U-WOA: an unsupervised whale optimization algorithm based deep feature selection method for cancer detection in breast ultrasound images

Breast cancer, if found to be malignant, is a serious health issue that affects millions of women every year. Hence, the presence of any such disease requires prior attention to be treated with the best possible medical advice. Computational scientists have been constantly striving to contribute towards screening breast cancer using different image modalities for effective and economical detection. In this research, we propose a technique for categorizing breast cancer from breast ultrasound images utilizing deep learning and feature-selection algorithms based on meta-heuristic methods. We first extract features from ultrasound breast images using a standard deep learning model and following which we choose an optimal set of features from this deep feature set using an unsupervised version of the Whale Optimization Algorithm (U-WOA). We select the leader agent in U-WOA using a pseudo-fitness value which is calculated using three rank-based feature filter methods. The overall framework aims to reduce the presence of redundant and non-informative features extracted from the deep learner to build a more robust framework. The proposed approach yields state-of-the-art classification performances on a standard benchmark breast cancer ultrasound dataset namely BUSI.