Alarm management of ventilator using machine learning techniques

Abstract

Digital transformation in healthcare has ushered in an era of interconnected, data-driven clinical environments, enabling continuous monitoring and analysis of patient status. As care complexity increases, smart decision-support systems have emerged to assist medical staff in interpreting vast streams of physiological data, addressing the growing need for timely and accurate insights at the point of care. Alarm fatigue poses a huge problem in the healthcare sector, which critically undermines patient safety by desensitizing clinicians to life-threatening alarms, increasing the risk of missed or delayed responses. Its pervasive cognitive overload and workflow disruptions also contribute to clinician burnout and medical errors. This work emphasizes the critical role of Alarm Management Systems within Intensive Care Units, where alarm overload and false alarms contribute to clinician fatigue and potential safety risks. The novelty of this project lies in the fact that no similar solution exists, we generated a dataset by simulating patient conditions and capturing sensor data for several disease states both with and without induced system or patient faults.