The eye sees what the mind knows: Eye tracking and artificial intelligence to assess surgical resident level of expertise.

BACKGROUND: Surgical education continues to strive to move beyond traditional time-based competency metrics. As gaze behaviors can reflect underlying cognitive strategies, eye-tracking technology may provide insight into individual differences. Hence, we sought to characterize gaze-based phenotypes in surgical learners to examine their relationship with training level. METHODS: We conducted an eye-tracking study involving general surgery residents from postgraduate year 1 through postgraduate year 5 using screen-based eye trackers. Gaze patterns were collected on 2 30-second video clips of operations. Three core metrics were extracted: average dwell time per fixation, spatial dispersion, and scan entropy. Principal component analysis was applied for dimensionality reduction, followed by k-means clustering to identify distinct visual learning phenotypes. RESULTS: Fifty-seven residents participated in this study. Principal component analysis identified 3 components explaining 87.7% of the variance in gaze behavior, with scan entropy and spatial dispersion being primary drivers. K-means clustering revealed 3 gaze phenotypes: (1) focused-stable-systematic, characterized by long, deliberate fixations, low spatial dispersion, and systematic scan paths. This expert-like pattern was predominantly noted among postgraduate year 4-5 residents; (2) brief-stable-systematic, defined by brief fixations but similarly low dispersion and entropy, suggesting a structured but superficial approach. This pattern was common among postgraduate year 1-3 residents; and (3) brief-active-exploratory, marked by high scan entropy and broad spatial dispersion. This phenotype was exclusively composed of postgraduate year 1-2 residents. CONCLUSION: Gaze-based phenotypes capture meaningful cognitive strategies in surgical learners that are not fully explained by postgraduate year level. Identification of these visual learning profiles offers a novel framework for personalized feedback, competency assessment, and adaptive curriculum design.