Smart Glasses–Based Eye-Tracking System for Cognitive Impairment Screening in Clinic and Community Adults: A Cross-Sectional Study (Preprint)
BACKGROUND Brief screening tests such as the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) yield summary scores that offer limited insight into how tasks are performed and may miss subtle early decline in community-dwelling adults. Because health status differs between community and clinic populations, screening tools adaptable to each setting are needed. Smart glasses with eye tracking capture reading behavior alongside task performance, but their relationships with screening scores, systemic health, and brain structure remain unclear. OBJECTIVE We examined associations of smart-glasses measures with MMSE and MoCA, developed setting-specific scores classifying low screening scores, and explored their links with laboratory and brain computed tomography (CT) measures. METHODS In this cross-sectional study (2024–2026), adults recruited from outpatient clinics and the community completed the MMSE, MoCA, physical measurements, and an approximately 10-minute smart-glasses assessment comprising a reading task and six cognitive tasks, which generated 30 parameters (8 reading eye-tracking measures and 22 cognitive task measures). Spearman correlations were calculated, and cohort-specific logistic regression models combining age with selected predictors classified MMSE ≤23 and MoCA ≤25; their predicted probabilities served as Cognitive Impairment Prediction Scores (CIPS). Associations of CIPS with laboratory abnormalities and CT-derived brain structure parameters were explored. RESULTS Among 911 participants, the smart-glasses cognitive total score correlated with MMSE and MoCA (Spearman ρ=0.43 and 0.39), whereas longer reading time and higher fixation count during eye-tracked reading correlated with lower MMSE and MoCA scores (ρ=−0.33 and −0.39; ρ=−0.25 and −0.35, respectively). Within each cohort, age and up to 5 or 10 top-ranked predictors from univariable screening were entered into logistic regression models, and the predicted probabilities served as CIPS. CIPS models achieved higher AUCs than age alone in both cohorts (MoCA ≤25: 0.813–0.953 vs 0.685–0.869; MMSE ≤23: 0.771–0.948 vs 0.677–0.827). In the clinic cohort, higher CIPS was associated with eGFR <60 mL/min/1.73 m² (OR per SD 1.72–1.90) and anemia (hemoglobin <12 g/dL; OR 1.56–1.77), whereas in the community cohort, higher CIPS-MoCA correlated with lower albumin (ρ=−0.46). Among participants with brain CT, CIPS correlated with larger ventricular volume (ρ=0.30–0.43 vs −0.18 and −0.23 for MMSE and MoCA) and smaller parietal lobe and lentiform nucleus volumes (ρ=−0.37 to −0.32). CONCLUSIONS CIPS, which combined age with reading and cognitive task measures generated by the smart-glasses eye-tracking system, classified low MMSE and MoCA scores in both clinic and community settings. Exploratory associations with laboratory and brain CT measures suggest that these setting-specific, multicomponent scores may also reflect broader health status. This approach may support a more comprehensive evaluation of cognitive health.
Authors
- Ying‐Ying Yang (ORCID: https://orcid.org/0000-0002-7719-0397)
- Hsiao‐Chin Shen (ORCID: https://orcid.org/0000-0001-8637-2937)
- Hsiao‐Yun Yeh (ORCID: https://orcid.org/0009-0002-6592-0097)
- Chung-Pin Li
- Shuu-Jiun Wang
- Chih-Hsun Wu
- Wan-Ting Chu
- Ming-Chih Hou
- Yen-Hsun Huang
- Kuang-Chen Yen
- Jen-Feng Liang
- Shinn-Jang Hwang
- Chia-Yu Tsai
- Wui-Chiang Lee
- Yu-Wei Lee
- Yan-Ru Chen
- Chao-En Li
- Meng-Ju Tsai
- Chien-Wei Su
Publication Details
- Published
- 2026-11-20
- DOI
- https://doi.org/10.2196/preprints.117449
- Primary Topic
- Dementia and Cognitive Impairment Research
- Type
- preprint