Automatic Diagnosis of Age-Related Macular Degeneration via Federated Learning

Artificial intelligence has taken healthcare a step forward by providing quick diagnosis and treatment recommendations. The traditional machine-learning approach requires massive data to train the model for better diagnosis. But, due to policy regulations, medical data is always guarded by the barricades of the law, making data accessibility difficult for researchers. To address this issue, a data-decentralized collaborative framework known as federated learning is adopted that reaps the benefits of huge private data without aggregating it into a single common store. A pre-trained model is employed to diagnose age-related macular degeneration and performance of the proposed framework is compared with other two model architectures, namely, MobileNet and InceptionV3. To investigate the effectiveness of the proposed framework, a comparison is made with a data-centralized learning approach. Using the MobileNet model, the federated and centralized frameworks have achieved an accuracy of 95% and 92%, respectively. These findings encourage clinicians around the globe to utilize wealthy private data without violating privacy laws using federated learning to build a powerful model for classifying any disorders while maintaining data privacy.