Online Federated Deep Probabilistic Learning-Based Smart Healthcare on Multi-Cloud Systems

Abstract

Healthcare management is one of the research topics that is being addressed intensively during the last decade. It is one of the corner stones that smart cities are built upon. This article addresses the problem of patient diagnoses. The article proposes an unsupervised deep Bayesian neural network architecture that learns the probabilities of different diagnoses based on factors like symptoms, patient history, and patient personal information. The outcome of the work is to optimize the healthcare process in a way to find the balance between minimizing test costs and maximizing the Precision of the diagnosis process. Every medical institution like hospitals and clinics provide data of symptoms, tests orders and the effectiveness of those tests in finding the actual diagnosis. An overall model is built based on federated learning to optimize the selection process of tests based on observed symptoms. Experiments show that the proposed model succeeds in improving the process with respect to time, cost, and Precision. The obtained results show that the more clouds involved in the experiment the less the error is and the faster the learning curve is.