Healthcare Diagnostics Service Using Federated Learning

Healthcare systems around the world rely on powerful computational prediction tools in order to make accurate diagnostics with regard to the human body. In order to estimate the severity of lung damage post-COVID infection, healthcare providers rely on AI prediction tools to perform diagnosis. While such tools exist at a rudimentary level, there is a growing demand for more reliable and democratised systems that train models over a diverse data-set. To that end, the focus of this research paper turns to federated learning, a distributed machine learning paradigm. The system proposed consists of a central server that pools features and weights across various nodes, thereby cutting bias in the prediction models. This also achieves data decentralisation which ensures patient privacy. An end-to-end application is realised that facilitates distributed training of batch data that is visualised in real-time with the help of sockets. The application also features an inference service, classifying chest x-rays based on whether the image displays damage in case of Pneumonia.