Federated Learning and Optimization for Lung Cancer Detection Using FedResNet Model

Lung cancer detection through medical imaging remains a critical challenge due to privacy concerns and limited data availability across institutions. This study presents a federated learning-based diagnostic model, FedResNet, integrating ResNet-50 with Federated Averaging (FedAvg) and differential privacy to enable secure, collaborative training across distributed medical nodes. The framework employs MRI image preprocessing involving grayscale normalization and contrast enhancement, followed by feature extraction using a modified ResNet-50 architecture. Federated optimization combines FedSGD for local gradient updates and FedAvg for global model aggregation, achieving efficient convergence. Experimental evaluation on a lung cancer MRI dataset comprising 2,500 images achieved 98.44% accuracy, 98.70% recall, and 98.12% precision, outperforming baseline CNN and CNN-FL models. The proposed FedResNet model demonstrates strong diagnostic performance while preserving patient data confidentiality, making it suitable for privacy-sensitive healthcare environments.