Development of An Automated Liquid Level Monitoring Technique for Urine Bag Based on Deep Learning

영상 딥러닝 기반 소변주머니 내부 소변량 자동 감시 기술 개발

In-hospital patients who need long-term catheterized urinary support requests repetitive flushing of urine in the urine bag, which can increase the workload of nursing staffs. In this study, a deep learning-based liquid level monitoring technique (based on EfficientDet Lite models) that can estimate the status of in-bag liquid level in real-time from the periodically photographed bag images captured using a camera installed in front of the urine bag. In experiments, the error rates between the manually-calculated in-bag liquid area and the model-extracted in-bag liquid area were 1.05 ± 1.58%, and those between the manually-calculated scale area and the model-extracted scale area were 2.57 ± 2.54%, respectively. In addition, the error rates between the actual amount of in-bag liquid and the model-calculated in-bag liquid amount were 2.28 ± 4.97%, respectively. The implemented technique could successfully estimate the time-varying amount of liquid in the urine bag. It can be applied to develop a smart flush-request alarm system for patients with long-term catheterized urinary support that can reduce the workload of nursing staffs while increasing the safety of patients.