Assessing Medical Training Skills via Eye and Head Movements

The dataset consists of multimodal eye and head movement recordings collected from 24 medical practitioners during simulated baby delivery training sessions. Each participant had 2 training sessions during which they wore eye-tracking glasses (Tobii Pro Glasses2) that captured detailed gaze and head movement data in real time. 

Each participant’s data is labeled based on their training level (trained vs. untrained), 14 skill scores given by expert doctor and expert based time segmentation of video indicating different stages of breach delivery process. 

This enables various supervised learning experiments. The dataset supports analysis of visual attention and motor coordination, and has been used to extract features that differentiate practitioner expertise. Performance metrics using these features include an F1 score of 0.85 (head-related features) and 0.77 (pupil-related features), and AUC values of 0.86 and 0.85 respectively. Trining scripts and models are available in separate repository (see paper for link and more details).

This dataset enables development of computational models for implicit skill assessment in clinical education and provides a foundation for integrating wearable sensing technologies into simulation-based training.