Detecting Task Difficulty of Learners in Colonoscopy: Evidence from Eye-Tracking

Area of Science:

• Medical Simulation
• Human Performance Monitoring
• Artificial Intelligence in Healthcare

Background:

• Physician training requires extensive practice and expert feedback for skill development.
• Personalized feedback is time-consuming and prone to bias.
• Eye-tracking offers a potential objective measure of performance and difficulty.

Purpose of the Study:

• To investigate the utility of eye-tracking data for detecting moments of navigation loss (MNL) during simulated colonoscopy.
• To develop and evaluate deep learning models for automated identification of MNLs.
• To establish a foundation for an AI-driven healthcare training and education system.

Main Methods:

• Trainees' eye movements were recorded during simulated colonoscopy procedures.
• Deep convolutional generative adversarial networks (DCGANs) were used to synthesize eye-tracking data.
• Long Short-Term Memory (LSTM) networks were employed to classify MNLs using real and synthesized eye-tracking data.
• Deep learning model performance was benchmarked against expert annotations of colonoscopy videos.

Main Results:

• Deep learning models successfully detected moments of navigation loss (MNL) in simulated colonoscopy.
• The optimal classification performance was achieved by combining real human eye data with 1000 synthesized data points.
• The best model achieved high accuracy (91.80%), sensitivity (90.91%), and specificity (94.12%) in identifying MNLs.
• The study demonstrated the feasibility of using AI to analyze eye-tracking data for performance assessment.

Conclusions:

• Eye-tracking combined with deep learning provides an objective method for detecting task difficulty in medical training.
• AI-driven analysis of eye movements can enhance the efficiency and reduce bias in performance feedback.
• This research lays the groundwork for developing advanced simulation-based education systems for healthcare professionals.