System for Breast Cancer Diagnosis: A Survey

Transcription

1 Chiew, T.K., et al. (Eds.): PGRES 2017, Kuala Lumpur: Eastin Hotel, FCSIT, 2017: pp System for Breast Cancer Diagnosis: A Survey Rasha, Maizatul Akmar Ismail, Amiruddin Kamsin and Saqib Hakak Department of Information Systems Faculty of Computer Science and Information Technology, University of Malaya, MALAYSIA rashaatallah@siswa.um.edu.my, maizatul@um.edu.my, amir@um.edu.my, saqibhakak@siswa.um.edu.my ABSTRACT One of the most cause of death in the world is breast cancer. This cancer increase the ration of the women death. Cancer happens when uncontrollable cell start appear in the human body and spread. But to reduce breast cancer fatality must detect and diagnose this disease early. Accurate classification of breast tumor is an important task in medical diagnosis. New technical computing starting help medical in diagnosis diseases and to improve the specialists doctors performance. The aim of this survey paper is to define the current state of research in breast cancer and to extract the limitation of the existing systems. There are many software based on neural network, support vector machine, fuzzy logic, deep learning and many others techniques are being used in medical. In this paper two type of technique have been studied deep learning and neural network. The comparison between existing approaches was done based on three main evaluation parameters i.e., accuracy, sensitivity and specificity along with data sets used. Keywords: Breast Cancer, Deep learning, Neural network INTRODUCTION Computer system techniques are used for classify and detect diseases in medical field. Breast cancer is one of the huge and most diagnosed cancers. A report by World Health Organization (WHO) published in 2013 has shown that cancer is one of the leading causes of death. Early detection of cancer saving approximately 37.3% patients (Salama, Abdelhalim et al. 2012). There are two type of cancer malignant and benign. The main issues is to diagnose malignant and benign cancer. Many studies have used computer technology to classify whether the cancer is benign or malignant tumours (Huaqing,et at 2016 ). However, this paper discovers the most recent works starting from 2015 onwards in the area of breast cancer detection. The main contribution of the article can be summarised as: This survey compare different approaches used for breast cancer involving latest works and discuss open issues for deep learning and neural network. Page 85

2 Rasha, Maizatul Akmar Ismail, Amiruddin Kamsin & Saqib Hakak Approaches used for Cancer Detection In order to carry out the survey work, several articles related to breast cancer techniques were reviewed and selected from credible sources such as: Web of Science (WoS), Science Direct, IEEE, Springer. The main methods used in these articles were classified as figure 1, Neural Network and Deep Learning. Figure 1 : The different approaches methods used for classifying the types of cancer The current paper integrates different classification techniques which are used at breast cancer. Table 1 provides list of the papers which dealt with single methods. Table 1: Publication of breast cancer techniques Classifier Types Author Aim Bayesian neural network (BNN) Cellular neural network (CNN) Neural network Cancer Breast Classification Using Extracted Parameters from a Terahertz Dielectric Model of Human Breast Tissue Detection of microcalcifications in digitized mammograms with multistable cellular neural networks using a new image enhancement method: automated lesion intensity enhancer (ALIE) Increase the classification tumor using BNN To propose a model using Cellular neural networks (CNNs) to remove pectoral muscle and unnecessary parts from the mammogram images Page 86

3 System for Breast CancerDiagnosis: A Survey Cellular neural network (CNN) Artificial Neural Network (ANN) Deep learning with Shear wave Elastography Convolution Neural Network Classification of benign and malignant breast tumors based on hybrid level set segmentation Robust mass classification-based local binary pattern variance and shape descriptors Deep learning based classification of breast tumors with shear-wave elastography A deep feature based framework for breast masses classification proposed two segmentation approaches for classifying benign and malignant tumours using CNN Detect masses on mammographic images based on the Local Binary Pattern Variance (LBPV) and shape descriptors,also using Artificial Neural Network (ANN) to classify the masses Deep Learning To propose deep learning architecture to classify malignant and benign breast tumours To implement? fine-tuning operation on the trained deep CNN model to acquire the feature extraction Deep learning with SAE Deep Learning based Visual Search Discrimination of Breast Cancer with Microcalcifications on Mammography by Deep Learning Probabilistic Visual Search for Masses Within Mammography Images using Deep Learning To improve the performance of an innovative deep learning model for classifying breast lesions. To work with an entire mammography image as input without the need for image segmentation The effectiveness of the previously listed techniques that were used at cancer breast classification is evaluated based on how correctly the methods classified cancer. The evaluation is made based on accuracy, specificity and sensitivity criteria. Table 3 indicates the analysis of these three key aspects in medical data using at breast cancer detect. Survey on Cancer Detection techniques As shown in table 3, this survey presents 2 Network and Deep Learning. The first method was Neural network, Cellular Neural Network (CNN) on of the methods when used it the accuracy increase from 82% to 96%, also used Artificial Neural Network (ANN) which improve the accuracy to 91%, the best become 100% when used Bayesian Neural network. The second method was deep learning, when deep learning combined with visual search, the accuracy was 85%, also when Stacked auto encoder (SAE) used to build deep learning network, the accuracy improved to 87.3%, The accuracy also improved to 91.3% by adding shear wave elastography (SWE), after add different layers at deep learning network the accuracy become 96.7%. Page 87

4 Rasha, Maizatul Akmar Ismail, Amiruddin Kamsin & Saqib Hakak As seen in table 3, the data set which used for evaluation collected from various resources as Digital Database for Screening Mammography (DDSM), Wisconsin Breast Cancer Dataset (WBCD), Mammography Image Analysis Society (MIAS) and Wisconsin Original Breast Cancer (WOBC). The least number of images were 22 images, the images were collected from Mammography Image Analysis Society (MIAS) dataset, and the result of evaluation was 82% accuracy, sensitivity was 90.9% and specificity was 52.2% by using cellular neural network. The most number of images were 2803 images. This images were collected from Wisconsin Diagnostic Breast Cancer (WDBC), Wisconsin Original Breast Cancer (WOBC), Ljubljana Breast Cancer Dataset, METABRIC Breast Cancer Dataset and Netherlands Cancer Institute (NKI) Dataset, through this paper the evaluation was 96% for accuracy by using deep learning network. Table 2 :Performance evaluation summary for the surveyed methods Paper Breast Cancer Classification Using Extracted Parameters from a Terahertz Dielectric Model of Human Breast Tissue Data set Data Method References Accuracy Sensitivity (TPF) Specificity (TNF) 74 Female Bayesian % patients neural Italy research network committee (BNN) Page 88

5 System for Breast CancerDiagnosis: A Survey Detection of microcalcifica tion in digitized mammogram s with multistable cellular neural networks using a new image enhancemen t method: automated lesion intensity enhancer (ALIE) Classification of benign and malignant breast tumors based on hybrid level set segmentatio n Robust mass classificationbased local binary pattern variance and shape descriptors 100 Mammographi c Image Analysis Society (MIAS) database 57 Mammographi c Image Analysis Society (MIAS) database 600 FARABI database from EL FARABI radiologic centre Cellular neural networks- CNN Cellular neural network (CNN) Local Binary Pattern Variance (LBPV) with Artificial Neural Network (ANN) % 92.70%, 90.54%, Page 89

6 Rasha, Maizatul Akmar Ismail, Amiruddin Kamsin & Saqib Hakak Deep learning based classification of breast tumors with shear-wave elastography 227 National Natural Science Foundation of China (and the Shanghai Natural Science Fund deep learning (DL) with shearwave elastograp hy (SWE %, 88.6%, 97.1%, A deep feature based framework for breast masses classification Discriminatio n of Breast Cancer with Microcalcific ations on Mammograp hy by Deep Learning 600 Digital Database for Screening Mammography (DDSM) 1204 Nanhai Affiliated Hospital of Southern Medical University convolutio nal neural network (CNN) Deep learning Probabilistic Visual Search for Masses Within Mammograp hy Images using Deep Learning 2500 Digital Database for Screening Mammography (DDSM) Deep Learning 19 85% 85% State-of-the art for breast cancer techniques Neural network (Civcik, Yilmaz et al. 2015) used Cellular neural networks (CNNs) to remove pectoral muscle and unnecessary parts from the mammogram images, supported the model by Automated Lesion intensity Enhancer (ALIE) to enhancing lesion intensities. They also used the multistable CNNs. After applying the combination of these methods on the MIAS database, the accuracy become 82.0%. In 2015 Masmoudi and others researchers (Masmoudi, Ben Ayed et al. 2015) proposed an approach for detect masses on mammographic images based on the Local Binary Pattern Variance (LBPV) and shape descriptors,also using Artificial Neural Network (ANN) to classify the masses. The results indicate that the approach achieves accuracy over 91% which are applied to 600 mammographic, 300 for training and 300 for testing. Also (Rouhi, Jafari et al. 2015) proposed two segmentation approaches for classifying benign and malignant tumours using mammograms. One segmentation method consists of region growing method and other method consists of cellular neural network (CNN) approach. After necessary pre-processing of input mammogram, using region growing method and application of artificial neural network (ANN), necessary features from benign and malignant classification are extracted. In the CNN based approach, existing CNN model is applied for segmentation and classification purposes. The CNN Page 90

7 System for Breast CancerDiagnosis: A Survey model is trained using genetic algorithm. In these both methods, tumour boundary information is preserved to detect malign and benign tumours in mammograms. The average of the accuracy is 96%. (Truong, Tuan el at 2015) used Bayesian neural network (BNN) instead of using support vector machine (SVM) to increase the accuracy of tumour classification by using combinations of four model parameters. The dataset is 74 sample, they use 80% of the data (59 samples) for training and 20% (15 samples) for testing. BNN successfully classifies the data using the combinations of four model parameters with an accuracy of 97.3%.. Table 3: Neural network Methods Authors Methods Classification Accuracy (%) (Civcik, Yilmaz et al. 2015) CNN 82.0% (Masmoudi, Ben Ayed et al. ANN 91% 2015) (Rouhi, Jafari et al. 2015) CNN 96% Truong, Tuan el at 2015 BNN 97.3%. Deep Learning A visual search engine developed based on deep learning to classify if the tissues is mass or not at the mammography images.(ertosun and Rubin 2015).The model obtained 85% accuracy. Also deep learning used with the stacked auto encoder (SAE) to create a deep network by stacking multiple auto encoders hierarchically to improve the performance of an innovative deep learning model for classifying breast lesions (Wang, Yang et al. 2016). It is applied at SunYat-sen University Cancer Center (Guangzhou, China) and Nanhai Affiliated Hospital of Southern Medical University (Foshan, China ). The accuracy increased to 87.3%. When Zhang, Xiao (Zhang, Xiao et al. 2016) proposed deep learning architecture to classify malignant and benign breast tumours. The (Zhang, Xiao et al. 2016) image representations using point-wise gated Boltzmann machine (PGBM) technique. SWE images of 291 patients after all necessary pre-processing were used to evaluate the proposed architecture. (Jiao, Gao et al. 2016) applied fine-tuning operation on the trained deep CNN model to acquire the feature extraction. It is applied at DDSM dataset accuracy 96.7%. Table 4 : Deep Learning Methods Authors Methods Classification Accuracy (%) (Ertosun and Rubin 2015) Deep learning-based visual 85% search (Wang, Yang et al. 2016) Deep learning with SAE 87.3% Zhang, Xiao et al Deep learning with Shear wave 91.3% Elastography (Jiao, Gao et al. 2016) Deep learning : Deep features from different layers 96.7% RESEARCH GAPS Challenges and Open Issues Accuracy An accurate classifier is the most important component of any CAD. (Abdel-Zaher and Eldeib 2016). To achieve accurate breast cancer classification is still challenging due to the unknown cause of the disease and the similarities between benign and malignant masse. (Verma, McLeod et al.) Page 91

8 Rasha, Maizatul Akmar Ismail, Amiruddin Kamsin & Saqib Hakak Pattern Breast tumour SWE images contain artefact, noise, and other irrelevant patterns, such as irregular stiffness distributions. (Zhang, Xiao et al.), also the challenge is how to learn robust representations that can distinguish useful (i.e., task-relevant) patterns from large amounts of distracting (i.e., task-irrelevant) patterns (Nair and Hinton 2009). The important challenge is how to understand and utilize the patterns that may be task-relevant but are difficult to interpret by human observers, such as the black holes absent of colour on SWE, the missing areas with invalid stiffness values. (Zhang, Xiao et al. 2016) CONCLUSION In this survey paper, we have discussed various techniques system designed using different soft computing techniques from 2015 to These expert systems are widely used in medical field for classification and diagnosis of breast cancer tumour. The key concept of soft computing technologies is to build an automated system that learns from the decision parameter of the diseases so that the designed system can be used to diagnose and treatment of patient with unknown disease symptoms. REFERENCES Salama, G. I., M. Abdelhalim and M. A.-e. Zeid (2012). "Breast Cancer Diagnosis on Three Different Datasets Using Multi-Classifiers." International Journal of Computer and Information Technology 1(1): Stewart BW, Wild CP, editors. World cancer report 2014 Lyon: International Agency for Research on Cancer; Abdel-Zaher, A. M. and A. M. Eldeib (2016). "Breast cancer classification using deep belief networks." Expert Syst. Appl. 46(C): Benndorf, M., E. Kotter, M. Langer, C. Herda, Y. R. Wu and E. S. Burnside (2015). "Development of an online, publicly accessible naive Bayesian decision support tool for mammographic mass lesions based on the American College of Radiology (ACR) BI-RADS lexicon." European Radiology 25(6): Cao, Y., X. Hao, X. Zhu and S. Xia (2010). "An adaptive region growing algorithm for breast masses in mammograms." Frontiers of Electrical and Electronic Engineering in China 5(2): Civcik, L., B. Yilmaz, Y. Ozbay and G. D. Emlik (2015). "Detection of microcalcification in digitized mammograms with multistable cellular neural networks using a new image enhancement method: automated lesion intensity enhancer (ALIE)." Turkish Journal of Electrical Engineering and Computer Sciences 23(3): Cordeiro, F. R., W. P. Santos and A. G. Silva (2016). "A semi-supervised fuzzy GrowCut algorithm to segment and classify regions of interest of mammographic images." Expert Systems with Applications 65: da Rochaa, S. V., G. Braz, A. C. Silva, A. C. de Paiva and M. Gattass (2016). "Texture analysis of masses malignant in mammograms images using a combined approach of diversity index and local binary patterns distribution." Expert Systems with Applications 66: Ertosun, M. G. and D. L. Rubin (2015). Probabilistic Visual Search for Masses Within Mammography Images using Deep Learning. Proceedings 2015 Ieee International Conference on Bioinformatics and Biomedicine. J. Huan, S. Miyano, A. Shehu et al. New York, Ieee: Huang, M. W., C. W. Chen, W. C. Lin, S. W. Ke and C. F. Tsai (2017). "SVM and SVM Ensembles in Breast Cancer Prediction." Plos One 12(1): 14. Page 92

9 System for Breast CancerDiagnosis: A Survey Jiao, Z., X. Gao, Y. Wang and J. Li (2016). "A deep feature based framework for breast masses classification." Neurocomputing 197: Kim, S. (2016). "Weighted K-means support vector machine for cancer prediction." Springerplus 5: 11. Kim, W., K. S. Kim and R. W. Park (2016). "Nomogram of Naive Bayesian Model for Recurrence Prediction of Breast Cancer." Healthcare Informatics Research 22(2): Korkmaz, S. A. and M. Poyraz (2015). "Least Square Support Vector Machine and Minumum Redundacy Maximum Relavance for Diagnosis of Breast Cancer from Breast Microscopic Images." Procedia - Social and Behavioral Sciences 174: Masmoudi, A. D., N. G. Ben Ayed, D. S. Masmoudi and R. Abid (2015). "Robust mass classification-based local binary pattern variance and shape descriptors." International Journal of Signal and Imaging Systems Engineering 8(1-2): Nair, V. and G. E. Hinton (2009). Implicit mixtures of restricted Boltzmann machines. Advances in neural information processing systems. Ohri, K., H. Singh, A. Sharma and Ieee (2016). "Fuzzy Expert System for diagnosis of Breast Cancer." Proceedings of the 2016 Ieee International Conference on Wireless Communications, Signal Processing and Networking (Wispnet): Onan, A. (2015). "A fuzzy-rough nearest neighbor classifier combined with consistencybased subset evaluation and instance selection for automated diagnosis of breast cancer." Expert Systems with Applications 42(20): Rouhi, R., M. Jafari, S. Kasaei and P. Keshavarzian (2015). "Benign and malignant breast tumors classification based on region growing and CNN segmentation." Expert Systems with Applications 42(3): Salama, G. I., M. Abdelhalim and M. A.-e. Zeid (2012). "Breast Cancer Diagnosis on Three Different Datasets Using Multi-Classifiers." International Journal of Computer and Information Technology 1(1): Singh, B. K., K. Verma and A. S. Thoke (2016). "Fuzzy cluster based neural network classifier for classifying breast tumors in ultrasound images." Expert Systems with Applications 66: Verma, B., P. McLeod and A. Klevansky A novel soft cluster neural network for the classification of suspicious areas in digital mammograms. Zhang, Q., Y. Xiao, S. Chen, C. Wang and H. Zheng "Quantification of Elastic Heterogeneity Using Contourlet-Based Texture Analysis in Shear-Wave Elastography for Breast Tumor Classification." Ultrasound in Medicine and Biology 41(2): Zhang, Q., Y. Xiao, W. Dai, J. Suo, C. Wang, J. Shi and H. Zheng (2016). "Deep learning based classification of breast tumors with shear-wave elastography." Ultrasonics 72: Zhang, Q., Y. Xiao, W. Dai, J. F. Suo, C. Z. Wang, J. Shi and H. R. Zheng (2016). "Deep learning based classification of breast tumors with shear-wave elastography." Ultrasonics 72: Zhang, Q., Y. Xiao, W. Dai, J. Suo, C. Wang, J. Shi and H. Zheng (2016). "Deep learning based classification of breast tumors with shear-wave elastography." Ultrasonics 72: Page 93

10 Rasha, Maizatul Akmar Ismail, Amiruddin Kamsin & Saqib Hakak Civcik, L., B. Yilmaz, Y. Ozbay and G. D. Emlik (2015). "Detection of microcalcification in digitized mammograms with multistable cellular neural networks using a new image enhancement method: automated lesion intensity enhancer (ALIE)." Turkish Journal of Electrical Engineering and Computer Sciences 23(3): Ertosun, M. G. and D. L. Rubin (2015). Probabilistic Visual Search for Masses Within Mammography Images using Deep Learning. Proceedings 2015 Ieee International Conference on Bioinformatics and Biomedicine. J. Huan, S. Miyano, A. Shehu et al.: Jiao, Z., X. Gao, Y. Wang and J. Li (2016). "A deep feature based framework for breast masses classification." Neurocomputing 197: Jiao, Z. C., X. B. Gao, Y. Wang and J. Li (2016). "A deep feature based framework for breast masses classification." Neurocomputing 197: Masmoudi, A. D., N. G. Ben Ayed, D. S. Masmoudi and R. Abid (2015). "Robust mass classification-based local binary pattern variance and shape descriptors." International Journal of Signal and Imaging Systems Engineering 8(1-2): Nair, V. and G. E. Hinton (2009). Implicit mixtures of restricted Boltzmann machines. Advances in neural information processing systems. Rouhi, R., M. Jafari, S. Kasaei and P. Keshavarzian (2015). "Benign and malignant breast tumors classification based on region growing and CNN segmentation." Expert Systems with Applications 42(3): Salama, G. I., M. Abdelhalim and M. A.-e. Zeid (2012). "Breast Cancer Diagnosis on Three Different Datasets Using Multi-Classifiers." International Journal of Computer and Information Technology 1(1): Verma, B., P. McLeod and A. Klevansky A novel soft cluster neural network for the classification of suspicious areas in digital mammograms. Wang, J. H., X. Yang, H. M. Cai, W. C. Tan, C. Z. Jin and L. Li (2016). "Discrimination of Breast Cancer with Microcalcifications on Mammography by Deep Learning." Scientific Reports 6. Zhang, Q., Y. Xiao, S. Chen, C. Wang and H. Zheng "Quantification of Elastic Heterogeneity Using Contourlet-Based Texture Analysis in Shear-Wave Elastography for Breast Tumor Classification." Ultrasound in Medicine and Biology 41(2): Zhang, Q., Y. Xiao, W. Dai, J. Suo, C. Wang, J. Shi and H. Zheng (2016). "Deep learning based classification of breast tumors with shear-wave elastography." Ultrasonics 72: Zhang, Q., Y. Xiao, W. Dai, J. F. Suo, C. Z. Wang, J. Shi and H. R. Zheng (2016). "Deep learning based classification of breast tumors with shear-wave elastography." Ultrasonics 72: Page 94