Bounded-abstaining classification for breast tumors in imbalanced ultrasound images (Q2023639)

The paper is concerned with a classification of breast tumors based on ultrasound images. Given the imbalanced nature of data implying a high uncertainty of results, the proposed bounded-abstaining classifier is investigated, which is aimed at maximizing area under the ROC curve while considering two abstention constraints. Described is an overall design process composed of the following phases: image acquisition, wavelet reduction (to eliminate speckles), feature extraction, feature selection, and classification (realized with the use of kNN and SVM classifiers). The performance of the classifier is assessed with the aid of several performance indicators such as accuracy, area under curve, G-mean, sensitivity, specificity, among others. An effect of imbalanced data on the performance of classification results is investigated. Comparative analysis is carried out leading to the conclusion of the better performance of the developed classification model over its competitors.