Deep features and metaheuristics guided optimization-based method for breast cancer diagnosis

Breast cancer is one of the most prevalent cancer types and the second leading cause of death among women. But fortunately, early diagnosis and treatment of breast cancer reduces mortality rates and improves the quality of life significantly. In this work, a computer-aided three-stage pipeline for the diagnosis of breast cancer from mammogram images has been proposed. Firstly, an attention-aided VGG16 network has been applied for deep feature extraction from mammogram images. Then, the extracted features are passed to the adaptive beta hill climbing-aided whale optimization algorithm for the most informative feature selection. Lastly, the features from stage 2 are again evaluated by the grey wolf optimizer combined with the adaptive beta hill climbing algorithm for the selection of the best subsets of features and subsequent identification of cancers using the k-nearest neighbours classifier. The proposed three-stage method has achieved an average accuracy of 97.63% and has reduced 81.78% of the original features. Considering the fact that well annotated imaging data is a costly resource for modern data intensive deep networks, this work shows that a fairly lightweight off the shelf network like VGG-16 can be coupled with a robust feature selection policy to outperform several recent approaches which employ more data intensive learning strategies with larger networks.