Abstract

The identification of breast cancer on histopathology images helps pathologists who need precise and reliable computational methods. This research proposes a unique framework that integrates transformer-based categorization, graph-based tissue modeling, federated learning and Generative Adversarial Network (GAN) to improve diagnostic accuracy. To enhance feature consistency by standardizing histopathology images and to reduce inter-laboratory variances, a stain normalization GAN is used. We use SLIC to divide the tissue regions while maintaining cellular interactions and spatial connectivity and display them as a graph. Connectivity- Aware graph transformer uses connectivity-biased self-attention to capture both global and local topological relationships, which is used to process the retrieved graph features. Federated learning allows collaborative learning while protecting sensitive patient data by ensuring privacy preserving decentralized model training across several institutions. This method improves model robustness and generalization without centralizing data. The experimental assessment on openly accessible breast cancer datasets shows that our suggested framework performs better in terms of accuracy and interoperability than deep learning models. This paper presents a scalable, privacy-preserving, and clinically practical method for automated breast cancer diagnosis integrating transformer-based classification, connective graph representation, and GAN-based stain normalization within a federated learning paradigm.

Authors

S. Shiny, G. Vaishnavi, R. Kavya Lakshmi
Mepco Schlenk Engineering College, India

Keywords

GAN based Stain Normalization, Federated Learning, Tissue Graph Construction, Connectivity-Aware Graph Transformer, Breast Cancer Classification