RiIG-WCP Image-based Breast Tumors Classification Using Swin Transformer

Breast cancer remains a significant global health challenge, necessitating continued research and innovation in early detection and classification. This study introduces a novel approach for the classification of breast tumors in medical images, leveraging the power of Rician Inverse Gaussian (RiIG) modeled Weighted Contourlet Parametric (RiIG-WCP) imaging and the state-of-the-art Swin Transformer model. In this investigation, a two-step process was employed. Before collecting specific information, high-frequency features are first retrieved via contourlet transformation from the B-mode breast ultrasonography (US) images. Then, statistical modeling is carried out using the RiIG distribution. A swin transformer model is then utilized to learn spatial associations and categorize the transformed images into several tumor types after weighting the generated images by cross-correlation. Using a publicly accessible breast ultrasound dataset, we carried out various studies. Additionally, the conventional swin transformer model has been improved to perform better in a low-configuration GPU environment and classification tasks. In terms of classification accuracy, sensitivity, and specificity, our customized swin transformer model performed remarkably well in classifying breast tumors, scoring 99%, 100%, and 98.04%, respectively.