Multi-Images Recognition of Breast Cancer Histopathological via Probabilistic Neural Network Approach

著者: Heba M. Afify, Kamel K. Mohammed, Aboul Ella Hassanien, A Azar, S El-Said, A Cruz-Roa, A Belsare, M Mushrif, M Pangarkar, N Meshram, A Khan, A Baig, H Asri, H Mousannif, H Al Moatassime, T Noel, M Aswathy, M Jagannath, J Burt, A Bentaieb, H Li-Chang, D Huntsman, G Hamarneh, K Deb, A Samir, P Amrit, T Meyarivan, M Dorigo, C Blum, F Spanhol, L Oliveira, P Cavalin, C Petitjean, L Heutte, Z Han, B Wei, Y Zheng, Y Yin, K Li, S Li, J Weston, C Watkins, Juanying Xie, Ran Liu, Joseph Luttrell, I, Chaoyang Zhang, C Lehman, M Kowal, P Filipczuk, M Vasantha, V Bharathi, R Dhamodharan, M Hu, N Suematsu, Y Ishida, A Hayashi, T Kanbara, M Nawaz, A Sewissy, T Soliman, P Filipczuk, T Fevens, A Krzyzak, R Monczak, F Spanhol, L Oliveira, C Petitjean, L Heutte, S Selvarajah, S Kodituwakku, M Tarek, Jin-Myung Hamdani, Adel Won, Fakhri Alimi, Karray, H Wu, J Phan, A Bhatia, B Shehata, M Wang, Y George, H Zayed, M Roushdy, B Elbagoury, Y Zhang, B Zhang, F Coenen, W Lu - Journal of System and Management Sciences 2020 被引用: 21

This paper suggests an approach for automated diagnosis of different types of breast carcinoma histopathological images via machine learning algorithms. The approach starts by extracting some features then use the nondominated genetic algorithm II (NSGA-II) algorithm and Ant Colony Optimization (ACO) to select the best features. Additionally, two classifiers including probabilistic neural network (PNN) and multi-class support vector machine (MSVM) is proposed to determine the type of breast cancer. These histopathological images database consider as a multi-classification task that used 848 BreakHis images with 400 magnification factors. The data divided into 160 samples for the testing phase and 688 samples for the training phase. The experiment results evaluated by four parameters including accuracy, sensitivity, specificity, and precision. For the NSGA-II algorithm, these outcomes refer to the maximum accuracy obtained through the combined features set with the accuracy being 100% by the training dataset and 82.5% by the testing dataset for the PNN classifier.