Title: New methods based on mRMR_LSSVM and mRMR_KNN for diagnosis of breast cancer from microscopic and mammography images of some patients

Authors: Sevcan Aytaç Korkmaz; Mustafa Poyraz; Abdullah Bal; Hamidullah Binol; Ibrahim Hanifi Özercan; Mehmet Fatih Korkmaz; Ayşe Murat Aydin

Addresses: Department of Electrical-Electronic Engineering, Engineering Faculty, Firat University, 23100 Elazığ, Turkey ' Department of Electrical-Electronic Engineering, Engineering Faculty, Firat University, 23100 Elazığ, Turkey ' Department of Electrical-Electronic Engineering, Engineering Faculty, Yildiz Technical University, Istanbul, Turkey ' Department of Electrical-Electronic Engineering, Engineering Faculty, Yildiz Technical University, Istanbul, Turkey ' Department of Pathology, Medicine Faculty, Firat University, 23100 Elazığ, Turkey ' Department of General Surgery, Medicine Faculty, Firat University, 23100 Elazığ, Turkey ' Department of Radiology, Medicine Faculty, Firat University, 23100 Elazığ, Turkey

Abstract: The aim of this study is to determine cancerous lesions in light microscopic and mammographic images taken from some patients. In this study, 23 features are used. These features obtained 92 features by rotating in variety of angles. Structure of the study composes three steps. These are feature select step, classification step and testing stage. In feature select step, optimal feature subset using minimum redundancy and maximum relevance via mutual information (mRMR) have been found. In classification step, Least Square Support Vector Machine (LSSVM) and fuzzy k-nearest neighbour (KNN) are used. For validation of the proposed methods accuracy rates are found. These accuracy rates, with mRMR_KNN, have obtained 100% and 98.33% in microscopic and mammographic images respectively. With mRMR_LSSVM 100% and 96.67% accuracies are obtained in microscopic and mammographic images respectively. When these microscopic and mammography images have been combined, mRMR_KNN and mRMR_LSSVM methods have found 100% and 100% accuracy rate respectively.

Keywords: breast histology images; mammography; least squares SVM; LSSVM; support vector machines; fuzzy k-NN classifier; k-nearest neighbour; feature selection; minimum redundancy; maximum relevance; breast cancer diagnosis; mammograms; microscopic images; classification accuracy; breast cancer screening.

DOI: 10.1504/IJBET.2015.072930

International Journal of Biomedical Engineering and Technology, 2015 Vol.19 No.2, pp.105 - 117

Received: 16 Oct 2014
Accepted: 10 Nov 2014
Published online: 08 Nov 2015 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article