Characteristic Facial Expressions During Visual Fixation in Japanese Older Adults With Lower Cognitive Screening...

Early identification of older adults with lower cognitive screening performance may facilitate appropriate clinical evaluation and intervention to delay the progress of cognitive decline and onset of dementia. Because facial information can be acquired noninvasively and unobtrusively during everyday life, it has attracted increasing attention as a potential digital biomarker. However, it remains unclear which facial regions are associated with cognitive decline under a neutral facial condition. Furthermore, it is unknown whether these facial characteristics primarily reflect static facial appearance or facial movements.

This study aimed to identify facial action unit (AU) patterns associated with lower cognitive screening performance during a fixation-cross viewing condition that is less susceptible to the influence of language and emotional elicitation.

Older adults were classified into either a lower cognitive screening performance group (Montreal Cognitive Assessment Japanese version [MoCA-J] score ≤25; n=31) or a healthy control group (MoCA-J score ≥26; n=14). Facial videos were recorded while participants viewed a fixation cross for 1 minute. OpenFace 2.0 was used to extract the intensities of 17 AUs and the presence of 18 AUs. Video analyses were first performed to identify AU characteristics associated with cognitive screening performance. For AU intensity, frame-level mixed-effects models with participant-specific random intercepts and a first-order autoregressive (AR(1)) correlation structure were used, adjusting for age, Geriatric Depression Scale score, and sex, and participant-level sensitivity analyses were conducted to assess the consistency of the frame-level findings. For AU presence, participant-level analyses were performed because the frame-level models did not yield reliable parameter estimates. Subsequently, single-frame analyses were performed to examine whether the AU characteristics identified in the video analyses could be explained solely by static facial appearance. As this study was exploratory and hypothesis-generating, P values were not adjusted for multiple comparisons.

For AU intensity, frame-level video analyses showed that the lower cognitive screening performance group had a higher probability of zero intensity for AU10 (upper lip raiser), with participant-level analysis based on the proportion of zero-intensity frames yielding consistent results. In contrast, single-frame analyses of AU intensity showed no significant group difference for AU10, suggesting that the observed difference may reflect facial movements rather than static facial appearance. For AU presence, video analyses showed no significant group differences.

Older adults with lower cognitive screening performance exhibited distinct AU patterns related to AU10 (upper lip raiser) during fixation-cross viewing, which may reflect subtle differences in facial movements around the mouth. These exploratory findings identify candidate facial biomarkers associated with lower cognitive screening performance and warrant validation in larger, balanced, independent samples.