Federated Learning CNNs for Robust and Secure Lung Disease Diagnosis

Determination of the exact nature of lung diseases is critical for successful treatment and superior client outcomes. This study focuses on implementing FL with CNN to enhance diagnostic accuracy while observing the privacy rights of the patient’s data. Traditional, centralized types of machine learning also have significant privacy issues because of the data-gathering process. On the other hand, the state in the southeastern United States, known as Florida (FL), permits multiple healthcare centers to collectively train a model without sharing the raw data. This paper introduced the FL-CNN model that we designed, and as evidenced by the performance indexes, the model presented much higher efficiency compared with the centralized CNN. Hence, FLCNN had a higher accuracy rate of 90%. 7% more than the centralized model, which achieved an accuracy level of 85%. 3%. Besides, the proposed FL-CNN model had an accuracy of 91: 2% and a recall of 89. Specifically, the Recall or True Positive Rate was 92%, Precision was 86%, and the F1-score was 90%. 5%, while the precision characteristic of the centralized model is 84%. 0%, recall of 83. 85%, respectively, for Mota, while achieving almost similar accuracy and F1score percentages of 96% and 95%, respectively, for each of the three levels of language proficiency for the targeted L2 writers. 7%. The proposed FL-CNN's AUC-ROC was found to be 0.95, which is greater than the value of the centralized model, which is 0. 89. Thus, it can be concluded that the FL-CNN has a superior capability to classify various classes.