Modernizing Lung Cancer Detection: The Federated Learning CNN Approach

Deep learning paradigms, particularly Convolutional Neural Networks (CNNs), have carved out a place in imaging-based illness identification in the modern age of medical diagnostics. Through a decentralised approach of federated learning among six different clients, the study penetrates the field of lung cancer diagnosis, dissecting its severity over four gradient levels. With its capacity for dispersed data processing, federated learning enables on-site data training without transferring any data, protecting data privacy and improving scalability. The first Result Analysis examined how the federated CNN model performed for specific customers. In contrast to Client 6, who fluctuated between 89.50% and 93.26%, Client 1 had diagnostic accuracy across severity levels ranging from 85.80% to 95.76%. These variations highlight the inherent variations in local data and the flexibility of the approach after leaving the microcosm of local data to start exploring how local data transforms into a more comprehensive Global Data Model. When results were averaged, they varied from Client 1’s 90.80% to Client 3’s 88.36%. The model’s capacity to combine a wide range of disparate information into a single, unified diagnostic framework was highlighted by this consolidation. Finally, the study used three averaging metrics to evaluate the model’s effectiveness: macro, weighted, and micro. The macro average ranged from 88.40% (Client 3) to 92.67% (Client 2). Client 3 had a weighted average factoring class distribution of 90.25%, while Client 2 peaked at 92.91%. Client 3 scored 90.27% on the micro average based on aggregate classifications, whereas Client 2 continuously dominated at 92.91%.In conclusion, this study demonstrates the revolutionary potential of CNNs and federated learning for determining the severity of lung cancer.