Classification of Breast Cancer in Mammograms Using an Optimized Hybrid Deep Learning Models and Feature Fusion Techniques

Abstract

Breast cancer is the most common deadly disease occurred in women. The major cause of the breast cancer agent is still not known. The early detection and treatment of breast cancer prevent the spreading of cancers to other parts and increase the lifetime of patients. Micro-calcification is one of the main signs of breast cancer. Mammography is a widely used digital screening approach to detect a microcalcification cluster in images. Compared to other image modalities, mammography is inexpensive and requires a low radiation dose. Image processing techniques with the aid of machine learning (ML) and deep learning (DL) techniques support radiologists to diagnose breast cancers earlier. In this work, the modified hybrid models with optimized feature selection models are proposed for accurate microcalcification classifications. The hybrid models of ResNet101V2 with Long short-term memory (LSTM) layers and ResNet101V2 with Support Vector Machine (SVM) classifiers are proposed for feature extraction. Then, the features are combined using serial-based feature fusion techniques. The performance of classifier models and feature selection process is improved using the metaheuristic algorithm of the Cheetahs optimizer. Experimental results on the MIAS breast cancer database show the superior performance of proposed models in terms of area under the curve (AUC), accuracy, and specificity and recall rates.