Using Pretrained Vision Transformer for Breast Cancer Binary Classification
Esam Mohammed Asem Othman · 2023
Breast cancer is presently the most prevalent cancer diagnosed among women on a global scale. Originating in the breast tissue, it stands as a primary cause of mortality in women. Timely detection of breast cancer plays a pivotal role in successful treatment, as early-stage diagnosis offers curative possibilities. Breast cancer is classified into two types: malignant and benign tumors. Malignant tumors pose a greater threat due to their accelerated growth rate, necessitating accurate identification of tumor type to determine the appropriate course of treatment for breast cancer patients. In this paper, we introduce an innovative approach employing Transformers for the classification of breast cancer images. Our proposed methodology leverages pretrained ViT Transformer, previously trained on an auxiliary domain, which are then integrated with an additional network comprising fully connected layers. Subsequently, this network is trained using breast cancer images in isolation. Our findings, based on the analysis of a benchmark dataset, demonstrate the remarkable classification accuracy of 97.5%achieved by our proposed model.