Force Profile as Surgeon-Specific Signature.

Area of Science:

• Neurosurgery
• Surgical Robotics
• Machine Learning

Background:

• Operating room surgeon performance is understudied.
• Sensorized surgical devices and deep learning offer new quantitative insights.
• Previous work identified surgeon force profiles, skill, and tasks, but not individual surgeon identification.

Purpose of the Study:

• To investigate if a surgeon's unique force profile can identify them.
• To explore deep learning architectures for surgeon identification based on tool-tissue forces.
• To analyze force data patterns for differentiating surgical techniques.

Main Methods:

• Utilized the SmartForceps System for time-series tool-tissue force data.
• Investigated multiple neural network architectures (ResNet, XGBoost, CNNs, LSTMs).
• Data split into training (80%), validation (10%), and testing (20%) sets; addressed data imbalance via subsampling.

Main Results:

• Best model (time-series ResNet + XGBoost) achieved AUC 0.97, F1-score 0.82.
• Convolutional neural networks (CNNs) outperformed LSTMs in performance and speed.
• An ensemble model identified expert surgeons with 83.8% accuracy on validation data.

Conclusions:

• Surgeons exhibit unique force profiles identifiable by deep learning.
• Developed models can provide surgeons with quantitative feedback and track skill progression.
• Surgeon identification facilitates correlating surgical performance with patient outcomes.