JournalVolume 25, No.2, Special Issue on Artificial Intelligence and Computational Methods for the Medical Cyber-Physical Systems
Learning-Based Decision Support System for Critical Care in Clinical Cyber-Physical Environments (pp. 245-269) S. Mansour1, M. Almofarreh2 , M.M. Asiri 3, E. Alyounis4, M.N. Alkhomsan5, A. Yafoz6, A.A. Alhashmi7,*
https://doi.org/10.30546/1683-6154.25.2.2026.2451Department of Radiological Sciences, College of Health and Rehabilitation Sciences, Princess Nourah bint Abdulrahman University,
P.O. Box 84428, Riyadh 11671, Saudi Arabia
2Department of Computer Science, College of Computer and Information Sciences, King Saud University, P.O.Box 51178,
Riyadh 11543, Saudi Arabia
3Department of Computer Science, Applied College at Mahayil, King Khalid University, Saudi Arabia
4Department of Health Information Management and Technology, College of Applied Medical Science, King Faisal University,
Al Ahsa 31982, Saudi Arabia
5Department of Software Engineering, College of Computer and Information Sciences, Jouf University, Sakaka 72341, Saudi Arabia
6Department of Information Systems, Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah,
Saudi Arabia
7Department of Computer Science, College of Science, Northern Border University, Arar, Saudi Arabia
∗Corresponding author e-mail: Asma.alhashmi@nbu.edu.sa
Abstract. This paper proposes a decision support framework that incorporates a safetyrestricted offline Deep Reinforcement Learning (DRL) agent with a post-hoc Explainable AI (XAI) module to a Clinical Cyber-Physical System (CCPS). The DRA agent is trained on past ICU data to acquire an optimal policy on the application of mechanical ventilation, introduction of vasopressors and similar interventions. The XAI layer uses Shapley Additive Explanations (SHAP) to produce feature-based explanations of the recommendations in the model that are clinically meaningful to increase transparency. This system is tested with offline policy testing in a simulated ICU based on real patient data. Outcomes show better prediction of patient outcome and safety of treatment with respect to baseline approaches. Combining interpretability and safety-conscious learning, the framework facilitates sound human-AI collaboration and opens the door to the responsible use of AI in the intensive care units.
Keywords: Explainable Artificial Intelligence (XAI), Deep Reinforcement Learning, Clinical Decision Support, Critical Care, Cyber-Physical Systems, Interpretability, Sepsis Management, Mechanical Ventilation.
AMS Subject Classification: 68T05, 68T07, 93E20, 92C50, 62P10.
|
|