Improving the classification performance of breast ultrasound image using deep learning and optimization algorithm

Breast cancer is a common and major health concern in women across the world. Early prognosis can ensure better and cost-effective treatment thus reducing the mortality rate. In recent years researchers are proposing different models for the early prediction of breast cancer. In this paper computer-aided diagnosis (CAD) using breast ultrasound images has been proposed for multi-classification that is to classify tumors into benign, malignant, or normal. We propose a hybrid approach by combining a pre-trained Convolutional neural network (CNN) with optimization and machine learning for tumor diagnosis. The CNN pre-trained model ResNet-50 is used in this study for feature extraction, binary gray wolf optimization (BGWO) for feature selection, and classification Support vector machine (SVM). We examine two different approaches: a Transfer Learning (TL) approach and a machine learning approach and we assess their performance with and without BGWO for feature selection. There was a total of 780 images in the BUSI dataset, with 437 benign, 210 malignant, and 133 normal images. The classification accuracy and the Area under the receiver operating characteristics (AUROC) of the proposed method were 84.9% and 0.97. The computation time taken is also reduced with feature selection to 4.41 seconds from 18.275 seconds (without feature selection). Finally, when tested on the BUSI dataset, the results show that the feature selection method improves overall performance.