Embracing Federated Learning and CNN for Distributed Diagnosis of Lung Diseases: A Novel Approach

This study distrubuted learning and machine learning to diagnose lung illnesses, is described in this research report. The experimental design examines four different healthcare customers and six different kinds of lung disorders. The main reason for implementing FL is the pressing necessity to protect patient data privacy. FL is unique because it enables training on decentralized data without explicit sharing. The suggested technique is used to show strong performance in identifying different lung illnesses. To prevent learning from being skewed toward anyone dataset, the local datasets of each client were fed into the model, resulting in a global model with more broad and nuanced learning. For each client and type of condition, the findings were astonishingly constant, with precision, memory, and F1 scores ranging from around 95% to 98% and an unchanging accuracy of 0.99 across all courses and customers. The model was assessed by three distinct averaging techniques-Macro, Weighted, and Micro. Results for the macro-average varied from 89.08% to 91.36%, showing the model's effectiveness across several classes. Results ranged from 90.38% to 91.89%, indicating the model's capacity to adjust to unbalanced datasets using weighted averages. Results ranged from 90.34% to 91.59% for the micro-average, representing the model's global accuracy and recall skills. The robustness of the strategy is shown by the consistently strong performance across a vast clientele and several classifications of lung disorders. This study advances healthcare analytics by showing how sophisticated machine learning models can accurately identify lung ailments while maintaining data privacy. These models are trained via federated learning approaches.