830 167 41 medical medical ADJ fcis-29830 167 42 image image NOUN fcis-29830 167 43 segmentation segmentation NOUN fcis-29830 167 44 , , PUNCT fcis-29830 167 45 2024 2024 NUM fcis-29830 167 46 . . PUNCT