Enhancing Lung Cancer Detection with Advanced Federated Learning Aggregation Techniques

Early detection of lung cancer is vital for improving patient survival rates, yet achieving high accuracy remains a significant challenge due to the heterogeneous nature of medical data originating from various institutions. Federated Learning (FL) has emerged as a promising paradigm that enables collaborative model training across decentralized datasets while ensuring data privacy by keeping sensitive patient information on local servers. Despite its advantages, FL struggles with data heterogeneity, particularly when handling non-Independent and Identically Distributed (non-IID) data, which can hinder model convergence and degrade performance. To address these challenges, this study proposes an enhanced FL-based model for lung cancer detection that integrates the K-Nearest Neighbors (KNN) classifier with advanced aggregation techniques. Specifically, the proposed framework employs three distinct FL aggregation methods “FedAvg+, FedProx, and FedMA” to assess their effectiveness in handling diverse, distributed medical imaging data. Each aggregation strategy was evaluated independently to identify the most suitable method for optimizing classification performance while preserving data confidentiality. Experimental results reveal that the FedMA aggregation method achieves the highest accuracy of 99.28%, outperforming the others in terms of sensitivity, specificity, and precision. These results demonstrate that incorporating advanced aggregation techniques within the FL framework significantly improves diagnostic accuracy, model robustness, and adaptability across diverse healthcare environments. By ensuring both high predictive performance and strong privacy protection, the proposed model offers a scalable and secure solution for implementing AI-powered diagnostic systems in real-world medical settings, thereby supporting more reliable and ethically responsible approaches to lung cancer detection.