Introduction: As image interpretation is a visual perception task, it is affected by context and task structure. The non-contrast CT Alberta Stroke Program Early CT Score (ASPECTS) is considered an objective radiologic measure of the extent of ischemic change in cases of acute ischemic stroke. We hypothesize that variability in ASPECTS reading is because readers are susceptible to interplay between higher-order, top-down information (clinical background or other imaging) and bottom-up sensory information, and that experts are more resistant to being swayed by these factors than novices.
Methods: We tested the effect of three top-down background-information conditions (no information; clinical information with affected side; clinical information plus CT angiography (CTA)) and three environment/task conditions (daylight, core lab, time pressure) on the inter-rater reliability (IRR) of NCCT ASPECTS. Three raters (trainee, fellow, neuro-radiologist) independently scored 150 NCCT scans from acute ischemic stroke patients, with 50 allocated to each environmental task. This was repeated over three sessions, with one for each background-information condition. ASPECTS on CT perfusion Tmax thresholded scans served as ground truth.
Results: IRR improved when clinical information and CTA were provided (ICC 0.55 - clinical + CTA, 0.41 - clinical, 0.10 - no information). In some situations (NIHSS <5, onset-to-imaging times >180 min), IRR was greater with clinical information but without CTA. Daylight and time pressure had greater IRR than core lab (ICC 0.59 and 0.7 vs. 0.34, respectively). In general, the trainee assigned lower ASPECTS relative to the fellow and expert. The expert showed greatest concordance with CT perfusion ASPECTS (ICC 0.69), and the trainee the least (ICC 0.47).
Conclusion: The cognitive framework that we propose, which includes top-down and bottom-up constraints on visual perception, can explain variability in ASPECTS interpretation on NCCT.