Decentralized Intelligence for Healthcare Decision Support

Abstract

Machine learning and Artificial intelligence (ML/AI) have progressed recently, as evidenced by the number of research carried out in this area, thus providing new opportunities for healthcare decision support systems (HDSS). As the world moves towards real-time decision making, AI on the edge is poising itself to be the next frontier for processing data locally while providing instantaneous insights and taking care of evident issues like data privacy and connectivity. While ML and AI significantly boost healthcare decision-making processes in a variety of clinical scenarios. This paper examines the implementation of the techniques on edge devices. Leveraging a rich real-world healthcare dataset of 55,500 patient records obtained from Kaggle, we examine the magnitude of the improvements in latency reduction, patient privacy enhancement, and clinical workflow efficiency improvements due to edge computing. We evaluate that inference times achieved by the optimized Random Forest models deployed at the edge are orders of magnitude smaller than those achieved by the networked alternatives in the cloud and that while the best predictive accuracy achieved was 92.3%, the edge AI models provide comparable results with only minor losses in predictive capacity in exchange for a significant gain in performance, indicating the ability for edge-based models to act as a future path in healthcare decision-support systems.