Eye Movements of Radiologists Reflect Expertise in CT Study Interpretation: A Potential Tool to Measure Resident Development

Purpose: To establish potential markers of visual expertise in eye movement (EM) patterns of early residents, advanced residents, and specialists who interpret abdominal computed tomography (CT) studies. Materials and Methods: The institutional review board approved use of anonymized CT studies as research materials and to obtain anonymized eye-tracking data from volunteers. Participants gave written informed consent. Early residents (n = 15), advanced residents (n = 14), and specialists (n = 12) viewed 26 abdominal CT studies as a sequence of images at either 3 or 5 frames per second while EMs were recorded. Data were analyzed by using linear mixed-effects models. Results: Early residents' detection rate decreased with working hours (odds ratio, 0.81; 95% confidence interval [CI]: 0.73, 0.91; P = .001). They detected less of the low visual contrast (but not of the high visual contrast) lesions (45% [13 of 29]) than did specialists (62% [18 of 29]) (odds ratio, 0.39; 95% CI: 0.25, 0.61; P < .001) or advanced residents (56% [16 of 29]) (odds ratio, 0.55; 95% CI: 0.33, 0.93; P = .024). Specialists and advanced residents had longer fixation durations at 5 than at 3 frames per second (specialists: beta = .01; 95% CI: .004, .026; P = .008; advanced residents: beta = .04; 95% CI: .03, .05; P < .001). In the presence of lesions, saccade lengths of specialists shortened more than those of advanced (beta = .02; 95% CI: .007, .04; P = .003) and of early residents (beta = .02; 95% CI:.008, 0.04; P = .003). Irrespective of expertise, high detection rate correlated with greater reduction of saccade length in the presence of lesions (beta = 2.10; 95% CI: 2.16, 2.04; P = .002) and greater increase at higher presentation speed (beta = .11; 95% CI: .04, .17; P = .001). Conclusion: Expertise in CT reading is characterized by greater adaptivity in EM patterns in response to the demands of the task and environment. (C) RSNA, 2016