A Novel Closed-Loop Deep Learning-Based Smart Infusion Rate Monitoring Technique for Safe Intravenous Medication Administration

Abstract

Medication administration via an intravenous (IV) catheter is a widely used medical procedure; however, incidents related to IV administration have consistently been reported to regulatory agencies. To improve patient safety during these incidents, it is essential to enhance the monitoring of IV administration. This study proposes an artificial intelligence-based technique for real-time infusion rate (IR) monitoring that automates several processes: the initial setup for image monitoring, the monitoring of variations in in-bag liquid volume and infusion pump settings, the determination of the relevance between infusion status and in-bag liquid residue, and the alarm processes for early detection of IV administration-related emergencies, using deep learning models and mathematical estimations. The experimental results demonstrate that the average error rate for estimating in-bag liquid volume is less than 5.00%, the average mismatch between the bag-extracted IR and the pump-extracted IR is under 3.00%, the average error rate for the “time-to-bag-empty” alarm is below 6.00%, and the error rate for the “low in-bag liquid volume” alarm is under 10.00%. The accuracy of detecting abnormal IR settings of the infusion pump was 100%. Based on these results, we conclude that the proposed artificial intelligence-based smart IR status monitoring technique shows promise as a prototype for autonomous IV administration monitoring with minimal human intervention, serving as a foundational step toward clinically deployable solutions in future healthcare settings.