Automatic Control System for Precise Intravenous Therapy Using Computer Vision Based on Deep Learning

• Published in: IEEE access Vol. 11; pp. 121870 - 121881
• Main Authors: Seol, Jaehwang, Lee, Sangyun, Park, Jeongyun, Kim, Kwang Gi
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Automatic control systems; Browsers; Computer vision; computer vision-based on deep learning; Deep learning; Electron tubes; Feedback; feedback control; Fluid injection; Fluids; infusion flow rate; Internet of Things; Intravenous therapy; Machine learning; Medical treatment; Monitoring; Monitors; Peristaltic pumps; Prototypes; Pumps; Remote monitoring; Side effects; Therapy; vision monitoring; Vision sensors
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2023.3328568

Intravenous (IV) therapy provides a rapid therapeutic effect. However, errors in infusion rate can lead to over-dosing and under-dosing, potentially causing severe side effects for patients. To prevent incidents, medical staff visually monitors medication injection information (MIF) such as current flow rates (CFR) and injection volume. In this study, we propose an internet of things (IoT) system that automatically controls the CFR and remotely monitors the MIF. First, a peristaltic pump is designed to infuse parenteral fluids based on the "drop-by-drop" phenomenon. Second, a computer vision-based on deep learning algorithm provides real-time video and counts the fluid dropping to derive the MIF. Finally, the CFR is automatically applied as a feedback signal to regulate the infusion cycle of the peristaltic pump. After embedding all systems in our prototype, we evaluate the system performance according to IV therapy protocols used in clinical practice. In our experiments, the mean accuracy of the fluid injection using the peristaltic pump was 99.23% and the dropping count was the highest at 98.25%. Furthermore, the average accuracy of the CFR using the feedback system was 99.3%. Based on our results, we confirmed that automated control of infusion rate is possible in IV therapy and computer vision-based on deep learning can be utilized for feedback sensors and monitoring system. Therefore, we believe that our method is the practical solution that can increase patient safety and work efficiency of clinical staffs.