Artificial intelligence-based waste detection in robotic surgery: a pilot validation study of computer vision...

Cost containment remains critical for widespread adoption of robotic surgery, yet operating room waste of single-use supplies represents a significant source of unnecessary expenditure that is often overlooked. Traditional waste tracking relies on time-intensive direct observation, limiting scalability and consistency. We developed and validated AIMs, a computer vision-based system designed to automatically detect and quantify robotic stapler load waste on the surgical back table. This prospective validation study evaluated system accuracy in detecting robotic stapler loads during 18 consecutive robotic procedures at a high-volume robotic surgery center. The system utilizes a convolutional neural network for real-time object detection via continuous back table video recording, with system-generated logs compared against concurrent manual observation by a trained observer. Performance was assessed using sensitivity (recall), precision (positive predictive value), and false negative/positive rates at four critical timepoints: initial count, addition events, removal events, and final waste quantification. The system achieved perfect performance for initial load detection (100% sensitivity and precision) and final waste quantification (100% sensitivity and precision). Mid-procedure tracking showed 87.7% sensitivity and 100% precision for addition events, and 97.1% sensitivity with 100% precision for removal events. All detection errors were false negatives with zero false positive detections recorded. System operation did not interfere with surgical workflow or require additional personnel. This computer vision-based approach provides objective, scalable data for preference card optimization and cost reduction initiatives, contributing to enhanced cost-effectiveness of robotic surgery programs and facilitating broader patient access to advanced surgical technology.