Gossip Neural Networks for Invasive Ductal Carcinoma Detection Using Histopathological Images

This work explores the use of Gossip Neural Networks (GossipNNs) for decentralized deep learning in the diagnosis of breast cancer, and specifically, Invasive Ductal Carcinoma (IDC), on the Breast Histopathology Images dataset. GossipNNs operate with a peer-to-peer communication model, allowing weight updates to be shared between neighboring nodes in a completely decentralized manner, without relying on a central structure or a single server. This approach reduces communication overhead and enhances scalability, which makes it suitable for large-scale applications like medical imaging. The study evaluates significant performance metrics like accuracy, ROC curve, F1 score, training time, and communication cost. The results show that GossipNNs retain their robust classification performance and also provide an effective and scalable training process, making it a promising approach for real-world clinical applications.