Lung Abnormalities Detection Using Adaptive Weighted Aggregation based Federated Learning on Chest X-Ray Images

The increasing global prevalence of COVID-19 underscores the urgent need for the rapid development of consistent diagnostic approaches.Personal health data is protected by severe privacy regulations, making it challenging to implement centralized data-driven approaches in the medical sector, where personalized training data plays a crucial role.Federated Learning (FL) offers a promising solution by providing a decentralized approach to this problem.Thus, this research proposes the Adaptive Weighted Aggregation based FL (AWA-FL) approach for the secure detection of lung abnormalities.The AWA approach allocates fewer resources to clients with less relevant updates, thereby supporting the improvement of the overall efficiency of the FL system.Initially, this research involves Chest X Ray (CXR) images, including COVID-19 and Pneumonia, for the implementation and estimation of the proposed AWA-FL approach.Next, pre-processing techniques like image resizing and data augmentation techniques are applied to enhance the dataset and achieve better detection results.Experimental results demonstrate that the proposed AWA-FL approach attains higher accuracy results of 98.29% and 98.39% on the COVID-19 and Pneumonia datasets, respectively, when compared to the existing approaches: SecureFed and Fl-DenseNet.