Optimizing Lung Disease Detection through Federated Learning and Convolutional Neural Networks

This paper presents a new approach to screening lung diseases by combining CNN with FL in a multi-centre setting. Thus, this methodology helps overcome the critical privacy concerns of centralized data processing. FL-CNN helps train many healthcare centres without transferring raw patient data, which is a benefit in terms of privacy and security of patient information. The model was tested on many sets and performed exceptionally well on lung diseases. For the FL-CNN model, the identified accuracy rate was 0. 92 and a working precision of 0. 91; a recall of 0. 90, a precision of 97% and an F1-score of 0. 91, which is higher than that of the archetypal centralized arrangement. The traditional model demonstrated 0. 85, precision of 0. 84, recall of 0. part B accuracy of 83, and F1-score of 0. 83. For the FL-CNN model, the AUC-ROC was None. 95, while that of the centralized model was 0. 88. From this, it would be significant to deduce that the FL-CNN model provided a better diagnosis than the centralized model. The heat map generated from the confusion matrix also ensures the findings; most predictions show the actual classes. Additionally, the accuracies & losses of training and validation depict that the FL-CNN model always provided a great result, and only a slight overfitting was observed during the training phase. Comparing the bar graphs, there are several performances where the FL-CNN model stands out.