Reinventing Respiratory Diagnostics: Federated Learning & CNNs in Pneumonia Classification

A persistent worldwide health issue is Pneumonia, a lung illness that affects many people. This study explores a unique method for categorizing five types of pneumonitis that combine federated learning with convolutional neural networks (CNN). The research uses decentralized learning to gather localized insights and incorporate them into a complete global model using data from five distinct customers, each with a private dataset reflective of the different Pneumonia classes. Results analysis revealed positive performance indicators with accuracy, recall, and F1 scores averaging between 92 and 95 per cent across customers. Federated Averaging was also used to combine the findings from regional data sets smoothly. A multi-dimensional performance evaluation was provided by methodically computing the macro, micro, and weighted averages. The Micro average showed values between 87.22% and 95.29%, the Macro average was between 86.72% and 94.79%, and the Weighted average varied between 87.23% and 95.29%. The local data underwent a revolutionary recalibration to contribute to the global data via federated averaging. In conclusion, this work emphasizes the effectiveness of CNN and federated learning in delivering a scalable and reliable solution to the classification issue of Pneumonia, reiterating its promise as a diagnostic tool in the future.