Predicting surgical skill from the first N seconds of a task: value over task time using the isogony principle

Area of Science:

• Robotics in Surgery
• Surgical Skill Assessment
• Machine Learning in Medicine

Background:

• Traditional surgical skill evaluations often use broad metrics, overlooking critical procedural nuances.
• Robot-assisted surgery offers rich data but requires advanced methods for skill analysis.

Purpose of the Study:

• To develop and evaluate temporal clustering methods for segmenting robot-assisted surgical procedures into clinically relevant tasks.
• To apply machine learning algorithms to these segmented tasks for objective surgeon skill level classification.

Main Methods:

• Utilized features derived from the isogony principle to train four machine learning algorithms.
• Employed K-fold and leave-one-user-out cross-validation for model accuracy assessment.
• Used dry-lab laparoscopic data from common training exercises.

Main Results:

• Achieved median accuracies of 85% for 2-class and 63% for 3-class skill level classification.
• Demonstrated that 2-class models can predict skill level with 90% of best-case accuracy using only 8 seconds of data.
• Successfully discerned novice and expert skill levels in unobserved trials.

Conclusions:

• State vector machines trained with isogony principle parameters can effectively differentiate novice and expert surgical skills.
• Accurate skill classification is achievable within 10 seconds of task initiation, significantly improving evaluation efficiency.