Personalized Federated Learning for Histopathological Prediction of Lung Cancer

Lung cancer is a leading contributor to cancer-related fatalities worldwide, and histopathological image analysis plays a critical role in cancer detection by identifying morphological abnormalities in tissue samples. Artificial intelligence (AI) in medicine has evolved but still faces challenges, notably in maintaining data privacy and security. Federated Learning (FL) has emerged as a promising solution, enabling the training of robust models without jeopardizing data privacy. However, the effectiveness of existing FL approaches often falters in non-independent and identically distributed (non-IID) scenarios, where data distributions vary across clients. Addressing this challenge, our research presents a personalized federated learning (PFL) method specifically designed for lung cancer prediction using histopathological scans. This novel framework leverages client-specific autoencoders coupled with hierarchical clustering for personalized federated learning for lung cancer prediction. Our findings demonstrate the efficacy of the proposed method for collaborative lung cancer prediction in medical heterogeneous data environments while preserving data privacy.