Objective <p>To develop a non-contact monitoring system for unplanned device removal (UDR) and to preliminarily evaluate its clinical feasibility.</p> Methods <p>We developed a non-contact monitoring system for UDR featuring an interactive user interface. It operates reliably on CPU hardware, eliminating the need for GPU acceleration or cloud-based processing while supporting adjustable sensitivity and incorporating multiple mechanisms to mitigate alarm fatigue. We applied the system to neurocritical care patients, using unplanned removal of cranial drainage tubes as a representative scenario, and conducted a proof-of-concept, prospective, randomized controlled trial to determine the number of hand-to-head occurrences detected by the system vs. direct observation.</p> Results <p>The system functioned continuously throughout the study period without any interruption. Individual monitoring durations ranged from 25.7 to 88.9&#xa0;h, with a median of 45.5 (IQR: 31.0–55.8) hours. No crashes or abnormal terminations are observed. A total of 65 neurocritical care patients were enrolled, with 34 in the intervention group and 31 in the control group. The median number of hand-to-head contacts was 18.0 (IQR: 15.3–20.0) in the intervention group and 12.0 (IQR: 11.0–13.0) in the control group, with a significant difference between groups (<i>p</i> &lt; 0.001). The incidence of UDR was 2.9% in the intervention group versus 16.1% in the control group (<i>p</i> = 0.10).</p> Conclusion <p>This study preliminarily validated the clinical value of an AI-based system for monitoring UDR in neurocritical care patients. The system provides a non-contact, low-cost, and efficient monitoring solution, and demonstrates strong potential for broader clinical applications. With further optimization, it may be adapted for use with various critical devices in pediatric, respiratory, and other intensive care settings.</p> Clinical trail number <p>Not applicable.</p>