Ranking surgical skills using an attention-enhanced Siamese network with piecewise aggregated kinematic data

Area of Science:

• Robotics in Medicine
• Surgical Training and Assessment
• Machine Learning for Healthcare

Background:

• Computerized surgical skill assessment is crucial for objective performance evaluation and expert training.
• Current methods often rely on manual annotations, which can be inconsistent, scarce, or biased.
• This leads to challenges in accurately categorizing surgical skill levels.

Purpose of the Study:

• To develop a novel approach for surgical skill assessment using a pairwise ranking task.
• To address the limitations of manual annotations in skill level categorization.
• To provide a more interpretable and objective method for evaluating surgical performance.

Main Methods:

• The study frames surgical skill assessment as a pairwise comparison of two actions.
• A model using an attention-enhanced Siamese Long Short-Term Memory Network was developed.
• Kinematic motion data from robot-assisted surgery sensors, processed by piecewise aggregate approximation, were utilized.

Main Results:

• The proposed model demonstrated higher accuracy in pairwise ranking compared to existing models.
• It outperformed existing regression models in their respective experimental setups.
• The model proved accurate for individual progress monitoring on a new dataset.

Conclusions:

• The relative assessment approach mitigates issues associated with inconsistent skill level annotations.
• This method offers a more interpretable means for objective surgical skill assessment.
• The model facilitates monitoring individual skill development over time through comparative analysis.