Federated Learning and Cloud-Based Artificial Intelligence for Real-Time Diagnosis of Rare Diseases in Healthcare Systems

Rare diseases present a unique diagnostic challenge.With only a handful of identified cases and wide-ranging clinical manifestations, these diseases rarely appear in the differential diagnosis of physician decision-support systems.Consequently, they are infrequently considered at the initial visit.Lack of awareness thus leads to severe, irreversible complications, considerable impairment or even death, ultimately resulting in a loss of human life comparable to that of common diseases like breast cancer.A real-time diagnostic service for several rare diseases-stargardt disease, idiopathic pulmonary fibrosis, systemic lupus erythematosus, scleroderma, Crohn disease, and Cushing syndrome-based on federated learning and cloud artificial intelligence can help overcome the problem.The objective of the federated-learning service is to train an artificial-intelligence model at each hospital site without collecting or sharing sensitive data in a central cloud.The multi-institutional architecture is designed to produce collaborative real-time diagnoses without the large time lags associated with multisite diagnosis requested from typical cloud-based platforms.The proposed research framework guarantees that highly sensitive data from different locations remains on-site during the training process, can receive real-time predictions through the AI model of other sites, and thus supports local specialists in correctly diagnosing rare diseases.A federated approach minimizes the potential presence of low-quality data and enhances the diagnostic reliability of models used to support the decision-making process.Given the richness of medical data from different areas supplied by different medical centers, the approach is applicable across a broad range of federated-learning scenarios.