Pneumonia Detection from X-Ray Images Using FederatedLearning- an Unsupervised Learning Approach

Abstract

Pneumonia continues to be a major worldwide health issue, particularly within a dynamicenvironment characterized by factors like climate change, the challenge of antibiotic resistance, and theconstant evolution of disease-causing agents. Despite efforts such as immunizations and maintaining hygiene,pneumonia continues to result in substantial illness and, in severe cases, fatalities. Doctors often use X-raysto find pneumonia because they show lung problems well. Utilizing machine learning or deep learning is a valuable method in aiding radiologists in examining the extensive collection of chest X-ray images.However, these approaches rely on extensive datasets for training, which necessitates centralizing the data.Yet, due to regulations safeguarding medical data privacy, gathering and sharing patient information on acentralized server is frequently unfeasible. Another issue within healthcare systems involves the accessibilityof labeled data. This study proposes a solution to address these obstacles. It involves the implementationof an unsupervised learning model, which was trained on decentralized datasets using the Federated Learningtechnique. Three healthcare institutions participated in the training process of this model. The assessmentof this model indicated that the Federated Learning approach produces results that are on par with theperformance of models based on centralized learning. These findings suggest that medical institutions shouldadopt collaborative strategies and leverage their diverse private data to develop such models efficiently.