MD Student Cumming School of Medicine, Project neuroArm, University of Calgary
Introduction: Does each surgeon operate with a unique technique? Automated surgeon identification can identify characteristics of individual surgeons, providing evidence for a surgeon’s unique “force profile” and, through such data, an opportunity for machine learning-driven surgical education. This study aimed to identify surgeons using their tool-tissue force profile from sensorized bipolar forceps during neurosurgery.
Methods: Three surgeons used the “SmartForceps System,” a force-sensing bipolar forceps developed to quantify surgical tool-tissue forces, during surgical resection of brain tumors on 50 adult patients between October 2021 and June 2022. The surgeons included an attending neurosurgeon with 30+ years of experience (n=684 tasks) and two PGY 6 resident surgeons (n=656 and 784 tasks, respectively). The recorded time-series force data was used to train a long short-term memory (LSTM) deep neural network to identify the surgeon who performed each surgical task. The input used to train the machine learning model included 5-second force segments.
Results: The LSTM model achieved an overall accuracy of 74%, macro-average of 74%, weighted average of 74%, one-vs-one receiver operating characteristic (ROC) area under the curve (AUC) of 0.89, and one-vs-rest AUC of 0.89. The model had an F1-score of 0.67 for the attending neurosurgeon, 0.85 for the surgical fellow, and 0.70 for the PGY 6 surgeon.
Conclusion : Our LSTM model accurately identified individual surgeons based on their force profile with 74% accuracy and an AUC of 0.89. The results suggest that surgeons have a unique force profile, a characteristic amenable to identification using machine learning. Based on tool-tissue forces alone, the notable finding may characterize one’s surgical technique, offering an automated platform to identify the nuances of surgery and surgeons by their skill level. Future work aims to validate the model with more surgeons across various skill levels to describe features that differentiate individual surgeons.