Federated Learning - a Novel Approach for Predicting Diseases in Unprecented Areas

According to the World Health Organization (WHO), approximately 75 percent of deaths in rural areas are attributed to delayed disease diagnosis. Some diseases exhibit unforeseeable symptoms, leading to life-threatening conditions. Moreover, emerging and rare diseases with recognizable mani-festations in advanced stages pose challenges due to physicians' limited knowledge. In this context, federated learning emerges as a privacy-conscious machine learning approach, ideally suited for smart healthcare applications. It facilitates collaboration among multiple hospitals to conduct training without the need to share raw data, thereby preserving sensitive information. The proposed solution showcases the feasibility of using federated learning to predict diseases based on frontal chest X-rays. The iterative training process of federated learning, occurring at predefined intervals, significantly enhances efficiency, allowing doctors to include specific symptoms for early predictions of novel diseases. A comprehensive evaluation using RESNET-50 on frontal chest X-rays demonstrated that the federated learning approach improves the efficiency of disease detection compared to a normal model by atleast 2 percent.