Near Real-Time Federated Machine Learning Approach Over Chest Computed Tomography for COVID-19 Diagnosis

ABSTRACT

During the COVID-19 pandemic, artificial intelligence (AI) plays a major role to detect and distinguish between several lungs diseases and diagnose COVID-19 cases accurately. This article studies the feasibility of the federated learning (FL) approach for identifying and distinguishing COVID-19 X-ray images. We trained and tested FL components by using the data sets that collect images of three different lungs conditions, COVID-19, common lungs and viral pneumonia. We develop and evaluate FL model horizontally with same parameters and compare the performance with the classic CNN model and the transfer learning approaches. We found that FL can quickly train artificial intelligence models on different devices during a pandemic, avoiding privacy leaks that may be caused by such a high resolution personal and private X-ray data. © 2022, Springer Nature Singapore Pte Ltd.