JMIR Publications

ABSTRACT

Background:

Taiwan’s transition into a super-aged society has increased the importance of dementia and the need for scalable, non-invasive indicators of cognitive decline. Conventional cognitive screening may be limited by administration time and language dependency. AI-based analysis of affective facial responses may provide digital behavioral markers, but its feasibility in aging populations remains underexplored.

Objective:

This study examined whether AI-derived affective expression congruency and onset latency differed across adults with healthy cognition (HC), mild cognitive impairment (MCI), or dementia.

Methods:

This feasibility study enrolled 60 participants, including 20 with HC, 20 with MCI, and 20 with dementia. Participants viewed positive and negative emotion-eliciting videos and completed Mini-Mental State Examination (MMSE) recall and calculation tasks. Facial expressions during emotional stimulation and the MMSE-based cognitive-task condition were video-recorded and analyzed using a convolutional neural network–based AI system. Outcomes included affective expression congruency and onset latency. Congruency was defined as the proportion of AI-detected expressions consistent with the intended emotional or task-related category, and onset latency as the time from stimulus or task onset to a non-neutral facial expression. Group differences were analyzed using analysis of variance, chi-square tests or Fisher exact tests, Bonferroni-adjusted post hoc comparisons, analysis of covariance adjusting for age and education level, and exploratory receiver operating characteristic analyses.

Results:

Affective expression congruency decreased across cognitive performance levels. During the 14-second positive stimulus, congruent positive expressions were observed in 65% (13/20) of HC participants, 20% (4/20) of MCI participants, and 0% (0/20) of participants with dementia. During the 48-second negative stimulus, congruent negative expressions were observed in 60% (12/20) of HC participants compared with 5% (1/20) of both MCI and dementia participants. During the MMSE-based cognitive-task condition, task-congruent negative expressions were observed in 50% (10/20), 15% (3/20), and 10% (2/20) of HC, MCI, and dementia participants. Onset latency differed significantly across groups for the 14-second positive stimulus (F(2,57)=19.631, P<.001), 52-second positive stimulus (F(2,57)=4.250, P=.019), 48-second negative stimulus (F(2,57)=15.605, P<.001), and MMSE-based cognitive-task condition (F(2,57)=11.281, P<.001). After adjustment for age and education level, the group effect remained significant for all latency outcomes. Exploratory receiver operating characteristic analyses showed moderate-to-good separation for HC versus MCI and HC versus dementia comparisons but limited MCI versus dementia separability.

Conclusions:

AI-assisted analysis of affective facial responses revealed observable differences in expression congruency and onset latency across cognitive performance levels. These preliminary findings suggest that AI-derived facial behavioral metrics may serve as non-invasive adjunctive features for observational cognitive assessment. The findings should be interpreted cautiously because of the exploratory design, sample size, and absence of external validation. Clinical Trial: Not applicable.