Breast thermal images classification using optimal feature selectors and classifiers

Abstract

In this study, a full automatic technique has been presented to assist physicians in early detection of breast cancer based on different degrees. First the region of interest is determined using full automatic operation and the quality of image is improved. Then, some features including statistical, morphological, frequency domain, histogram and grey-level co-occurrence matrix features are extracted from segmented right and left breasts. Subsequently, to achieve the best features and increase the accuracy of the proposed method, feature selectors such as minimum redundancy and maximum relevance, sequential forward selection, sequential backward selection, sequential floating forward selection, sequential floating backward selection and genetic algorithm have been used. Finally, to classify and TH labeling procedures, supervised learning techniques such as AdaBoost, support vector machine, nearest neighbor, Naive Bayes and probability neural network are applied and compared with each other. The results obtained on native database showed the significant performance of the proposed algorithm in comprising to the similar studies. The experimental results gave the best mean accuracy of 88.03% for only using 0° image with combination of mRMR and AdaBoost and for combination of 3 degrees with combination of GA and AdaBoost.