![International Journal of Electrical and Computer Engineering (IJECE)
Vol. 8, No. 4, August 2018, pp. 2442~2450
ISSN: 2088-8708, DOI: 10.11591/ijece.v8i4.pp2442-2450 2442
Journal homepage: http://iaescore.com/journals/index.php/IJECE
The Utilization of Physics Parameter to Classify Histopathology
Types of Invasive Ductal Carcinoma (IDC) and Invasive
Lobular Carcinoma (ILC) by using K-Nearest Neighbourhood
(KNN) Method
Anak Agung Ngurah Gunawan1
, I Wayan Supardi2
, S. Poniman3
, Bagus G. Dharmawan4
1,2,3
Physics Department, Udayana University, Bali, Indonesia
4
Radiology Department, Prima Medika Hospital, Bali, Indonesia
Article Info ABSTRACT
Article history:
Received Jan 31, 2018
Revised Apr 2, 2018
Accepted Apr 20, 2018
Medical imaging process has evolved since 1996 until now. The forming of
Computer Aided Diagnostic (CAD) is very helpful to the radiologists to
diagnose breast cancer. KNN method is a method to do classification toward
the object based on the learning data which the range is nearest to the object.
We analysed two types of cancers IDC dan ILC. 10 parameters were
observed in 1-10 pixels distance in 145 IDC dan 7 ILC. We found that the
Mean of Hm(yd,d) at 1-5 pixeis the only significant parameters that
distingguish IDC and ILC. This parameter at 1-5 pixels should be applied in
KNN method. This finding need to be tested in diffrerent areas before it will
be applied in cancer diagnostic.
Keyword:
Invasive ductal carcinoma
Invasive lobular carcinoma
Nearest neighbourhood
Copyright © 2018 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Anak Agung Ngurah Gunawan
Departement of Physics,
Udayana University,
Bali, Indonesia.
Email: ngurah_gunawan@unud.ac.id
1. INTRODUCTION
The diagnosing of early breast cancer is very important to reduce the mortality rate for women. The
breast cancer is the health problem in the world, many women die because of it. Most of the patients who
come to have treatment have had advanced stage. Therefore, the early detection of breast cancer and its
treatment is the only way to survive longer and to improve the patients‟ life quality. CAD system that is
evolved is very helpful in diagnosing breast cancer. Besides, CAD system can also be used as the comparison
of diagnostic result of the radiologists and the pathology and anatomy specialist doctors. In this kind of CAD
system, the accuracy of result is very important. A misguided detection can cause a misguided treatment too.
Because the problem is sensitive, there are many researchers do the research in breast cancer specialty and
compete each other to achieve the better result.
The technologies developed all this time to detect early breast cancer are Ultrasonography (USG)
device [1], [2], Mammography, Magnetic Resonance Imaging (MRI) [4], [5], and Positron Emission
Tomography (PET) Scan [6]. All early detection tools above are unable to classify IDC and ILC
histopathology types. Therefore, we propose a new technique to classify IDC and ILC histopathology types
by using physics parameter as the input variable by using KNN method. Software that we produced would be
planted in Mammography tool so that it could improve its activity as an early detection of breast cancer.
The research that we proposed was focused on mammogram images from Dokter Soetomo
(Surabaya, Indonesia) Hospital, Sanglah (Denpasar, Bali, Indonesia) Hospital, and from Prima Medika](https://image.slidesharecdn.com/v5214mar186jan1828aug178504-13503-3-ededit-201116071451/85/The-Utilization-of-Physics-Parameter-to-Classify-Histopathology-Types-of-Invasive-Ductal-Carcinoma-IDC-and-Invasive-Lobular-Carcinoma-ILC-by-using-K-Nearest-Neighbourhood-KNN-Method-1-320.jpg)
![Int J Elec & Comp Eng ISSN: 2088-8708
The Utilization of Physics Parameter to Classify Histopathology… (Anak Agung Ngurah Gunawan)
2443
(Denpasar, Bali, Indonesia) Hospital producing classification of IDC and ILC histopathology types. IDC was
currently categorized into invasive carcinoma of no special type was terrace breast carcinoma, namely it was
from 45% to 75% case, whereas ILC was only from 5 to 15% invasive breast carcinoma [7]. The disparity
between these two types was clinicopathology characteristic and response regarding systemic therapy [8].
IDC Histopathology gave the growth image of invasive epithelium malignant cells which mostly form solid
and sinsisial patterns, and part with glandular and tubules differentiation. ILC consists of epithelium
malignant cells arranged in the spreading of individual cell or arranged in infiltrative linear pattern between
fibrus connective tissue stroma and it was usually connected with lobular carcinoma in situ (LCIS) [7].
In our paper, we proposed a new method to classify histopathology types of ILC and IDC breast
cancers. Because of the physics parameter value range of ILC and IDC was different [9]. So, physics
parameter containing on the mammogram could be used as the input variable for KNN method to determine
whether it belonged to ILC or IDC types.
The superiority of the method that we proposed was the output of our method was numerical form
which its value was certainly to be very objective, it was different with the previous method that still used the
visual reading which the result was very subjective and depending on the readers. Why did we do the
research? We wanted to help decreasing the mortality rate of women who have had breast cancer.
The aim of the writing is to introduce a new method to detect the types of IDC and ILC
histopathology. CAD system that we have developed is used as the comparison of FNAB result before doing
the operation.
2. METHODS
2.1. Sample
The research was approved by ethics committee of Sanglah Hospital, Denpasar, Bali, Indonesia.
Number: 1204/UN.14.2/KEP/2017. The samples were taken at random from the year 2013 to the year 2017
from the database of Dokter Soetomo Hospital (Surabaya, Indonesia), Prima Medika Hospital (Denpasar,
Bali, Indonesia), Sanglah General and Central Hospital (Denpasar, Bali, Indonesia). The samples consisted
of 7 images of ILC type and 145 images of IDC type.
2.2. Developed method
Gunawan [9] used physics parameter to determine histopathology types of breast cancer by using
Special Pattern Cropping method. In this research, we developed the use of physics parameter as the input of
KNN method to determine histopathology type of breast cancer.
It has been observed that the abnormality, especially the suspicious are with the higher density than
the neighbor pixel like on the Figure 1 and Figure 2. We counted 9 physics parameters like entropy, contrast,
angular second moment, inverse difference moment, mean of Hm(y,d), deviation, entropy of Hdiff, angular
second moment of Hdiff and mean of Hdiff for every pixel with the range between pixels from 1 pixel to 10
pixels. By using Anova statistics to get significant parameter that was able to distinguish IDC and ILC
histopathology types. Then, we used chosen physics parameter as input variable from KNN method to take
the final decision. We applied the pre-processing steps at the early algorithm to fix the image quality. The
aim of repairing application of image quality was to clarify the image on mammogram. The block diagram
explaining the method used was showed on Figure 3.
(a) (b)
Figure 1. (a) IDC, (b) Subtract the background image of the original image
Image: From the data base of Radiology Department of RSUP. Dr. Soetomo Surabaya Hospital.](https://image.slidesharecdn.com/v5214mar186jan1828aug178504-13503-3-ededit-201116071451/85/The-Utilization-of-Physics-Parameter-to-Classify-Histopathology-Types-of-Invasive-Ductal-Carcinoma-IDC-and-Invasive-Lobular-Carcinoma-ILC-by-using-K-Nearest-Neighbourhood-KNN-Method-2-320.jpg)
![Int J Elec & Comp Eng ISSN: 2088-8708
The Utilization of Physics Parameter to Classify Histopathology… (Anak Agung Ngurah Gunawan)
2445
∑ ∑ [ ( )] ( ) (1)
∑ ∑ (2)
∑ ∑ ( ) (3)
∑ ∑ [
( )
] (4)
for yr ≠ yq
∑ ∑ ( ) ( )
( )
(5)
with
( ) ∑ (6)
∑ (7)
∑ (8)
√∑ ∑ ( ) (9)
∑ ∑| |
(10)
∑ (11)
∑ (12)
∑ (13)
With H(yq,yr,d), d, y each is the probability of a pair of gray-level, the distance between the pixel and gray
level value, respectively [10], [11].
2.5. The selection of physics variable
Selection of significant physics variables as input KNN method that is able to distinguish type IDC
and ILC using T test statistical analysis. The main reason why we used KNN method to determine invasive
lobular carcinoma and invasive ductal carcinoma in breast cancer, because the categorization based on the
nearest range between examined data with learning data used the Euclidean Distance formula like the
Equation (14).
2.6. KNN method
KNN method is a method to do classification toward the object based on the learning data which the
range is nearest to the object.
To calculate Euclidean Distance by using the Equation (14).
ni
i
UT iiD
1
2
(14)
It consists of 2 input variables, namely mean1 and mean2 and 2 histopathology types, namely ILC
and IDC. If there is a new data with mean1 value 180.81088 and mean2 = 181.11186, is like Table 1.](https://image.slidesharecdn.com/v5214mar186jan1828aug178504-13503-3-ededit-201116071451/85/The-Utilization-of-Physics-Parameter-to-Classify-Histopathology-Types-of-Invasive-Ductal-Carcinoma-IDC-and-Invasive-Lobular-Carcinoma-ILC-by-using-K-Nearest-Neighbourhood-KNN-Method-4-320.jpg)
![Int J Elec & Comp Eng ISSN: 2088-8708
The Utilization of Physics Parameter to Classify Histopathology… (Anak Agung Ngurah Gunawan)
2449
examination result from the radiologists and the experts of Pathology Anatomy (PA) was the same with the
system decision of CAD that we developed. If the decision was assumed to be different, so it was assumed
FP.
It was like shown at table 5, our algorithm, the sensitivity was 70%, the specificity was 71%, the
accuracy was 70%, the precision was 98%, and error rate was 30%. Two radiologists and experts of
pathology and anatomy found our result was quite satisfying and could be reliable.
4. CONCLUSION
The CAD system that we developed by using physics parameter was very helpful in classifying
histopathology type of breast cancer. Mammogram image was very difficult to detect histopathology types.
Even, the Radiology experts were not able to identify it for 100%.
The utilization of physics parameter to classify histopathology types of breast cancer helped the
experts of Radiology to get the second opinion. The final aim of the CAD system that we developed for
mammography was to detect untouchable lesion which its size was often neglected on mammography.
Detection of histopathology type increased the opportunity for women to be successful in the treatment of
breast cancer.
Our research was especially focused on histopathology types of IDC and ILC. We used physics
parameter to classify histopathology types of breast cancer, after getting significant variable to distinguish
histopathology types of IDC and ILC. Then, we used the parameter as the input variable of KNN method to
take decision whether it included IDC or ILC histopathology types.
According to the experts of radiology, the result produced by the CAD system that we developed
was quite satisfying and could be reliable, and could assist the expert of radiology in diagnosing breast
cancer.
ACKNOWLEDGMENTS
The authors would like to thank for the assistance given during the research to: Directorate of
Research and Community Service (DRPM) as the research funder. Prof.Dr.dr.Ketut Suastika,SpPD-KEMD
(Rector of Udayana University, Bali), Prof.Dr.Ir. I Nyoman Gde Antara,M.Eng (Chairman of LPPM
Udayana University, Bali), Drs. I.B. Suaskara,M.Si (Dean of FMIPA Faculty, Udayana University, Bali),
Prof.Dr.dr.Sri Maliawan,SP.BS(K) (Chairman of Research Ethics Commission FK UNUD/RSUP Sanglah,
Denpasar), Prof. Dr.dr.Triyono KSP,Sp.Rad( K) (Chairman of Radiology Department of RSUP Dokter
Soetomo Hospital, Surabaya), dr. A.A.A.N.Susraini,SpPA(K) (Chairman of Pathology Anatomy Department
of RSUP Sanglah Hospital, Denpasar Bali), dr. Bagus G. Dharmawan, Sp.Rad (Chairman of Radiology
Department of RSU. Prima Medika Hospital, Denpasar, Bali), Wirabrata (Director of Human Resource and
Education of RSUP Sanglah Hospital Denpasar, Bali), dr. Putu Dian Ekawati, MPH (President Director of
RSU. Prima Medika Hospital, Denpasar, Bali).
REFERENCES
[1] Satoko Nakano, Masahiko Ohtsuka, Akemi Mibu, at al., “Diagnostic imaging strategy for MDCT- or MRI-detected
breast lesions: use of targeted sonography”, BMC Medical Imaging, 201212:13.
[2] Jimmy Okello, Harriet Kisembo, Sam Bugeza, at al., “Breast cancer detection using sonography in women with
mammographically dense breasts”, BMC Medical Imaging, 201414:41.
[3] Sylvia H. Heywang-Köbrunner,Astrid Hacker, Stefan Sedlacek, “Advantages and Disadvantages of Mammography
Screening”, Breast Care (Basel), 2011 Jun, vol. 6, no. 3, pp. 199-207.
[4] Fernanda Philadelpho AP, Gabriela M, Maria Julia GC, at al., “Magnetic resonance imaging-radioguided occult
lesion localization (ROLL) in breast cancer using Tc-99m macro-aggregated albumin and distilled water control”,
BMC Medical Imaging, 201313:33.
[5] Arjan P Schouten van der Velden, Carla Boetes, Peter Bult, at al., “Magnetic resonance imaging in size assessment
of invasive breast carcinoma with an extensive intraductal component”, BMC Medical Imaging, 20099:5.
[6] Bateman, Timothy, “Advantages and disadvantages of PET and SPECT in a busy clinical practice”, Journal of
Nuclear Spinger Journal, Jan 19, 2012.
[7] Romualdo Barroso-Sousa and Otto Metsger-Filho, “Difference between invasive lobuler and invasive duktal
carcinoma of the breast: result and therapeutic implications”, Therapeutic Advances in Medical Oncology, vol. 8
no. 4, pp. 261-266.
[8] Sunil RL., Lan OE., Stuart JS., at al., “WHO Classification of Tumours of The Breast”, International Agency for
Research on Cancer, 4th
Edition, Lyon, 2012.
[9] A.A.N. Gunawan, Supardi, Dharmawan, “Readability Increase Of Mammography X-Ray Photos Results In
Determining The Breast Cancer Histopathology Types Using Special Pattern Cropping With Physical Parameter”,
Advances in Applied Physics, 2014, vol. 2, no. 1, pp. 43-52.](https://image.slidesharecdn.com/v5214mar186jan1828aug178504-13503-3-ededit-201116071451/85/The-Utilization-of-Physics-Parameter-to-Classify-Histopathology-Types-of-Invasive-Ductal-Carcinoma-IDC-and-Invasive-Lobular-Carcinoma-ILC-by-using-K-Nearest-Neighbourhood-KNN-Method-8-320.jpg)
![ ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 8, No. 4, August 2018 : 2442 – 2450
2450
[10] A.A.N. Gunawan, “A novel model determination of breast cancer stage using physical parameter”, Far East
Journal of Mathematical Sciences, 2014, vol. 87, no. 1, pp. 23-35.
[11] Dhawan, A.P., “Analysis of Mammofraphic Microcalcification Using Gray Level Image Structure Feature”,
IEEE,Trans.Medical Imaging, 1996, vol. 15, pp. 246-257.
BIOGRAPHIES OF AUTHORS
Dr. Anak Agung Ngurah Gunawan, M.T., was born in Denpasar on September 25, 1962. He
obtained his Masters degree in 1999 in computer Engineering, Institut Teknologi 10 Nopember
Surabaya Indonesia and his Dr degree from the faculty sains and teknologi, University of
Airlangga Surabaya Indonesia. His main interess are image procecing. He has published 24 peper
in journal international. He is a Senior Lecture Departement of physics University of Udayana at
Bali Indonesia.
I Wayan Supardi, S.Si, M.Si. Born on March 31, 1971, undergraduate degree (S1) in 1998,
Department of Physics (Physics of Instrumentation of Electronics and Output) Faculty of
Mathematics and Natural Sciences Udayana University, Strata 2 (S2) in 2004 Department of
Physics (Instrumentation Physics) Bandung Institute of Technology (ITB ). Becoming a Lecturer
at Physics Department in 1999. Field of instrumentation and energy penetration.
Ir. S. Poniman, M.Si, born on June 6, 1956, has been a staff of physics faculty of mathematics
and natural sciences of udayana university since 1986. Undergraduate degree (S1) at surabaya
technological institute in 1985, postgraduate program (S2) Bandung in 1996. Field of
superconductor research. The experience of the organization became chairman of the faculty of
mathematics faculty of university udayana since 2011 until now.](https://image.slidesharecdn.com/v5214mar186jan1828aug178504-13503-3-ededit-201116071451/85/The-Utilization-of-Physics-Parameter-to-Classify-Histopathology-Types-of-Invasive-Ductal-Carcinoma-IDC-and-Invasive-Lobular-Carcinoma-ILC-by-using-K-Nearest-Neighbourhood-KNN-Method-9-320.jpg)
The Utilization of Physics Parameter to Classify Histopathology Types of Invasive Ductal Carcinoma (IDC) and Invasive Lobular Carcinoma (ILC) by using K-Nearest Neighbourhood (KNN) Method
- 1. International Journal of Electrical and Computer Engineering (IJECE) Vol. 8, No. 4, August 2018, pp. 2442~2450 ISSN: 2088-8708, DOI: 10.11591/ijece.v8i4.pp2442-2450 2442 Journal homepage: http://iaescore.com/journals/index.php/IJECE The Utilization of Physics Parameter to Classify Histopathology Types of Invasive Ductal Carcinoma (IDC) and Invasive Lobular Carcinoma (ILC) by using K-Nearest Neighbourhood (KNN) Method Anak Agung Ngurah Gunawan1 , I Wayan Supardi2 , S. Poniman3 , Bagus G. Dharmawan4 1,2,3 Physics Department, Udayana University, Bali, Indonesia 4 Radiology Department, Prima Medika Hospital, Bali, Indonesia Article Info ABSTRACT Article history: Received Jan 31, 2018 Revised Apr 2, 2018 Accepted Apr 20, 2018 Medical imaging process has evolved since 1996 until now. The forming of Computer Aided Diagnostic (CAD) is very helpful to the radiologists to diagnose breast cancer. KNN method is a method to do classification toward the object based on the learning data which the range is nearest to the object. We analysed two types of cancers IDC dan ILC. 10 parameters were observed in 1-10 pixels distance in 145 IDC dan 7 ILC. We found that the Mean of Hm(yd,d) at 1-5 pixeis the only significant parameters that distingguish IDC and ILC. This parameter at 1-5 pixels should be applied in KNN method. This finding need to be tested in diffrerent areas before it will be applied in cancer diagnostic. Keyword: Invasive ductal carcinoma Invasive lobular carcinoma Nearest neighbourhood Copyright © 2018 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: Anak Agung Ngurah Gunawan Departement of Physics, Udayana University, Bali, Indonesia. Email: [email protected] 1. INTRODUCTION The diagnosing of early breast cancer is very important to reduce the mortality rate for women. The breast cancer is the health problem in the world, many women die because of it. Most of the patients who come to have treatment have had advanced stage. Therefore, the early detection of breast cancer and its treatment is the only way to survive longer and to improve the patients‟ life quality. CAD system that is evolved is very helpful in diagnosing breast cancer. Besides, CAD system can also be used as the comparison of diagnostic result of the radiologists and the pathology and anatomy specialist doctors. In this kind of CAD system, the accuracy of result is very important. A misguided detection can cause a misguided treatment too. Because the problem is sensitive, there are many researchers do the research in breast cancer specialty and compete each other to achieve the better result. The technologies developed all this time to detect early breast cancer are Ultrasonography (USG) device [1], [2], Mammography, Magnetic Resonance Imaging (MRI) [4], [5], and Positron Emission Tomography (PET) Scan [6]. All early detection tools above are unable to classify IDC and ILC histopathology types. Therefore, we propose a new technique to classify IDC and ILC histopathology types by using physics parameter as the input variable by using KNN method. Software that we produced would be planted in Mammography tool so that it could improve its activity as an early detection of breast cancer. The research that we proposed was focused on mammogram images from Dokter Soetomo (Surabaya, Indonesia) Hospital, Sanglah (Denpasar, Bali, Indonesia) Hospital, and from Prima Medika
- 2. Int J Elec & Comp Eng ISSN: 2088-8708 The Utilization of Physics Parameter to Classify Histopathology… (Anak Agung Ngurah Gunawan) 2443 (Denpasar, Bali, Indonesia) Hospital producing classification of IDC and ILC histopathology types. IDC was currently categorized into invasive carcinoma of no special type was terrace breast carcinoma, namely it was from 45% to 75% case, whereas ILC was only from 5 to 15% invasive breast carcinoma [7]. The disparity between these two types was clinicopathology characteristic and response regarding systemic therapy [8]. IDC Histopathology gave the growth image of invasive epithelium malignant cells which mostly form solid and sinsisial patterns, and part with glandular and tubules differentiation. ILC consists of epithelium malignant cells arranged in the spreading of individual cell or arranged in infiltrative linear pattern between fibrus connective tissue stroma and it was usually connected with lobular carcinoma in situ (LCIS) [7]. In our paper, we proposed a new method to classify histopathology types of ILC and IDC breast cancers. Because of the physics parameter value range of ILC and IDC was different [9]. So, physics parameter containing on the mammogram could be used as the input variable for KNN method to determine whether it belonged to ILC or IDC types. The superiority of the method that we proposed was the output of our method was numerical form which its value was certainly to be very objective, it was different with the previous method that still used the visual reading which the result was very subjective and depending on the readers. Why did we do the research? We wanted to help decreasing the mortality rate of women who have had breast cancer. The aim of the writing is to introduce a new method to detect the types of IDC and ILC histopathology. CAD system that we have developed is used as the comparison of FNAB result before doing the operation. 2. METHODS 2.1. Sample The research was approved by ethics committee of Sanglah Hospital, Denpasar, Bali, Indonesia. Number: 1204/UN.14.2/KEP/2017. The samples were taken at random from the year 2013 to the year 2017 from the database of Dokter Soetomo Hospital (Surabaya, Indonesia), Prima Medika Hospital (Denpasar, Bali, Indonesia), Sanglah General and Central Hospital (Denpasar, Bali, Indonesia). The samples consisted of 7 images of ILC type and 145 images of IDC type. 2.2. Developed method Gunawan [9] used physics parameter to determine histopathology types of breast cancer by using Special Pattern Cropping method. In this research, we developed the use of physics parameter as the input of KNN method to determine histopathology type of breast cancer. It has been observed that the abnormality, especially the suspicious are with the higher density than the neighbor pixel like on the Figure 1 and Figure 2. We counted 9 physics parameters like entropy, contrast, angular second moment, inverse difference moment, mean of Hm(y,d), deviation, entropy of Hdiff, angular second moment of Hdiff and mean of Hdiff for every pixel with the range between pixels from 1 pixel to 10 pixels. By using Anova statistics to get significant parameter that was able to distinguish IDC and ILC histopathology types. Then, we used chosen physics parameter as input variable from KNN method to take the final decision. We applied the pre-processing steps at the early algorithm to fix the image quality. The aim of repairing application of image quality was to clarify the image on mammogram. The block diagram explaining the method used was showed on Figure 3. (a) (b) Figure 1. (a) IDC, (b) Subtract the background image of the original image Image: From the data base of Radiology Department of RSUP. Dr. Soetomo Surabaya Hospital.
- 3. ISSN: 2088-8708 Int J Elec & Comp Eng, Vol. 8, No. 4, August 2018 : 2442 – 2450 2444 (a) (b) Figure 2. (a) ILC, (b) Subtract the background image of the original image Image: From the data base of Radiology Department of RSUP. Dr. Soetomo Surabaya Hospital. Figure 3. The steps of proposed algorithm 2.3. Processing The aim of Processing step is to group mammogram images of ILC and IDC taken from the anatomic Pathologic research as the standard goal. Cropping the suspicious area. Then, it is done the reparation of the image quality to clarify the mammogram image. 2.4. The calculation of physics parameter After the processing step, our algorithm counts 9 physics parameters, namely entropy, contrast, angular second moment, inverse differential moment, mean, deviation, entropy of Hdiff, angular second moment of Hdiff and mean of Hdiff at the range from 1 pixel to 10 pixels by using the Equations (1)-(13). The counted entropy from the histogram of order two provides the measurement of irregularity and defined like Equation (1). Histogram of order 2 illustrates the distribution of possibilities from the event of the pair of gray-level. Input Image Fix the Image Quality by Using Histogram Equlization Calculation of Physics Parameter Selecting Significant Physics Parameter by Using Logistic Multinomial Inserting Chosen Physics Parameter as Input Variable of KNN Method Output
- 4. Int J Elec & Comp Eng ISSN: 2088-8708 The Utilization of Physics Parameter to Classify Histopathology… (Anak Agung Ngurah Gunawan) 2445 ∑ ∑ [ ( )] ( ) (1) ∑ ∑ (2) ∑ ∑ ( ) (3) ∑ ∑ [ ( ) ] (4) for yr ≠ yq ∑ ∑ ( ) ( ) ( ) (5) with ( ) ∑ (6) ∑ (7) ∑ (8) √∑ ∑ ( ) (9) ∑ ∑| | (10) ∑ (11) ∑ (12) ∑ (13) With H(yq,yr,d), d, y each is the probability of a pair of gray-level, the distance between the pixel and gray level value, respectively [10], [11]. 2.5. The selection of physics variable Selection of significant physics variables as input KNN method that is able to distinguish type IDC and ILC using T test statistical analysis. The main reason why we used KNN method to determine invasive lobular carcinoma and invasive ductal carcinoma in breast cancer, because the categorization based on the nearest range between examined data with learning data used the Euclidean Distance formula like the Equation (14). 2.6. KNN method KNN method is a method to do classification toward the object based on the learning data which the range is nearest to the object. To calculate Euclidean Distance by using the Equation (14). ni i UT iiD 1 2 (14) It consists of 2 input variables, namely mean1 and mean2 and 2 histopathology types, namely ILC and IDC. If there is a new data with mean1 value 180.81088 and mean2 = 181.11186, is like Table 1.
- 5. ISSN: 2088-8708 Int J Elec & Comp Eng, Vol. 8, No. 4, August 2018 : 2442 – 2450 2446 Table 1. Mean1 and Mean2 and 2 Histopathology Types Mean1 Mean2 Type of Histopathology 114.26825 113.97313 IDC 91.31764 91.71834 IDC 150.37279 151.51907 IDC 142.07158 142.72827 IDC 155.47071 156.30185 IDC 159.96244 161.16492 ILC 149.01149 148.98700 ILC 153.83340 153.93231 ILC 149.59902 149.84301 ILC 161.06060 161.83271 ILC The Steps: 1. Determine k parameter= the nearest neighbor number. For instance k =3. 2. Calculate the range between a new data and a learning data, is like Table 2. Table 2. Range Between a New Data and a Learning Data Mean1 Mean2 The range quadrate with the new data (180.81088, 181.11186) 114.26825 113.97313 (114.26825- 180.81088)2 + (113.97313- 181.11186)2 = 8935.530673 91.31764 91.71834 (91.31764- 180.81088)2 + (91.71834- 181.11186)2 = 16000.24142 150.37279 151.51907 (150.37279- 180.81088)2 + (151.51907- 181.11186)2 = 1802.210543 142.07158 142.72827 (142.07158- 180.81088)2 + (142.72827- 181.11186)2 = 2974.033346 155.47071 156.30185 (155.47071- 180.81088)2 + (156.30185- 181.11186)2 = 1257.660812 159.96244 161.16492 (159.96244- 180.81088)2 + (161.16492- 181.11186)2 = 832.5378658 149.01149 148.98700 (149.01149- 180.81088)2 + (148.98700- 181.11186)2 = 2043.207834 153.83340 153.93231 (153.83340- 180.81088)2 + (153.93231- 181.11186)2 = 1466.512365 149.59902 149.84301 (149.59902- 180.81088)2 + (149.84301- 181.11186)2 = 1951.921185 161.06060 161.83271 (161.06060- 180.81088)2 + (161.83271- 181.11186)2 = 761.7591848 3. The range order and the determination of the nearest neighbor based on the k minimum, is like Table 3. Table 3. Determination of the Nearest Neighbor Based on the k Minimum Mean1 Mean2 The range quadrate with the new data (180.81088, 181.11186) Minimum range level Including the nearest neighbor 114.26825 113.97313 (114.26825- 180.81088)2 + (113.97313- 181.11186)2 = 8935.530673 9 IDC 91.31764 91.71834 (91.31764- 180.81088)2 + (91.71834- 181.11186)2 = 16000.24142 10 IDC 150.37279 151.51907 (150.37279- 180.81088)2 + (151.51907- 181.11186)2 = 1802.210543 5 IDC 142.07158 142.72827 (142.07158- 180.81088)2 + (142.72827- 181.11186)2 = 2974.033346 8 IDC 155.47071 156.30185 (155.47071- 180.81088)2 + (156.30185- 181.11186)2 = 1257.660812 3 IDC 159.96244 161.16492 (159.96244- 180.81088)2 + (161.16492- 181.11186)2 = 832.5378658 2 ILC 149.01149 148.98700 (149.01149- 180.81088)2 + (148.98700- 181.11186)2 = 2043.207834 7 ILC 153.83340 153.93231 (153.83340- 180.81088)2 + (153.93231- 181.11186)2 = 1466.512365 4 ILC 149.59902 149.84301 (149.59902- 180.81088)2 + (149.84301- 181.11186)2 = 1951.921185 6 ILC 161.06060 161.83271 (161.06060- 180.81088)2 + (161.83271- 181.11186)2 = 761.7591848 1 ILC From the three levels, ILC histopathology type comes out two times, whereas IDC comes out one, it means mean1 value = 180.81088 and mean2 = 181.11186 including ILC group.
- 6. Int J Elec & Comp Eng ISSN: 2088-8708 The Utilization of Physics Parameter to Classify Histopathology… (Anak Agung Ngurah Gunawan) 2447 3. RESULTS AND DISCUSSION Table 4 shows the average of physics parameter of idc and ilc at various distance (pixels) at Dr Sutomo Hospital Surabaya at 2017. Table 4. Average of Physics Parameter of IDC and ILC at various distance (pixels) at Dr Sutomo Hospital Surabaya at 2017 Pixel IDC (n=148) ILC (n=7) Signifi kansi Average Standart deviation Standart error Average Standart deviation Standart error Entropy 1 3.6104536 0.15085807 0.01252808 3.6171657 0.08066793 0.03048961 0.907 2 3.6568468 0.15332612 0.01273304 3.6617500 0.07960890 0.03008933 0.933 3 3.6777772 0.15452700 0.01283277 3.6813429 0.07883346 0.02979625 0.952 4 3.6885450 0.15457833 0.01283703 3.6888114 0.07873740 0.02975994 0.996 5 3.6947606 0.15360226 0.01275597 3.6952657 0.07904419 0.02987590 0.993 6 3.6949498 0.15554954 0.01291769 3.6998771 0.07793913 0.02945822 0.934 7 3.6982033 0.15187368 0.01261242 3.7000300 0.07765963 0.02935258 0.975 8 3.6965695 0.15078884 .01252233 3.6977043 00.07783147 0.02941753 0.984 9 3.6941683 0.14933956 0.01240198 3.6966014 0.07457322 0.02818603 0.966 10 3.6909294 0.14800686 0.01229130 3.6916357 0.07334534 0.02772193 0.990 Contras 1 315.1388486 177.70143713 14.75730112 339.7269071 251.49710240 95.05696977 0.726 2 499.2650374 300.94027582 24.99172961 552.7721929 479.90755321 181.38800544 0.656 3 656.7864110 395.39781826 32.83600155 669.0755014 489.84205140 185.14289281 0.937 4 801.8035399 484.31155031 40.21988509 779.0552014 512.88993163 193.85417272 0.904 5 937.9186496 571.04196611 47.42245409 885.5693447 553.34877610 209.14617855 0.813 6 1063.7877057 653.23207514 54.24797114 988.7686614 599.97791773 226.77033749 0.766 7 1181.3326292 732.63307678 60.84186542 1085.7139914 649.32149292 245.42045588 0.735 8 1291.0414610 809.84046662 67.25359016 1177.5454200 701.23899871 265.04342860 0.716 9 1396.3682258 889.44363204 73.86427323 1264.6367414 758.53064057 286.69763383 0.701 10 1494.5215899 966.88745656 80.29563279 1347.9700171 813.56216396 307.49759456 0.694 Anguler second moment 1 0.0003488 0.00015695 0.00001303 0.0003714 0.00017102 0.00006464 0.711 2 0.0002964 0.000012514 0.00001039 0.0002857 0.00005593 0.00002114 0.823 3 0.0002806 0.00012200 0.00001013 0.0002714 0.00005146 0.00001945 0.843 4 0.0002726 0.00012036 0.00001000 0.0002643 0.00005062 0.00001913 0.857 5 0.0002673 0.00011942 0.00000992 0.0002614 0.00005146 0.00001945 0.897 6 0.0002639 0.00011959 0.00000993 0.0002571 0.00004786 0.00001809 0.883 7 0.0002623 0.00011742 0.00000975 0.0002571 0.00004786 0.00001809 0.909 8 0.0002616 0.00011728 0.00000974 0.0002571 0.00004786 0.00001809 0.921 9 0.0002619 0.00011727 0.00000974 0.0002571 0.00004192 0.00001584 0.916 10 0.0002629 0.00011695 0.00000971 0.0002571 0.00004192 0.00001584 0.897 Invers differensia l moment 1 0.0547979 0.01312559 0.00109002 0.0520400 0.00878544 0.00332058 0.584 2 0.0473543 0.03097366 0.00257222 0.0437914 0.00605996 0.00229045 0.762 3 0.0401514 0.01083754 0.00090001 0.0385729 0.00577497 0.00218273 0.703 4 0.0384927 0.02187408 0.00181654 0.0361086 0.00613606 0.00231921 0.775 5 0.0344490 0.01014603 0.00084258 0.0340371 0.00624037 0.00235864 0.916 6 0.0326346 0.00972193 0.00080736 0.0323700 0.00596515 0.00225461 0.943 7 0.0310297 0.00974142 0.00080898 0.0312786 0.00615010 0.00232452 0.947 8 0.0296746 0.00940930 0.00078140 0.0299643 0.00546359 0.00206504 0.936 9 0.0285969 0.00945333 0.00078506 0.0285969 0.00945333 0.00078506 0.930 10 0.0276533 0.00925452 0.00076855 0.0279657 0.00546788 0.00206666 0.930 Mean of Hm(y,d) 1 133.5669281 27.19744514 2.25862488 154.4756100 16.19232902 6.12012510 0.046 2 133.9853676 27.26778725 2.26446648 154.9300529 16.22254719 6.13154650 0.046 3 134.3513773 27.32300711 2.26905224 155.2295157 16.21027822 6.12690927 0.047 4 134.6905923 27.37766341 2.27359120 155.4813714 16.17205405 6.11246189 0.049 5 135.0195647 27.41287327 2.27651522 155.6828743 16.11552700 6.09109667 0.050 6 135.3111239 27.45034260 2.27962687 155.8540200 16.07657784 6.07637527 0.052 7 135.5660110 27.48385842 2.28241021 156.0133371 16.08660616 6.08016562 0.053 8 135.8055795 27.49340697 2.28320317 156.1406957 16.10091118 6.08557241 0.055 9 136.0024805 27.52114376 2.28550659 156.2764029 16.13230995 6.09744003 0.056 10 136.1667878 27.54103176 2.28715820 156.4100486 16.18074467 6.11574663 0.056 Deviation 1 31.6089927 10.37731414 0.86178903 31.2089386 8.17771668 3.09088638 0.920 2 31.4520058 10.32891615 0.85776980 30.8363843 8.00177988 3.02438852 0.877 3 31.1554626 10.29381994 0.85485522 30.6475200 8.17395880 3.08946603 0.898 4 31.2711453 10.34187188 0.85884571 30.4715129 8.29049564 3.13351282 0.841 5 31.2027498 10.34497031 0.85910302 30.3944043 8.50868609 3.21598105 0.839 6 31.1005330 10.32576034 0.85750772 30.3165057 8.66131959 3.27367109 0.844 7 31.5819366 11.55983913 0.95999239 30.2518243 8.78782378 3.32148518 0.765 8 31.0323292 10.36960425 0.86114876 30.1991443 8.91417439 3.36924122 0.835 9 30.9988406 10.36905312 0.86110299 30.1733129 9.01785132 3.40842742 0.836 10 31.0513794 10.40834147 0.86436571 30.1827486 9.11795614 3.44626349 0.829 Entropy of Hdiff 1 1.5257865 0.10911401 0.00906143 1.5202071 0.07754101 0.02930775 0.894
- 7. ISSN: 2088-8708 Int J Elec & Comp Eng, Vol. 8, No. 4, August 2018 : 2442 – 2450 2448 Pixel IDC (n=148) ILC (n=7) Signifi kansi Average Standart deviation Standart error Average Standart deviation Standart error 2 1.6213262 0.11459042 0.00951622 1.6125686 0.07693016 0.02907687 0.842 3 1.6791019 0.12226794 0.01015380 1.6695557 0.07801198 0.02948576 0.838 4 1.7207762 0.12816197 0.01064327 1.7120586 0.08066301 0.03048775 0.859 5 1.7530858 0.13349584 0.01108623 1.7447471 0.08343252 0.03153453 0.870 6 1.7790535 0.13767171 0.01143301 1.7715600 0.08655722 0.03271555 0.887 7 1.7974672 0.14557475 0.01208932 1.7942429 0.09061765 0.03425025 0.954 8 1.8190492 0.14463779 0.01201151 1.8133386 0.09409468 0.03556445 0.918 9 1.8348149 0.14744603 0.01224473 1.8293657 0.09614239 0.03633841 0.923 10 1.8486772 0.15050684 0.01249891 1.8435843 0.09751869 0.03685860 0.930 Anguler second moment of Hdiff 1 0.0373317 0.00945719 0.00078538 0.0362171 0.00625452 0.00236399 0.758 2 0.0301529 0.00819906 0.00068090 0.0297257 0.00492143 0.00186012 0.892 3 0.0264829 0.00760508 0.00063157 0.0262943 0.00447068 0.00168976 0.948 4 0.0241821 0.00738957 0.00061367 0.0239829 0.00421100 0.00159161 0.944 5 0.0225104 0.00723454 0.00060080 0.0223143 0.00406143 0.00153508 0.943 6 0.0212292 0.00709542 0.00058924 0.0210329 0.00390195 0.00147480 0.942 7 0.0202617 0.00701517 0.00058258 0.0200086 0.00394125 0.00148965 0.925 8 0.0194140 0.00691035 0.00057387 0.0190671 0.00394584 0.00149139 0.896 9 0.0187663 0.00687225 0.00057071 0.0184157 0.00376733 0.00142392 0.894 10 0.0181572 0.00679038 0.00056391 0.0178014 0.00369565 0.00139682 0.891 Mean of Hdiff 1 13.0044315 3.28830418 0.27307880 12.8764514 2.58576160 0.97732602 0.919 2 16.3007437 4.35912747 0.36200583 16.0138443 3.47013242 1.31158677 0.864 3 18.7080357 5.24950020 0.43594726 18.1383900 3.75149392 1.41793142 0.777 4 20.6860886 5.98356049 0.49690765 19.8918086 4.10911592 1.55309983 0.729 5 22.3636332 6.65117494 0.55235001 21.4161543 4.52950474 1.71199187 0.710 6 23.8832428 7.25739935 0.60269421 22.7707043 4.91520935 1.85777451 0.689 7 25.2233172 7.82210544 0.64959050 23.9815157 5.38847605 2.03665251 0.679 8 26.4331844 8.35874188 0.69415573 25.0965229 5.79428972 2.19003566 0.677 9 27.5453989 8.89030016 0.73829924 26.0930457 6.16308459 2.32942702 0.670 10 28.5556697 9.39921379 0.78056222 27.0679543 6.53656128 2.47058794 0.680 Anova was conducted using IBM SPSS 20 software. The mean parameter is the only parameter that significantly distingguish IDC and ILC. At the mean analyis, the distance of 1-5 pixels are significantly distinguish IDC and ILC, while the higher distances are not sigificantly different. The decision system of CAD that we developed can fall into one of the four categories. The image area can be called ILC and IDC and a decision can be true or false. The CAD system that we developed can produce two false output types, namely False Positive (FP) and False Negative (FN). True Positive (TP) and True Negative (TN) is the true decision. Two working measurements of classification system which are related with identified decision above are „Sensitivity and Specificity‟. Sensitivity (Recall) is TP/(TP+FN) whereas Specificity is TN/(TN+FP). The high values of sensitivity and specificity are very expected. „Accuracy‟ and „Precision‟ are also used to evaluate the performance of KNN system. Accuracy is (TN + TP)/(TP+FN+TN+FP) and Precision is TP/(TP+FP). Error Rate is (FP + FN)/(TP+FN+TN+TP). To assess our algorithm, we examined 152 images from the data base of Dokter Soetomo Hospital, Surabaya, Indonesia. 145 (95%) images have IDC character and 7 (5%) images have ILC character. The results of our analysis are shown at Table 5. Table 5 The Result of Analysis Performance measure IDC cases ILC cases 145 (95%) 7 (5%) TP FN TN FP 102 (70%) 43(30%) 5 (71%) 2 (29%) Sensitivity (Recall) 70 % Specificity 71 % Accuracy 70 % Precision 98 % Error Rate 30 % Evaluating of the performance of the CAD system that we developed to classify histopathology types of IDC and ILC needed the definite criteria to determine cluster detection of TP and FP. To evaluate our result, true classification cluster was identified by a radiologist and an expert of Pathology Anatomy. The criteria we used to calculate the number of TP detection, assuming that a detection cluster was true if the
- 8. Int J Elec & Comp Eng ISSN: 2088-8708 The Utilization of Physics Parameter to Classify Histopathology… (Anak Agung Ngurah Gunawan) 2449 examination result from the radiologists and the experts of Pathology Anatomy (PA) was the same with the system decision of CAD that we developed. If the decision was assumed to be different, so it was assumed FP. It was like shown at table 5, our algorithm, the sensitivity was 70%, the specificity was 71%, the accuracy was 70%, the precision was 98%, and error rate was 30%. Two radiologists and experts of pathology and anatomy found our result was quite satisfying and could be reliable. 4. CONCLUSION The CAD system that we developed by using physics parameter was very helpful in classifying histopathology type of breast cancer. Mammogram image was very difficult to detect histopathology types. Even, the Radiology experts were not able to identify it for 100%. The utilization of physics parameter to classify histopathology types of breast cancer helped the experts of Radiology to get the second opinion. The final aim of the CAD system that we developed for mammography was to detect untouchable lesion which its size was often neglected on mammography. Detection of histopathology type increased the opportunity for women to be successful in the treatment of breast cancer. Our research was especially focused on histopathology types of IDC and ILC. We used physics parameter to classify histopathology types of breast cancer, after getting significant variable to distinguish histopathology types of IDC and ILC. Then, we used the parameter as the input variable of KNN method to take decision whether it included IDC or ILC histopathology types. According to the experts of radiology, the result produced by the CAD system that we developed was quite satisfying and could be reliable, and could assist the expert of radiology in diagnosing breast cancer. ACKNOWLEDGMENTS The authors would like to thank for the assistance given during the research to: Directorate of Research and Community Service (DRPM) as the research funder. Prof.Dr.dr.Ketut Suastika,SpPD-KEMD (Rector of Udayana University, Bali), Prof.Dr.Ir. I Nyoman Gde Antara,M.Eng (Chairman of LPPM Udayana University, Bali), Drs. I.B. Suaskara,M.Si (Dean of FMIPA Faculty, Udayana University, Bali), Prof.Dr.dr.Sri Maliawan,SP.BS(K) (Chairman of Research Ethics Commission FK UNUD/RSUP Sanglah, Denpasar), Prof. Dr.dr.Triyono KSP,Sp.Rad( K) (Chairman of Radiology Department of RSUP Dokter Soetomo Hospital, Surabaya), dr. A.A.A.N.Susraini,SpPA(K) (Chairman of Pathology Anatomy Department of RSUP Sanglah Hospital, Denpasar Bali), dr. Bagus G. Dharmawan, Sp.Rad (Chairman of Radiology Department of RSU. Prima Medika Hospital, Denpasar, Bali), Wirabrata (Director of Human Resource and Education of RSUP Sanglah Hospital Denpasar, Bali), dr. Putu Dian Ekawati, MPH (President Director of RSU. Prima Medika Hospital, Denpasar, Bali). REFERENCES [1] Satoko Nakano, Masahiko Ohtsuka, Akemi Mibu, at al., “Diagnostic imaging strategy for MDCT- or MRI-detected breast lesions: use of targeted sonography”, BMC Medical Imaging, 201212:13. [2] Jimmy Okello, Harriet Kisembo, Sam Bugeza, at al., “Breast cancer detection using sonography in women with mammographically dense breasts”, BMC Medical Imaging, 201414:41. [3] Sylvia H. Heywang-Köbrunner,Astrid Hacker, Stefan Sedlacek, “Advantages and Disadvantages of Mammography Screening”, Breast Care (Basel), 2011 Jun, vol. 6, no. 3, pp. 199-207. [4] Fernanda Philadelpho AP, Gabriela M, Maria Julia GC, at al., “Magnetic resonance imaging-radioguided occult lesion localization (ROLL) in breast cancer using Tc-99m macro-aggregated albumin and distilled water control”, BMC Medical Imaging, 201313:33. [5] Arjan P Schouten van der Velden, Carla Boetes, Peter Bult, at al., “Magnetic resonance imaging in size assessment of invasive breast carcinoma with an extensive intraductal component”, BMC Medical Imaging, 20099:5. [6] Bateman, Timothy, “Advantages and disadvantages of PET and SPECT in a busy clinical practice”, Journal of Nuclear Spinger Journal, Jan 19, 2012. [7] Romualdo Barroso-Sousa and Otto Metsger-Filho, “Difference between invasive lobuler and invasive duktal carcinoma of the breast: result and therapeutic implications”, Therapeutic Advances in Medical Oncology, vol. 8 no. 4, pp. 261-266. [8] Sunil RL., Lan OE., Stuart JS., at al., “WHO Classification of Tumours of The Breast”, International Agency for Research on Cancer, 4th Edition, Lyon, 2012. [9] A.A.N. Gunawan, Supardi, Dharmawan, “Readability Increase Of Mammography X-Ray Photos Results In Determining The Breast Cancer Histopathology Types Using Special Pattern Cropping With Physical Parameter”, Advances in Applied Physics, 2014, vol. 2, no. 1, pp. 43-52.
- 9. ISSN: 2088-8708 Int J Elec & Comp Eng, Vol. 8, No. 4, August 2018 : 2442 – 2450 2450 [10] A.A.N. Gunawan, “A novel model determination of breast cancer stage using physical parameter”, Far East Journal of Mathematical Sciences, 2014, vol. 87, no. 1, pp. 23-35. [11] Dhawan, A.P., “Analysis of Mammofraphic Microcalcification Using Gray Level Image Structure Feature”, IEEE,Trans.Medical Imaging, 1996, vol. 15, pp. 246-257. BIOGRAPHIES OF AUTHORS Dr. Anak Agung Ngurah Gunawan, M.T., was born in Denpasar on September 25, 1962. He obtained his Masters degree in 1999 in computer Engineering, Institut Teknologi 10 Nopember Surabaya Indonesia and his Dr degree from the faculty sains and teknologi, University of Airlangga Surabaya Indonesia. His main interess are image procecing. He has published 24 peper in journal international. He is a Senior Lecture Departement of physics University of Udayana at Bali Indonesia. I Wayan Supardi, S.Si, M.Si. Born on March 31, 1971, undergraduate degree (S1) in 1998, Department of Physics (Physics of Instrumentation of Electronics and Output) Faculty of Mathematics and Natural Sciences Udayana University, Strata 2 (S2) in 2004 Department of Physics (Instrumentation Physics) Bandung Institute of Technology (ITB ). Becoming a Lecturer at Physics Department in 1999. Field of instrumentation and energy penetration. Ir. S. Poniman, M.Si, born on June 6, 1956, has been a staff of physics faculty of mathematics and natural sciences of udayana university since 1986. Undergraduate degree (S1) at surabaya technological institute in 1985, postgraduate program (S2) Bandung in 1996. Field of superconductor research. The experience of the organization became chairman of the faculty of mathematics faculty of university udayana since 2011 until now.