Evaluation of surgical skill using machine learning with optimal wearable sensor locations.

Area of Science:

• Biomedical Engineering
• Surgical Education
• Machine Learning in Medicine

Background:

• Assessing surgical skills is crucial for recruiting new surgeons, particularly in minimally invasive procedures.
• Wearable electromyography (EMG) and accelerometer sensors offer a feasible method for evaluating surgical performance through muscle activation and movement.
• Sensor placement challenges can hinder accurate skill assessment during surgical tasks.

Purpose of the Study:

• To identify optimal muscles and machine learning features for accurate surgical skill evaluation using wearable sensors.
• To investigate the feasibility of quick skill assessment during surgical tasks.

Main Methods:

• Twenty-six surgeons (novice, intermediate, expert) participated, with twelve wireless sensors (EMG and accelerometers) placed on specific arm and forearm muscles.
• Muscle activation and movement variability profiles were analyzed using features like approximate entropy, sample entropy, and multiscale entropy.
• Machine learning classifiers (Random Forest, Support Vector Machines, Naïve Bayes) were employed to differentiate skill levels.

Main Results:

• Features related to movement complexity were critical for identifying surgeon skill levels.
• The extensor carpi ulnaris (ECU) muscle with a Random Forest Classifier achieved the highest individual accuracy (61%).
• Combined muscle features, particularly ECU with deltoid or biceps, improved classification accuracy.

Conclusions:

• Quick surgical skill evaluation is achievable using wearable sensors and machine learning.
• The extensor carpi ulnaris (ECU), deltoid, and biceps muscles provide important features for differentiating surgical expertise.
• This technology has the potential to aid in the objective assessment and recruitment of surgeons.