RadFormer: Transformers with global-local attention for interpretable and accurate Gallbladder Cancer detection

Open Webpage

Soumen Basu, Mayank Gupta, Pratyaksha Rana, Pankaj Gupta, Chetan Arora

Published: 01 Jan 2023, Last Modified: 09 Nov 2025Medical Image Anal. 2023EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•RadFormer is a novel DNN model for detecting Gallbladder Cancer from US images.•Global attention is fused with local bag of words style features using a transformer.•The bag of words style features are used to generate explanations for predictions.•Detection accuracy of RadFormer surpasses state-of-the-art DNNs and radiologists.•Semantically meaningful, human-readable explanations of decisions are provided.•Explanations are consistent with the radiological standards for diagnosis of GBC.