A Robust Hybrid CNN+ViT Framework for Breast Cancer Classification Using Mammogram Images
Vasudha Rani Patheda, Gunda Laxmisai, Gokulnath BV, Siddique Ibrahim S P, Selva Kumar S
Abstract
Breast cancer is the most frequent type of cancer largely experienced by women currently, although it could happen to men also. It appears when abnormal breast tissue cells grow rapidly and form tumors. Mammogram is a technique that is employed by doctors to analyse the breast in the diagnosis of early cancer. These mammograms are classified into Benign and Malignant. This research addresses the variability and potential oversight in radiologists’ manual mammogram interpretations, aiming to enhance classification accuracy by combining Convolution Neural Networks (CNNs) and Vision Transformers (ViTs). CNN is a successful image classification that uses hierarchical feature extraction, ViTs capture the global context but require substantial data and computation. In this research, we have used CLAHE-enhanced mammogram images from Kaggle for training and applied a CNN+ViT model. We have also used a few pre-trained models such as DenseNet, Inception, SE Resnet, and XceptionNet for comparative analysis. The CNN+ViT model gave us an accuracy of 90.1% showing robust performance. Although XceptionNet achieved perfect accuracy, it may indicate overfitting.