Format Template


Volume 1, No. 2 | July – December 2017 

SJCMS | P-ISSN: 2520-0755 | E-ISSN: 2522-3003 © 2017 Sukkur IBA University – All Rights Reserved 
48 

Initiative for Thyroid Cancer Diagnosis: Decision Support 

System for Anaplastic Thyroid Cancer
Jamil Ahmed Chandio1, M. Abdul Rehman Soomrani1, Attaullah Sehito1, 

Shafaq Siddiqui1 

Abstract 
Due to the high level exposure of biomedical image analysis, Medical image mining 

has become one of the well-established research area(s) of machine learning. AI (Artificial 

Intelligence) techniques have been vastly used to solve the complex classification problems of 

thyroid cancer. Since the persistence of copycat chromatin properties and unavailability of nuclei 

measurement techniques, it is really problem for doctors to determine the initial phases of nuclei 

enlargement and to assess the early changes of chromatin distribution. For example involvement 

of multiple transparent overlapping of nuclei may become the cause of confusion to infer the 

growth pattern of nuclei variations. Un-decidable nuclei eccentric properties may become one 

of the leading causes for misdiagnosis in Anaplast cancers. In-order to mitigate all above stated 

problems this paper proposes a novel methodology so called “Decision Support System for 

Anaplast Thyroid Cancer” and it proposes a medical data preparation algorithm AD 

(Analpast_Cancers) which helps to select the appropriate features of Anaplast cancers such as 

(1) enlargement of nuclei, (2) persistence of irregularity in nuclei and existence of hyper 

chromatin. Proposed methodology comprises over four major layers, the first layer deals with 

the noise reduction, detection of nuclei edges and object clusters. The Second layer selects the 

features of object of interest such as nuclei enlargement, irregularity and hyper chromatin. The 

Third layer constructs the decision model to extract the hidden patterns of disease associated 

variables and the final layer evaluates the performance evaluation by using confusion matrix, 

precision and recall measures. The overall classification accuracy is measured about 97.2% with 

10-k fold cross validation. 

Keywords: biomedical image;algorithm;classification;dicision support system 

1. Introduction 
Recently biomedical image inference of 

DICOM (Digital communication in medicine) 

images have been witnessed one of the active 

research area(s) of machine learning and AI 

(Artificial Intelligence) base techniques have 

shown significant impact upon the diagnostic 

process. Various CAD (computer added 

diagnosis) systems have been proposed to 

solve the classification problems of malignant 

diseases such as lung, breast, head & neck, 

lymphatic system, thyroid and other cancers. 

Some of the very nice approaches 

[1],[2],[3],[4] were proposed to solve the 

classification problem of cancer disease. 

                                                           
1 Department of Computer Science Sukkur IBA University Sukkur, Pakistan 

Corresponding author: Jameelahmed.phdcs@iba-suk.edu.pk  

 Infact, it is really one of the challenging 

field to identify the object of interest for 

different organs of human body as stated above 

because the classification of poorly 

differentiated, well differentiated and 

undifferentiated cancers have been found with 

lots of variations and divergent properties. For 

example, thyroid Anaplast cancer is one of the 

aggressive malignancy and its growth rate is 

higher than the other type of cancers.  

Since the DICOM images of thyroid 

disease provides sufficient information to 

diagnose the Anaplast cancers but due to the 

use of poor staining material such as H and E, 

it may deceive doctors while examining the 

mailto:Jameelahmed.phdcs@iba-suk.edu.pk


Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

49 

mimic chromatin properties either it is hyper 

chromatin or not. Secondly, due to the 

unavailability of nuclei measurement 

techniques, at very early stage, irregular set of 

nuclei may produce point of confusion for 

doctors to determine the initial stage of 

enlargement from concerned group of nuclei. 

Involvement of multiple transparent 

overlapping nuclei may also be prone to one of 

the causes confusion to infer the growth pattern 

of nuclei grooves and early eccentric 

properties of nuclei are considered one of the 

difficult decisions because detachment of cell 

wall may be initial phase. In-order to solve all 

the above stated problems this paper proposes 

a system “Decision Support System for 

Anaplast Thyroid Cancer” to assist the doctors 

to diagnose the Anaplast cancer at early stage. 

Methodology of proposed system comprises 

over four layers and it proposes an algorithm 

for data preparation for Anaplast cancer 

[Algorithm 1]. In first layer noise reduction 

techniques have been used to reduce the noise 

of DICOM image by using adaptive threshold 

method and edges of objects have been 

detected by using canny edge detection 

algorithm whereas regional clusters have been 

tagged by using watershed algorithm. In 

second layer our proposed data preparation 

algorithm detects the properties and behaviors 

of Anaplast cancer such as enlargement of 

nuclei, irregularity of nuclei and hyper 

chromatin related features. In the third layer, 

random forest AI base technique has been used 

to construct the decision model to decide about 

the existence of cancerous material. In final 

layer performance evaluation have been 

conducted by observing confusion matrix, 

precision and recall measures. The overall 

performance has been shown by using AUC 

(area under curve). The measured accuracy of 

proposed system is about 97.20%. A real world 

dataset has been used for Anaplast cancers 

received from SMBBMU (Shaheed Muhtarma 

Benazir Bhutto Medical University) Pakistan 

as such datasets are unavailable in literature. 

Rest of paper is organized in five sections. 

Introduction is presented in section one. 

Related works in section two. Proposed 

methodology is shown in section three. Results 

in section four and conclusion are described in 

section five. 

2. Related Works 
Basically this paper falls into the category 

of productive mining and deals with the 

classification problem of DICOM (Digital 

communication in medicine) images of biopsy. 

Specially; this paper proposes a preprocessing 

algorithm for Thyroid related Anaplast cancers 

which are most aggressive type of malignancy 

comprises over dissimilar growth patterns in 

terms of shape, size and other morphological 

properties, which are significantly important to 

diagnose at proper stage. Following related 

works have been seen in the recent past.       

SVM (support vector machine) and 

AdaBoost machine learning techniques were 

compared for breast cancer classification 

problem [1]. The best classification accuracy 

was approximated as 87.42%. This paper 

presents a novel algorithm in data preparation 

phase and uses random forest AI base 

technique to construct the classification model 

for cancerous and non-cancerous thyroid 

Anaplast malignancies.  

A system [2] using Convolutional neural 

network based machine learning technique 

was proposed for thyroid disease 

classification. The DICOM images need 

significant pre-processing techniques for every 

individual class of disease such as well-

differentiated, poorly differentiated and others. 

All the cancer types may not be pre-processed 

with the same procedure because the key 

building blocks nuclei are considered most 

important micro architectural components, 

which usually have been found with different 

properties for every histo-pathological class. In 

data preparation phase proposed algorithm 

selects enlargement, irregularity and hyper 

chromatin related features. A comparative [3] 

study was conducted by using various 

segmentation algorithms based on clustering 

i.e. K-means and watershed than a supervised 

learning approach was used to construct the  

decision model obtained accuracies for 

template matching strategy were measured as 



Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

50 

72% and 87%.  Since the histopathology 

focuses on dissimilar micro structures of 

different human cells and tissues, whereas in a 

single organ various type of cancer diseases 

may be appeared because every disease has its 

own properties which needs significant efforts 

in data preparation phases.  

On the concept of Neural Network [4] 

different AI base algorithms i.e. Scaled 

Conjugate Gradient, BFGS Quasi-Newton, 

Gradient Descent method and Bayesian 

regularization algorithms were used and best 

approximated accuracies for thyroid disease 

datasets were recorded as respectively 90.5%, 

86.30% and 83.50%. Since the DICOM 

images of thyroid disease provide sufficient 

information to diagnose the Anaplast cancers, 

it deceives doctors when hyper chromatin may 

be observed with nearby properties due to the 

use of poor staining material and at very early 

stage irregular nuclei could not be seen 

properly due to the unavailability of nuclei 

measurement techniques, whereas 

enlargement of cells could easily be picked by 

the doctors but due to the involvement of 

multiple transparent overlapping nuclei in 

deceiving nature may create confused 

situations. It is direly needed to propose an 

efficient system with the support of systematic 

data preparation algorithm which recommends 

deepest decisions about the evidences of 

cancerous material at early stages. 

3. Methodlogy 
This paper proposes a system which deals 

with the predictive mining in the field of machine 

learning.  The system comprises over four layers, 

in the first layer noise reductions have been done 

and in the second layer several disease related 

properties have been selected under the umbrella 

of behavior detection and feature selection. In the 

third layer classification models have been 

constructed and in the final layer performance 

evaluation have been done. Since the DICOM 

images of Anaplast cancers need significant 

methodology to pre-process, the nuclei behaviors 

of Anaplast cancers have dissimilar behaviors 

that are very difficult to select object of interest 

in proper way. For example enlargement of 

nuclei, irregularity in the nuclear patterns and 

variations in chromatin distribution are very 

difficult to interpret as digital set of objects. This 

paper also contributes a data preparation 

algorithm AD (Analpast_Cancers) which 

performs all the above stated tasks effectively 

and every observation is recorded carefully with 

the assistance of expert medical panel. 

 
Algorithm1: AD (Anaplast Diagnosis) 

 



Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

51 

3.1. Layer 1: Noise Reduction        
Proposed algorithm has to perform three 

numbers of tasks. Firstly to detect the 

enlargement of cells. Secondly to find out the 

irregularity and third the hyper chromatin. Let 

us consider a medical image of biopsy 

consisting of the several number of nuclei and 

each nuclei occupies size, shape, color and 

many features. Computationally, the 

algorithm reduces the noise by applying the 

image pixel derivation as per eq. (1). 
 

‖∆ 𝑓‖   ← √(
𝜕𝑓

𝑑𝑥
)  2 + (

𝜕𝑓

𝑑𝑦
)  2             (1) 

 

Medical image 𝑀 = {𝑥+ℎ1, 𝑥 +
ℎ2, 𝑥 + ℎ3 … … … … 𝑥 + ℎ𝑛 } is consisting 
upon the h spaces in vertical and 𝑁 =
{𝑦 + ℎ1, 𝑦 + ℎ2, 𝑦 + ℎ3 … … … … 𝑦 + ℎ𝑛 }  
horizontal vectors. If complexity of huge 

vectors is selected as features of large size 

medical images, it may require high 

computational power to compute the 

correlation of the associated pixels. Proposed 

algorithm reduces the unnecessary 

information of medical image and detects the 

nuclei by eq. (1), where homogenous intensity 

based pixels regions are formed as foreground 

of second derivation. Since g(x, y)  is position 
of particular pixel p which represents the each 

nucleus considering homogenous threshold 

eq. (2). 

 

𝑔(𝑥, 𝑦) = (𝑥𝑖 , 𝑦𝑖 ) ≤ 0 (𝑃) ≤ 1 (2) 
 

Proposed algorithm visits each pixel and 

selects appropriate set of weighted pixels p 

designated as regions and tagged as 

Enlargement of nuclei where shape of various 

nuclei behaviors are considered with same 

visual properties.  the task of this subsection 

of algorithm is to extract the statistical 

morphological features. Additionally, canny 

edge detection algorithm is used to record the 

shape related changes between the set of 

nuclei in medical images. 

3.2. Canny edge detection: 
Nuclei edges provide additional 

information about the selected regions of 

enlarged objects where shape sizes could be 

recorded for further analysis and eq.(3) 

describes that all the edges of medical image 

objects are the lies between the (G) =

√G(xi, yi)
2 , since the G is targeted 

connected set of lines to be formed around the 

(xi, yi) spatial locations. Gx and Gy Angles 

could also be used to measure the directions of 

objects which are involved in expected 

expansion over multiple regions eq. (4). This 

Figure 1: Decision Support System for Anaplast Thyroid Cancer workflow 

 



Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

52 

helps to quantify the max and min distances 

for overlap objects and by  (
Gx

Gy
) dividing the 

spatial positions logical separation could be 

done. 

 

𝐸𝑑𝑔𝑒𝑁𝑢𝑐𝑙𝑒𝑖 (𝐺) = √𝐺(𝑥𝑖 , 𝑦𝑖 )
2            (3) 

 

𝐴𝑛𝑔𝑙𝑒𝑁𝑢𝑐𝑙𝑒𝑖   (𝜃 ) =  𝑡𝑎𝑛
−1(

𝐺𝑥

𝐺𝑦
)              (4) 

3.3. Layer 2: Behavior detection and 
feature selection: 

 Enlargement: 
𝑬𝒏𝒍𝒂𝒓𝒈𝒎𝒆𝒏𝒕 = ∑ 𝑺𝒊𝒛𝒆𝒐𝒇𝒑 

𝒊=𝟏
𝒋=𝒏   

= 𝟏          (5) 

        Let’s consider (xi, yi) are the spatial 
locations residing in the vector space of  f(G ) 
where shape are not equal to the same size eq. 

(5). It is said to enlarge shapes that are counted 

with the assistance of auto image separation 

techniques as per eq (6). The image particles 

are considered as seed and each seed is 

cropped with its mean value and surrounded 

area is mapped with edges that the necessary 

information of particular object can be 

obtained.  

 

𝐬𝒆𝒆𝒅(𝝁𝒙, 𝒚) = {(
𝝅𝒙
𝝅𝒚

)𝑺𝒖𝒎(𝒂(𝒙, 𝒚)}
 

 

 

          

(6) 

 Irregularity: 
Mostly the enlargement detection 

deceives in terms of object detection, since the 

proposed preprocessing technique measure 

the center of enlarged sequences of nuclei and 

rings acquired through canny edge detection 

are transformed around the large objects(xi +

 hi, yi +  hj) where (hi, hj) are transformed as 

boundaries of multiple objects.  

 

  𝐗𝐜𝐢𝐫𝐜𝐥𝐞 𝐜 =
𝟏

𝐌
∑ 𝐱𝐢

𝐧
𝐢=𝟏 𝐦𝐢(

𝛑𝐱+𝐡

𝛑𝐲+𝐡
) 

 

 

                    (7) 

The next task of our algorithm is to 

encircle the closed objects Xcircle c where 
every point is denoted by M and numbers of 

associated pixels have been calculated with 

aggregated values eq. (7). Since the locations 

(xi +  hi(xi +  hi, yi +  hj)) are encircled. 

The irregular margins M have been counted 

and recorded in each observation of training 

and testing datasets. 

 Hyper chromatin: 
The hyper chromatin is also known as 

access of chromatin could be found in 

cells. The cell wall of nuclei has irregular 

quantities of the chromatin fluid. It is a 

condition where nuclei could not able to 

maintain its shape related properties due 

to the detachment of surrounding nuclei 

walls and loss of nucleus properties at 

DNA levels. Computationally proposed 

subsection of algorithm records a high 

concentrated mean value of chromatin 

color movement based features and color 

deviation is quantified on the basis of 

mass value and recorded into the 

observation. 

 

*

1

2

*

ˆ

n

yy

atinHyperchrom

n

i

ii





(8) 

Hyper chromatin is one of the key 

feature in which each nuclei has to lose the 

chromatin. Since the loss of chromatin may 

incurred due to the eccentric nucleolus of 

nuclei or the expansion in the size of nuclei 

may be affected to nuclei shape, size and 

behaviors. Aggregated Color movements 

were measured with the assistance of distance 

matrix, where color spectrum is shown in 

[Figure 2] to represent the behaviors of 

chromatin in each nuclei with sounding set of 

nuclei. Let us suppose every DICOM image D 

is represented as collection of pixels contain 

the information based upon X = {X =
 x1, x2 … … … . xn} andY = {y1, y2 … … … . yn} 
where each set of pixels P(X, Y) =
{m1, m2 … … … . mn} qualifies an object with 
distinct set of featuresF = {f1, f2 … … … . fn} 
onH = {h1, h2 … … … . hn}. 

3.4. Layer 3: Classification of nuclei: 
Random forests machine learning 

algorithm is widely used to predict the 

different classification problems due to 



Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

53 

several advantages to regress the rank 

relationships of important variables. Deep 

decision trees have ability to aggregate more 

than one decision related to every class label 

attribute. 

Let’s consider DICOM dataset D 

consists having several no of important 

variablesD = {(Xi , Yi)} i=1
n . the decision 

model can be constructed as aggregated 

decision variable by fitting by measuring the 

Jth feature in training stage, since the decision 
tree Jth features are distressed datasets because  
of aggregated permutation operations (out of  

bag error ) conducted for overall trees. 

  

 ŷ = ∑ W(xi, x́)yi
n
i=1                               (9) 

 

       Let’s suppose ŷ is a class of functions in 
x,set of points which formalized by weighted 
function known as W. thus following 

weighted decision tree can be constructed as 

per eq. (9) and (10). 

 

ŷ =
1

m
∑ ∑ Wj(xi, x́)yi

n
i=1

m
j=1 =

∑ (
1

m
∑ Wj(xi, x́)

m
j=1 )

n
i=1 yi                         (10) 

3.5. Layer 4: performance evaluation: 
A total number of 20 biopsy images 

were used for training and testing purposes 

and 1829 number of nuclei were detected from 

those images. The confusion metrics [Table 1] 

shown for cancerous class which classified 

948 observations and non-cancerous classified 

a number of 830 instances. . The precision eq. 

(11) and recall eq. (12) was approximated for 

cancerous classes 98.63% and about 97.73% 

was measured for non-cancerous classes. 

 
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =

            
𝑁𝑢𝑚𝑏𝑒𝑟𝑜𝑓𝑇𝑟𝑒𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠

𝑁𝑢𝑚𝑏𝑒𝑟𝑜𝑓𝑇𝑟𝑢𝑒𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠+𝐹𝑎𝑙𝑠𝑒𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
     (11)   

 

𝑅𝑒𝑐𝑎𝑙𝑙 = 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 =  
𝑇𝑟𝑢𝑒𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒

𝐹𝑎𝑙𝑠𝑒𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒
      

(12)      
 

𝑆𝑒𝑝𝑒𝑐𝑖𝑓𝑦 =  
𝑇𝑟𝑢𝑒𝑁𝑒𝑔𝑒𝑡𝑖𝑣𝑒

𝑇𝑟𝑢𝑒𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒+𝐹𝑎𝑙𝑠𝑒𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒
        (13) 

 

4. Results 
In [Figure 2] there are five columns. The 

column one is presented as an image input 

where all the images are belonging to the 

Anaplast thyroid cancer with different histo-

pathological material H and E. In second 

column the points of individual cells have 

been shown, where each point denotes number 

of enlarge nuclei. In column number three the 

irregular nuclei are placed where each set of 

overlapping of nuclei are represented. The 

system generated circles show that there is 

several numbers of nuclei which are eccentric 

due to the loss of cell wall. In column number 

four hyper chromatin, level is shown by 

distance matrix in a color spectrum which 

reveals that the all those nuclei are considered 

as hyper chromatin where red color entropies 

are found with highest aggregated values. The 

ratio of chromatin is said to be hyper when the 

surrounded features are present with low red 

color ratio using color spectrum properties. 

Column five is placed to show the 

summarization of each observation. All these 

detected features have been recorded in a 

dataset for training and testing purposes. 

 

 
Figure 2: Results of preprocessing algorithm 

Anaplast Diagnosis 



Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

54 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table-I: Confusion matrix 

 Cancerous  Non-Cancerous 

Cancerous  948 31 

 Non-Cancerous 20 830 

 

Table-II: Overall performance of proposed methodology 

 
Raw 

images 

No of 

Extracted 

nuclei 

No of 

classified 

Nuclei 

No of miss-

classified 

Nuclei 

Precessio

n  
Recall  

Cancerous 10 979 948 31 98.63% 98.83% 

Non-Cancerous  10 850 830 20 97.73% 97.64% 

 

Table-III: Comparison of our system with literature 

Approaches Image Type Cancer Type Technique Accuracy 

1 

Ultrasound Image Follicular SVM  97.50 % 

AdaBoost 87.42%. 

2 FNAC Images Follicular NN 91.00 % 

3 
FNAC Images Medullary Templated matching 

strategy 

87.00% 

4 

 

 

FNAB Images 

 

 

Papillary 

Scaled Conjugate Gradient 90.5%,  

BFGS Quasi-Newton, 85.40% 

Gradient Descent method 86.30% 

Bayesian regularization 83.50% 

Our 

Proposed 

approach 

FNAB Images Anaplast  Random forest 97.20% 

Figure 3: Estimated ROC Curve for the proposed 

system 



Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

55 

5. Conclusion  
This paper contributes a novel data 

preparation algorithm AD (Anaplast 

Diagnosis) for Anaplast cancers, since the 

feature selection is one of the difficult tasks in 

medical images diagnosis. It needs significant 

efforts from the perspective of image feature 

engineering to select the related features from 

DICOM (Digital Communication in 

Medicine) images of thyroid biopsy such as 

enlargement of nuclei, irregular nuclei as 

individual & also in sequences and different 

chromatin distribution level features.   

The methodology of proposed system 

comprises over four interconnected layers, 

where layer one has been assigned the job to 

reduce the noise and the layer two has been 

assigned core tasks to select appropriate 

features by measuring the  aggregate values of 

involved pixels as color movements of 

selected objects. The layer three has been used 

to construct the classification model by using 

the random forest algorithm and the final layer 

has to perform the performance evaluation of 

the system.  

The numbers of 1829 nuclei were detected 

from 20 images of cancerous and non-

cancerous classes. The confusion matrix show 

that a total number of 948 instances out of 979 

observations were classified as cancerous 

class and out of 850 instances 830 

observations were classified for non-

cancerous class.  

The overall accuracy of the system was 

recorded as 97.2% with 10-k fold cross 

validation. Since such datasets are unavailable 

in literature, used dataset was received form 

SMBBMU, Pakistan. By observing the 

experiment in this research, we conclude that 

special data preparation algorithms are 

required to be developed for each histo-

pathological medical image classification 

problem distinctly, because every DICOM 

image can be identified with its own 

properties which are mostly different in nature 

from each other. 

 

ACKNOWLEDGEMENT 
This research work is carried out due to 

motivational attitude of my beloved father. He 

remained seriously ill for ten years and left me 

alone with wet eyes, but even in tight 

situations he encouraged me to carry on my 

studies. I am also highly thankful to my 

teachers and medical partners, who guided 

and motivated me to keep my step forward to 

complete my studies 

REFERENCES 
[1] Lashkari, AmirEhsan, and Mohammad 

Firouzmand. "Early Breast Cancer 

Detection in Thermogram Images using 

Supervised and Unsupervised 

Algorithms." The Journal of 

Engineering Vol 7, No 3, pp. 114-124, 

(2016)  

[2] Xu, Y., Mo, T., Feng, Q., Zhong, P., Lai, 

M. and Chang, E.I. , May. "Deep learning 

of feature representation with multiple 

instance learning for medical image 

analysis". In Acoustics, Speech and 

Signal Processing (ICASSP), IEEE 

International Conference pp. 1626-1630. 

(2014) 

[3] Cheng Chen, Wei Wang, John A. Ozolek, 

and Gustavo K. Rohde., "A flexible and 

robust approach for segmenting cell 

nuclei from 2D microscopy images using 

supervised learning and template 

matching". Journal of Cytometry A. 2013 

May ; 83(5): p.p 495–507, (2013) 

[4] Pourahmad, S., Azad, M., Paydar, S., & 

Reza, H. "Prediction of malignancy in 

suspected thyroid tumour patients by 

three different methods of classification 

in data mining". In First International 

Conference on advanced information 

technologies and applications pp. 1-8, 

(2012). 

[5] Glotsos, D,  Tsantis, S. Kybic J, 

Daskalakis, A. "Pattern recognition based 

segmentation versus wavelet maxima 

chain edge representation for nuclei 

detectionin microscopy images of thyroid 

nodules". Euromedica medical center, 



Jameel. Ahmed Chandio (et al.) Initiative for Thyroid Cancer Diagnosis: Decision Support System for Anaplastic 

Thyroid Cancer                          (pp. 48 - 56) 

Sukkur IBA Journal of Computing and Mathematical Sciences – SJCMS | Volume 1 No. 2 July – December 2017 © Sukkur IBA University 

56 

Department of Medical Imaging, Athens, 

Greece (2013). 

[6] Gopinath, B, Shanthi, N..  "Support 

Vector Machine Based Diagnostic 

System for Thyroid Cancer using 

Statistical Texture Features", Asian 

Pacific J Cancer Prev, 14 (1), pp. 97-102. 

(2013). 

[7] Gopinath, B, Shanthi, N. "Computer-

aided diagnosis system for classifying 

benign and malignant thyroid nodules in 

multi-stained FNAB cytological images”, 

Australas Phys Eng Sci Med Vol (36), pp. 

219–230, (2013). 

[8] Stavros T., “Improving diagnostic 

accuracy in the classification of thyroid 

cancer by combining quantitative 

information extracted from both 

ultrasound and cytological images". 1st 

IC-SCCE-Athens, pp. 8-10, (2004). 

[9] http://www.thyroidmanager.org/chapter/f

ine-needle-aspiration-biopsy-of-the-

thyroid-gland/ accessed on 15-01-2014. 

[10] Thyroid Cancer: Web Site, 

http://www.medicinenet.com/thyroid_ca

ncer/article.htm accessed on 15-01-2014 

[11] Han J., Kamber M.,   and    Pei P. “Data 

Mining: Concepts and Techniques”,. 3rd 

(ed.) The Morgan Kaufmann Series in 

Data Management Systems (2011).  

[12] Rafeal, C. G., Richard, E. W.,  (2007). 

"Digital Image Processing" , 3rd (ed).  

Prentice Hall. (2007). 

[13] Hussein, Mohamed, Amitabh Varshney, 

and Larry Davis. "On implementing 

graph cuts on cuda." First Workshop on 

General Purpose Processing on Graphics 

Processing Units. Vol. (1). (2007). 

[14] Boykov, Y., & Funka-Lea, G. "Graph 

cuts and efficient ND image 

segmentation". International journal of 

computer vision, 70(2), pp. 109-131, 

(2006). 

[15] Malik, J., Belongie, S., Leung, T., & Shi, 

J. “Contour and texture analysis for image 

segmentation". In Perceptual 

Organization for artificial vision 

systems Vol(1) pp. 139-172. (2000). 

[16] Al-Kofahi, Y., Lassoued, W., Lee, W., & 

Roysam, B. "Improved automatic 

detection and segmentation of cell nuclei 

in histopathology images".Biomedical 

Engineering, IEEE Transactions 

on, 57(4), pp. 841-852. (2010). 

[17] Liu, X., Huo, Z., & Zhang, J. "Automated 

segmentation of breast lesions in 

ultrasound images", In Engineering in 

Medicine and Biology Society, 2005. 

IEEE-EMBS 2005. 27th Annual 

International Conference of the IEEE pp. 

7433-7435, (2006).  

[18] I.S. Jacobs and C.P. Bean, “Fine particles, 

thin films and exchange anisotropy,” in 

Magnetism, vol. III, G.T. Rado and H. 

Suhl, Eds. New York: Academic, 1963, 

pp. 271-350. (1963).

 

http://www.thyroidmanager.org/chapter/fine-needle-aspiration-biopsy-of-the-thyroid-gland/
http://www.thyroidmanager.org/chapter/fine-needle-aspiration-biopsy-of-the-thyroid-gland/
http://www.thyroidmanager.org/chapter/fine-needle-aspiration-biopsy-of-the-thyroid-gland/
http://www.medicinenet.com/thyroid_cancer/article.htm%20accessed%20on%2015-01-2014
http://www.medicinenet.com/thyroid_cancer/article.htm%20accessed%20on%2015-01-2014
http://store.elsevier.com/authorDetails.jsp?authorId=ELS_1014284
http://store.elsevier.com/authorDetails.jsp?authorId=ELS_1014605
http://store.elsevier.com/authorDetails.jsp?authorId=ELS_1109696