Explainable Breast Cancer Detection Model

March–October 2025Independent ResearchAdvisor: Dr. Mehmet GulumCollaborators: Hieu Tran, Tri Dang, and Dat Nguyen

Abstract

In this work, we developed a Vision Transformer for breast cancer prediction with attention-based explanations to help clinicians interpret and verify model predictions.

Results

Correct benign mammogram prediction with its attention map — A correctly classified benign calcification example and the corresponding Vision Transformer attention map.

Malignant mammogram prediction with its attention map — A malignant prediction with attention concentrated on a localized region of the mammogram.

Second correct benign mammogram prediction with its attention map — A second correctly classified benign calcification example with localized model attention.

Misclassified malignant mammogram with its attention map — A malignant case predicted as benign, included to show a model failure and its diffuse attention pattern.