Instruction FACTA UNIVERSITATIS Series: Electronics and Energetics Vol. 29, N o 2, June 2016, pp. 233 - 241 DOI: 10.2298/FUEE1602233J ALGORITHM FOR UPTAKE ASSESSMENT IN SMALL LESIONS BASED ON DYNAMIC SCINTIGRAPHY SCANS * Milica M. Janković 1 , Vera Miler Jerković 1 , Ana Koljević Marković 2 , Dejan B. Popović 1 1 University of Belgrade – Faculty of Electrical Engineering, Belgrade, Serbia 2 National Cancer Research Center of Serbia, Belgrade, Serbia Abstract. The aim of our research was to develop an algorithm for estimation and visualisation of radiopharmaceutical uptake based on time-activity-curve (TAC) analysis in small regions of interest (ROI) in scintigraphic studies. The algorithm is implemented in Labview environment (National Instruments, Texas, Austin) and comprises the following steps: 1) delineation of grid of small ROIs over the examined tissue and corresponding TAC processing; 2) background vs tissue separation; 3) the extraction of all “suspected“ ROIs where TACs are not exponentially descendent; 4) correlation analysis between a TAC corresponding to the central suspected ROI and TACs of neghboring ROIs; 5) the extraction of representative TAC for “suspected“ area by Principal Component Analysis technique; and 6) visual interpretation of radiopharmaceutical distribution in the “suspected“ area. The application of algorithm is presented in data recorded in case of histopathologically proven parathyroid tumors. Key words: uptake, time activity curve, Principal Component Analysis, parathyroid tumor 1. INTRODUCTION Scintigraphy is a nuclear medicine diagnostic test for the visualization of spatial distribution of radioactivity uptake in a tissue. Radioactivity is taken by injection, inhalation or swallowing of medical agents (radiopharmaceuticals) with incorporated radioisotopes and the spatial distribution of radioactivity uptake is monitored by planar scintillation camera, SPECT (Single Photon Emission Computer Tomography) or PET (Positron Emission Tomography) camera. Dynamic scintigraphy is a diagnostic test for examining the function of organs and physiological systems. The result of this type of scintigraphy is a series of frames (dynamic scintigrams) recorded in short time intervals (10 seconds to 1 minute apart, depending on the type of organ and disease). Time activity curve (TAC) is a quantitative indicator of * An earlier version of this manuscript received the Best Oral Paper Award of the Biomedical Section at the 58 th ETRAN Conference, Vrnjačka Banja, 2-5 June, 2014 [1]. Received February 23, 2015; received in revised form June 14, 2015 Corresponding author: Milica M. Janković University of Belgrade – Faculty of Electrical Engineering, Bulevar kralja Aleksandra 73, Belgrade, Serbia (e-mail: piperski@etf.rs) 234 M.M. JANKOVIĆ, V. MILER JERKOVIĆ, A. KOLJEVIĆ MARKOVIĆ, D.B. POPOVIĆ radioactivity uptake changes in a specific region of interest (ROI) over time. Distinguishing typical TAC patterns is of great importance for diagnostic purposes. Beside the diagnostic application, scintigraphy has a very important place as a technique of preoperative imaging whose main goal is the precise localization of lesions in order to perform minimally invasive surgery [2-5]. In our previous work, we presented a Submarine method, based on TAC monitoring in small ROIs and finding abnormal TAC patterns corresponding to lesions [6]. Submarine method has proven useful for preoperative dynamic scintigraphic imaging of small lesions, especially in case of parathyroid imaging [6-8]. In this paper we introduce an algorithm that allows the precise uptake assessment in small lesions based on dynamic scintigrams and visual interpretation of uptake distribution in lesion area based on visualization of correlation matrix [1,8]. This algorithm is implemented as an additional tool in Submarine software. 2. METHODS AND MATERIALS Typical TAC pattern of health tissue consists of three phases: increasing vascular phase (the radioactivity in the target ROI is rapidly growing), accumulation uptake phase (radioactivity is accumulated in the target ROI) and washout phase (phase of exponential radioactivity decrease in the target ROI), [9]. In the case of lesions, the atypical TAC pattern (prolonged retention of radiopharmaceutical in the target tissue or even a peak of radioactivity in washout phase) could be observed, Fig. 1. Fig. 1 Difference in TAC patterns for healthy tissue and lesion In case of small lesions (<1 cm 3 ), it is very difficult or impossible to visually detect abnormal radioactivity uptake in individual frames, while it is clearly visible in the washout phase of TAC, Fig. 2. Central ROI of lesion is delineated by black color in Fig. 2A, and another three ROIs shifted relative to the central ROI are also delineated. TACs corresponding to highlighted ROIs are presented in Fig. 2B (TAC1, TAC2, TAC3, TAC4). A high degree of correlation between curves TAC1 and TAC2 (r=0.93), TAC1 and TAC3 (r=0.97) could be observed, versus substantially less correlated TAC1 i TAC4 (r=0.67). TACs corresponding to regions positioned over the lesion are not exponential in washout phase and are strongly correlated. This fact is used for defining the algorithm for uptake assessment and its visualization in small lesions. Uptake Visualisation in Small Lesions by Dynamic Scintigraphy 235 Fig. 2 A) A single frame from a dynamic image sequence, taken at the 23 th minute, with delineation of ROIs in the region of lesion (44 pixels, 66 mm) B) TACs corresponding to ROIs delineated in A) 2.1. Software The algorithm is implemented in the software for reading and processing dynamic studies introduced by authors in previous work [10]. Software is developed in Labview 8.6 environment (National Instruments, Texas, Austin) and additional NI Labview Biomedical Toolkit. Realized application enables:  Selection and readout of a dynamic scintigraphic study consisted of DICOM [11] images (each frame is archived as a separate .dcm file);  Rectangular cropping of frames to the region that is to be processed further – selection cropping position is performed on selected frame with visual inspection of cropping position in all frames;  Localization and visualization of small lesions by algorithm for uptake assessment presented in Section 1.2. 236 M.M. JANKOVIĆ, V. MILER JERKOVIĆ, A. KOLJEVIĆ MARKOVIĆ, D.B. POPOVIĆ Checking Principal Component Analysis conditions (see Section 1.3) for examples presented in Section 2 was performed in RStudio, version 0.98.976. 2.2. Algorithm description The algorithm for small lesion localization and visualization consists of six steps shown in Fig. 3. Fig. 3 Flowchart of the algorithm for the uptake assessment in small lesions. ROI – region of interest, TAC – time activity curve Step 1 Cropped area, containing the tissue that will be examined, is automatically divided into N small square ROIs, equal in size n x n, where n is a number of pixels (n=4 is a default value, but user can change it). This number of ROIs (N) will be reduced in Step 2 into the number of ROIs (T) which belongs to the tissue (TN). The number of tissue ROIs (T) will be reduced in Step 3 into the number of ROIs (M) whose TACs are not exponentially descendent (MT) and thereby indicate the abnormal radioactivity uptake and the potential lesion. TACs corresponding to all N ROI cells are calculated and smoothed by cubic spline technique using Labview function Cubic Spline Fit.vi [12]. User can adjust the value (range [0,1]) of balance parameter (input parameter of cubic spline function) taking into consideration the requirement that the coefficient of determination (R-square) is greater than 80% (user sets minimum balance parameter for which R-square>80%). Labview function Goodness of Fit.vi is used for estimation of R-square value based on the raw TAC and the cubic spline filtered TAC. Step 2 Maximum values of radioactivity TAC CS i max are calculated for all TAC CS i (i=1, N). Reference value P for discriminating tissue from background is calculated according to the following equation: NiTACP i ,1),max( max CS  (1) Uptake Visualisation in Small Lesions by Dynamic Scintigraphy 237 Further analysis continues only for those T ROIs (TN) that belong to the tissue, which means that satisfy the condition TAC CS i max > m  P, 00.8), because the prerequisite for PCA is good correlation, or too high in order to avoid multicollinearity (cxy<0.9) [13]. The value of determinant indicates on multicollinearity or singularity among original variables and it should not be less than 0.00001. In the case when the value of determinant is less than 0.00001, it means that some variables are highly correlated. The Kaiser-Meyer-Olkin (KMO) is a measure of sampling adequacy [15]. It compares correlation and partial correlations between variables. KMO takes values between 0 and 1. The value of KMO should be greater than 0.5 if the sample is adequate. The Bartlett's test of sphericity is a test used to examine the null hypothesis: “Variables are uncorrelated, correlation matrix is an identity matrix”. Therefore, we need to get p-value < 0.05 in this test and conclude that null hypothesis can be rejected. For choosing the number of principal components we used the Kaiser rule and Screeplot combined with the amount of total variance that the chosen principal components have (the amount of total variance above 80 % is usually suggested) [16]. The rotation of principal components is used for improving interpretation of results. We have chosen the orthogonal rotation – varimax [16]. Uptake Visualisation in Small Lesions by Dynamic Scintigraphy 239 3. RESULTS AND DISCUSSION We demonstrated the results of suggested algorithm for radioactivity uptake assessment in two patients who underwent parathyroid scintigraphy in the National Cancer Research Center of Serbia, Belgrade. Scintigraphic recording was performed in patients suspected of having primary hyperparathyroidism (PHPT) based on previous biochemical analysis (increased level of parathyroid hormone) and positive ultrasound findings. Patient data (biochemical, ultrasound, biopsy) are shown in Table 1. Table 1 Patients: biochemical, ultrasound and biopsy data. PHPT – primary hyperparathyroidism, PTH - parathyroid hormone Data Patient 1 Patient 2 Gender female male Age [years] 69 19 PTH [pg/ml] 223 125 PHPT ultrasound positive positive Previous thyroidectomy no yes, right Histopathology parathyroid adenoma parathyroid cancer Tumor position right inferior left superior Tumor volume [mm 3 ] 60 55 Siemens e.cam camera and Siemens Syngo e.soft 2007 software (Siemens AG, Erlangen, Germany) have been used for image acquisition. After intravenous 99m Tc MIBI administration (with the radioactivity of 500 MBq, 13.5 mCi), 35 minutes of dynamic parathyroid scintigraphy (1 frame/min, dimension of image matrix: 128x128, pixel size 1.5 mm, zoom 3.2, anterior view) were performed. Results of pre-analysis PCA data are presented in Table 2. All PCA criteria from Section 1.3 are satisfied (determinant value>0.00001, KMO>0.5, p-value<0.05 for Bartlett's test). First principal component carries more than 80% of total variance, which means that it is representative of TAC changes. Fig. 5 shows results of algorithm applied in Patient 1 for two dimensions of ROI cells (33 pixels and 44 pixels). Visual inspection of standard dynamic scintigram at the moment of radioactivity peak in washout phase cannot distinguish lesion from healthy tissue, unlike suggested parametric imaging (Fig. 5A). Better discrimination between lesion and healthy tissue is evident in case of smaller ROI dimension, closer to the real lesion localization (compare Fig. 5A left and right). Representative TAC patterns are presented in Fig. 5B. The position of small parathyroid adenoma (right inferior) was surgically confirmed. Table 2 Results of pre-analysis PCA data Parameters Patient 1 Patient 2 33 pixels (4.54.5 mm) 44 pixels (66 mm) 44 pixels (66 mm) Determinant value 0.00006 0.00009 0.00008 Kaiser-Meyer-Olkin value 0.855 0.806 0.805 Bartlett's test (p-value) 0.000 0.000 0.000 Number of principal components 1 1 1 Amount of total variance [%] 89.97 91.19 89.02 240 M.M. JANKOVIĆ, V. MILER JERKOVIĆ, A. KOLJEVIĆ MARKOVIĆ, D.B. POPOVIĆ Fig. 5 A) A single frame from a dynamic image sequence, taken at the 22 nd minute and visual interpretation of lesion localization by introduced algorithm B) Representative TAC patterns obtained by Principal Component Analysis Fig. 6A shows results of visualization algorithm applied in Patient 2. Representative TAC pattern is presented in Fig. 6B. The position of small parathyroid cancer (left superior) was surgically confirmed. Fig. 6 A) A single frame from a dynamic image sequence, taken at the 24 th minute and a visual interpretation of lesion localization by suggested algorithm B) Representative TAC pattern obtained by Principal Component Analysis Uptake Visualisation in Small Lesions by Dynamic Scintigraphy 241 4. CONCLUSION In this paper, we introduced an algorithm that enables the display of orientation, shape and boundaries of lesions. The algorithm visualizes the propagation of TAC correlation in the lesion area. The application of such algorithms is desirable in preoperative diagnostics in order to plan surgery. Further investigation will be related to the development of fully automated algorithm for lesion localization from dynamic scintigrams and its evaluation in a larger population with different oncological diseases. Acknowledgement: The paper is financially supported by the Ministry of Education, Science and Technological Development of the Republic of Serbia (no. 175016) and the company National Instruments (Slovenia, Ljubljana). REFERENCES [1] M. M. Janković, V. Miler Jerković, A. Koljević Marković and D. B. Popović, "Algorithm for the uptake assessment in small lesions in dynamic scintigraphy ", In Proceedings of the 58 th ETRAN Conference, 2- 5 June, Vrnjačka Banja, 2014, pp. ME 1.1 1-4 [In Serbian]. [2] D. Fuster, S. Vidal-Sicart, T. José-Vicente, P. Paredes, D. Rubello and F. Pons, "What is the role of preoperative scintigraphic imaging and the intraoperative gamma probe in secondary hyperparathyroidism?" Nucl Med Commun, vol. 35, no. 5, pp. 443-445, 2014. [3] I. Stoffels, M. Müller, M.H. Geisel, J. Leyh, T. Pöppel, D. Schadendorf and J. Klode, "Cost-effectiveness of preoperative SPECT/CT combined with lymphoscintigraphy vs. lymphoscintigraphy for sentinel lymph node excision in patients with cutaneous malignant melanoma", Eur J Nucl Med Mol, [Epub ahead of print] 2014. [4] M. Ibusuki, Y. Yamamoto, T. Kawasoe, S. Shiraishi, S. Tomiguchi, Y. Yamashita, Y. Honda, K. Iyama and H. Iwase, "Potential advantage of preoperative three-dimensional mapping of sentinel nodes in breast cancer by a hybrid single photon emission CT (SPECT)/CT system", Surg Oncol, vol. 19, no. 2, pp. 88- 94, 2010. [5] M. Giuliano, S.A. Gulec, D. Rubello, G. Boni, M. Puccini, M.R. Pelizzo, G. Manca, D. Casara, G. Sotti, P. Erba, D. Volterrani and A.E. Giuliano, "Preoperative localization and radioguided parathyroid surgery", J Nucl Med, vol. 44, no. 9, pp. 1443-1458, 2003. [6] A. Koljević Marković, M. M. Janković, I. Marković, G. Pupić, R. Džodić and A. B. Delaloye, "Parathyroid dual tracer subtraction scintigraphy: small regions method for quantitative assessment of parathyroid adenoma uptake", Ann Nucl Med, vol. 28, pp. 736-745, 2014. [7] M. Đurović M, M. M. Janković and A. Koljević Marković, " Semi-automatic localization of parathyroid tumors in dynamic sestamibi scintigrams ", In Proceedings of the 22 nd Telecommunications forum TELFOR 2014, 25-27 November, Belgrade, 2014, pp. 955-958 [In Serbian]. [8] M. M. Janković, " Computer system for acquiring, storing, retrieving and processing images obtained by gamma camera ", PhD thesis, University of Belgrade – Faculty of Electrical Engineering, 2014 [In Serbian]. [9] M. P. Sandler, R. E. Coleman, J. A. Patton, F. J. Th. Wackers and A. Gottschalk, Diagnostic Nuclear Medicine, 4th ed. Philadelphia: Lippincott Williams & Wilkins, 2003. [10] M. M. Janković, A. Koljević Marković and D. B. Popović, " Labview application for analysis of time activity curves in regions of small lesions in nuclear medicine ", In Proceedings of the 57 th ETRAN Conference, 3-6 June, Zlatibor, 2013, pp. ME 1.9 1-5 [In Serbian]. [11] http://dicom.nema.org/ [12] J. S. Fleming and R. W. Kenny, "A comparison of techniques for the filtering of noise in the renogram, " Phys Med Biol, vol. 22, no. 2, pp. 359-364, Mar. 1977. [13] J. E. Jackson and J. Wiley, A user's guide to principal components. New York: John Wiley and Sons, Inc., 1991. [14] T. M. Lehmann, C. Gönner and K. Spitzer, "Interpolation Methods in Medical Image Processing, " IEEE Trans Med Imag, vol. 18, no. 11, pp. 1049-1075, Nov. 1999. [15] H. F. Kaiser, "An index of factorial simplicity," Psychometrika, vol. 39, pp. 31-36, 1974. [16] I. T. Jolliffe, Principal Component Analysis. 2nd ed. New York, USA: Springer, 2002. http://dicom.nema.org/