Performance Comparison of Classification Models for Identification of Breast Lesions in Ultrasound Images

Abstract

AbstractGlobally, breast cancer is the most common disease among women. A region endures from damage through any disease then the region is known as lesion. It is important to differentiate different types of breast lesions for proper treatment. Therefore, there is a significant impetus for the researchers in the development of computer-aided diagnostic (CAD) system that can assist clinicians in breast lesion diagnosis. This work presents the performance comparison of different renowned classifiers in terms of accuracy with which they identify the type of breast lesion in ultrasound (US) images of the given dataset. The CAD system is developed to assist the clinicians as a second opinion tool in identifying the type of breast lesion. In this system, ultrasound images are taken as inputs and the region of interest (ROI) for each US image is marked according to the shape of abnormality. Different texture features are extracted from the US images and further these images are classified into binary classes of malignant and benign using different classifiers. The performance of these classifiers is compared and it is observed that the law’s mask texture features of dimension 5 provided a maximum classification of 97.4% than other feature extraction methods applicable for the classification of two-class breast lesions.KeywordsBreast cancerUltrasound imagingCAD systemTexture featuresMachine learning