Automated estimation of drain output in postoperative patients using deep learning on clinical images

Postoperative drains are essential components of care in general surgery and intensive care units, where accurate monitoring of drain output is critical for detecting complications such as hemorrhage, anastomotic leakage, or infection. Despite its importance, output measurement is still performed manually, which is time-consuming, exposes staff to biohazardous fluids, and is prone to documentation errors. In this study, we present a deep learning based automated system for estimating postoperative drain output from clinical images of Jackson-Pratt drains and drainage bags collected under real hospital conditions. The dataset includes a wide range of effluents, such as blood, gastric content, and serous fluid, representing the visual variability encountered in daily clinical practice. The proposed pipeline combines object detection and semantic segmentation to localize drains and mark out fluid boundaries, enabling precise and contact-free volume estimation of the fluids inside the drainage bags. Our system achieved high segmentation accuracy, with Intersection over Union (IoU) scores of approximately 0.99 and clinically acceptable mean absolute errors (5.1 mL for Jackson-Pratt drains and 43.6 mL for drainage bags). Robustness analyses demonstrated consistent performance across varying lighting conditions and viewing angles commonly encountered at the bedside. This work introduces a novel approach for automated drain output measurement validated on real patient data. By reducing staff workload and minimizing exposure to potentially infectious fluids, our system has the potential to improve both patient safety and occupational safety in surgical and intensive care unit (ICU) workflows, and also supports early detection of postoperative complications in surgical and intensive care settings. Our data and code are available at https://zenodo.org/records/17599824 .