Accelerating Lung Disease Diagnosis: The Role of Federated Learning and CNN in Multi-Institutional Collaboration

Abstract

This research employs federated learning using Convolutional Neural Networks (CNN) across multi-institutional datasets to classify the severity of lung disease. The project attempts to respect the strict privacy restrictions inherent to healthcare data while demonstrating the efficacy of this technique in a real-world, multi-client situation. The study included five customers with lung imaging data representing five different lung disease severity levels. A robust federated learning model and a CNN were trained and validated on these various datasets, allowing group learning without direct data exchange. Utilizing precision, recall, F1-Score, and accuracy measures, the model's performance was carefully examined at both the local and global levels. According to the local investigation, all customers showed remarkable performance, with an average accuracy between 97% and 99%. Each client's model effectively identified and categorized the degree of lung illness. Macro, micro, and weighted averages showed how the model might be adjusted to different clinical situations and data distributions, which the overall results confirmed further. The global model's performance was then assessed, and the results showed that it performed well in diagnosing lung disease severity across various institutional datasets, with an overall accuracy rate of 99%. This work highlights the enormous potential of federated learning in imaging and diagnostics in medicine. High-performance metrics attained at local and global levels point to a bright future for healthcare applications, especially in determining lung disease severity and promoting international partnerships in medical research.

Topics & Concepts