Reimagining Surgical Skill Assessment: From Subjective Evaluation to Real-Time AI Coaching

Deconstructing Surgery into Surgemes

The research focuses on breaking down complex surgical tasks into their most basic units of action, called "surgemes." By analyzing trials from the JIGSAWS dataset, the team maps the kinematic signatures of both success and failure.

Pinpointing the "Smoking Gun" of Failure

The study found technical failures are predictable and gesture-specific. A key finding was that Right Rotational Velocity (with a KL Divergence of ~8.0) was the critical kinematic marker distinguishing a clean needle drive from a botched one.

Transition to a Real-Time "Black Box"

For patients, this represents a step toward an operating room "black box" that doesn't just record events but actively warns against errors. The aim is to provide immediate feedback before a mistake becomes a medical complication.

The Link Between Time and Error

In suturing tasks, the frequency of executional errors—physical slips like dropping a needle—correlated strongly with the time spent on the task (r=0.837, p < 0.001). When a surgeon struggled, the increased time revealed a clear proficiency gap.

Identifying Vulnerable Surgical Gestures

The research highlights specific "vulnerable" gestures where errors are most likely:

• In suturing, gesture G6 (pulling the suture) had the highest error rate at 74%.
• Gesture G3 (pushing the needle through tissue) followed with a 51% error rate.
• A common novice error is failing to move along the curve of a needle, which causes unnecessary tissue trauma.

The Expert's "Procedural Error"

Surprisingly, breaking procedure isn't always bad. Experts frequently deviated from the standard sequence to "re-arrange" their tools—a mark of surgical style that yields high performance. This contrasts with novices, who made random errors even while following the "grammar" of the task.

Dataset & Modeling Constraints

• Sample Size: A small sample led to the exclusion of knot-tying data from the analysis.
• Camera View: The fixed-camera setup of JIGSAWS resulted in a high number of "out of view" errors.
• Model Assumption: The current model assumes motion data follows a Gaussian distribution, which may oversimplify the complex, non-linear movements of a surgeon.