Intra-operative estimation of surgical progress

Abstract

Driven by rising health care costs due to factors including advancing technology and an aging population, cost-effectiveness has become an increasingly important aspect of care delivery, with the operating room (OR) being a specific area of interest. By means of a surgical process model (SPM), OR systems can gain an understanding of clinical context and surgical workflow, hereby generating ample opportunities to improve OR logistics and surgical care. Applications of SPM's include intra-operative end-time predictions, improved surgical training and assessment, computer-aided surgery and increased autonomy in robotic surgery. This thesis evaluates the use of an SPM for intra-operative recognition of surgical phases in laparoscopic hysterectomy cases (n=40), based on manually annotated instrument usage data. Using a Random Forest model, an out-of-sample accuracy of 77% is achieved. The phase-recognition model is shown to predict surgical end-times with a mean absolute error of 16 minutes and is additionally found useful in the task of surgical phase extraction. Further research should specifically be aimed at replicating the promising simulated findings of this thesis in-vivo, using intra-operative sensor recordings in the OR.