A Federated Learning and Explainable AI-Enabled Multi-Modal Framework for Privacy-Preserving Breast Cancer Detection Using Capsule Networks, Transformers, and Feature Fusion with Interactive Clinician-Centric Dashboard Support

Abstract

Abstract - Breast cancer is a major global concern, where timely and precise diagnosis is essential for effective treatment. This research introduces a Multi-Modal AI-Based Breast Cancer Detection System that achieves 99% accuracy by combining Capsule Networks for the analysis of mammograms, Transformers for structured biopsy and genetic information, and a Feature Fusion Network to improve diagnostic reliability. To tackle privacy issues, Federated Learning (FL) facilitates decentralized model training across various hospitals without revealing sensitive patient information. Furthermore, Explainable AI (XAI) methods, such as SHAP for assessing feature importance, Grad-CAM for highlighting mammogram regions, and Contrastive Explanations for justifying decisions, enhance the transparency of AI predictions. An Interactive XAI Dashboard enables doctors to upload data, obtain real-time AI-supported diagnoses, and examine explanations, ensuring both trust and usability. This method improves breast cancer identification with notable precision, protection of privacy, and clarity of interpretation, positioning it as a viable option for clinical implementation.