Federated Learning approach in Healthcare ecosystems for efficient Lung Cancer classification: Insights from Model Generic Training to Fine-Tuning and Transfer Learning

Early cancer detection is vital for improving treatment effectiveness, survival rates, and patient well-being. It enables better treatment options, reduces healthcare costs, and minimizes the emotional impact of late-stage cancer. However, most recent studies are conducted in centralized learning environments, raising privacy concerns. AI technology’s role in accurately diagnosing and predicting cancer can significantly enhance patient outcomes. To tackle privacy concerns, we created a federated learning (FL) method that, in contrast to centralized systems, collects characteristics from several contexts. FL allows collaborative model training using data from multiple sources while protecting patient privacy, making it a valuable tool in cancer detection. It combines datasets from different locations to create stronger models for more accurate cancer diagnostics. This study focuses on identifying malignant lung nodules in pulmonary CT scans and classifying lung cancer severity using Deep Learning (DL) methods like fine-tuning, transfer learning, and the CatBoost algorithm within a federated learning framework. The federated learning approach achieved an accuracy of 89.0% for lung cancer classification using a fine-tuned InceptionV3-based model, demonstrating its potential to improve patient outcomes and support more effective screening programs.