M. J. Kleiman, J. E. Galvin
Early detection of mild cognitive impairment (MCI) can be reliably achieved using measures of executive functions, however executive compensation can mask these changes in the earliest stages. Gaze behavior, specifically during visual search, has been hypothesized to measure executive functions and attention within MCI, but prior study has been inconclusive. This study characterized naturalistic visual search in MCI across distractor conditions and evaluated multivariate gaze-based classification of MCI from cognitively normal aging. In a cross-sectional, propensity-matched cohort of 62 CN and 62 MCI participants, participants completed a naturalistic scene-viewing search task under five distractor conditions (none, dissimilar, similar, background, and unexpected). Significant interactions were observed including in target dwell ratio and scanpath entropy, with trial-to-trial variability in area-of-interest coverage found to be higher in MCI. Repeated 10x5-fold cross-validation with feature selection discriminated CN from MCI with an area under the curve of 0.732 (95% CI [0.645, 0.818]); an expert-curated 14-feature panel reached 0.898 (95% CI [0.840, 0.949]). MCI-associated search disruption is displayed across qualitatively distinct distractor contexts, including in the absence of distractors, suggesting that naturalistic visual search merits evaluation as a candidate early-detection measure.