RM-CC Image-Based Breast Tumor Classification Using Vision Transformer

This research aimed to develop an effective method for classifying breast tumors using a combination of contourlet transformation, parametric imaging, and vision transformer techniques applied to RiIG (Rician and inverse Gaussian) modeled contourlet coefficient (RM-CC) images. This research employed a two-step approach. First, contourlet transformation is applied to the B-mode breast ultrasound (US) images to extract high-frequency features and capture detailed information. Then, statistical modeling is applied using RiIG distribution, and a vision transformer (ViT) model is utilized to learn spatial relationships and classify the transform domain statistically modeled images into different tumor classes. A publicly available dataset is induced. Moreover, the traditional ViT model is fine-tuned to operate in a low-configuration GPU environment and ensure better classification performance. Our customized ViT model has shown remarkable performance in breast tumor classification, which in terms of classification accuracy, sensitivity, and specificity are 99.72%, 99.79%, and 99.57%, respectively.