Breast Cancer Classification on Enhanced Segmented Mammograms Using Optimized Convolutional Neural Networks

Abstract

Breast cancer ranks as the second most common malignancy among women and the second-most common reason for cancer deaths worldwide. Digital Mammogram screening can offer low-cost early diagnosis and reduce the breast cancer fatality rate among victims. This research aims to build a model that automatically assists in classifying malignant and benign lesions depicted on digital mammograms without any human interventions. The Mammographic Image Analysis Society (mini-MIAS) image dataset, which contains 322 mammograms, is employed in the present study. This research focuses on the Background Preserved and Feature-Oriented Contrast Improvement (BPFO-CI) method for contrast enhancement that uses the Weighted Cumulative Distribution Function. The Region of Interest (RoI) is then extracted from the improved mammograms using the Thresholding Segmentation method. Then extracted RoIs are used as input for classification using optimal Convolutional Neural Networks (CNN). Data augmentation is applied to the pre-processed dataset. The suggested pre-processed CNN model's performance is compared to various classification algorithms in pertaining to accuracy and confusion matrix. The simulation results confirm the importance and effectiveness of the suggested model in comparison to other well-known conventional approaches. As a result, this classification method is predicted to aid in the diagnosis of breast cancer.