Pushing Diagnostic Frontiers: Federated Learning CNN for Diverse Lung Disease

This study explores the possibility of Federated Learning combined with Convolutional Neural Networks (CNN) for improved lung disease diagnoses in the age of developing digital health technologies. The research uses data from five different customers. It is based on a dataset separated into five kinds of lung disorders, providing a broad and detailed perspective on the diagnostic skills of the suggested model. With accuracy rates ranging between 88.43% and 96.69%, it was determined from the local data analysis of five clients that each customer had a distinct diagnostic aptitude, highlighting the CNN model’s resilience in various regional settings. The overall picture was startling when these individual results were combined using Federated Averaging. Client 3 demonstrated, for example, an exceptional accuracy of 93.68% with a recall of 93.65%, resulting in an F1-score of 93.66% for a massive 900.00 pictures, while the AUC metric generally stayed around 0.97-0.98 for most customers. Moving away from the averaging methods, the macro-averaging approach produced accuracy, recall, and F1-score values of 92.46%, 92.46%, and 93.66%, respectively, supporting the model’s capacity to treat all classes equally. The accuracy, recall, and F1-score of the weighted averaging, which is sensitive to class imbalances, were 92.93%, 92.85%, and 93.83%, respectively. The weighted average and the Micro Averaging, which examines the model’s overall performance, were similar, pointing to a consistent dataset across the many lung disease subtypes. In conclusion, this study’s ground-breaking combination of federated learning with CNNs signals a new era in the diagnosis of lung diseases. Setting the scene for unmatched diagnostic skills in lung illnesses, the rigorous analysis of local customer data, supported by complex averaging procedures, guarantees that the obtained insights are both granular and internationally applicable.