Automated surgical skill assessment in RMIS training

Area of Science:

• Robotics
• Surgical Training
• Machine Learning

Background:

• Manual feedback in robot-assisted minimally invasive surgery (RMIS) is time-consuming and subjective.
• Objective skill assessment is crucial for effective surgical training.

Purpose of the Study:

• To explore holistic features from robot kinematic data for automated skill assessment.
• To propose a weighted feature fusion technique for improved score prediction.
• To develop a method for generating 'task highlights' for targeted feedback.

Main Methods:

• Utilized the JHU-ISI Gesture and Skill Assessment Working Set (JIGSAWS) dataset.
• Evaluated four holistic features: sequential motion texture (SMT), discrete Fourier transform (DFT), discrete cosine transform (DCT), and approximate entropy (ApEn).
• Implemented individual feature analysis and a proposed weighted combination technique for skill classification and score prediction.

Main Results:

• Holistic features outperformed previous Hidden Markov Model (HMM)-based methods for skill classification.
• The proposed feature fusion strategy significantly improved skill score prediction, achieving a 0.61 average Spearman correlation coefficient.
• Analysis showed task highlights correlate with specific surgical gestures.

Conclusions:

• Holistic features from robot kinematic data enable effective skill evaluation in basic RMIS tasks.
• The presented framework can facilitate real-time score feedback in RMIS training.
• This approach supports more focused training for surgical trainees.