An accessible and efficient mobile eye-tracking application for community-based cognitive impairment screening in China

An accessible and efficient mobile eye-tracking application for community-based cognitive impairment screening in China | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article An accessible and efficient mobile eye-tracking application for community-based cognitive impairment screening in China Mingxia Wei, Jincheng Li, Tongyao You, Yu Yu, Jiaying Lu, Suzhen Liang, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7141899/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 17 Nov, 2025 Read the published version in Alzheimer's Research & Therapy → Version 1 posted 11 You are reading this latest preprint version Abstract Background Cognitive impairment (CI) screening remains critically inaccessible in resource-limited Chinese communities. Methods We developed a tablet-based Mobile Eye-Tracking Application (m-ETA) using a three-step approach. A dementia discrimination model based on six oculometric features was trained in a hospital cohort (N = 204) and validated for biological relevance with Alzheimer's biomarkers (N = 101). Generalizability and accuracy were further assessed in a community cohort (N = 433) and two real-world populations (N = 2,685). Results m-ETA achieved high diagnostic accuracy for dementia (AUC = 0.99). The oculometric features were significantly associated with cognitive performance, brain atrophy, and tau deposition (all P < 0.05). m-ETA accurately detected CI (AUC = 0.80), with excellent negative predictive value for ruling out CI, and identified individuals with lower cognition performance across diverse communities. Conclusions m-ETA offers a low-cost, non-invasive, and efficient tool for large-scale CI screening, particularly suited to underserved and low-literacy communities in China. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction With an aging population, dementia has become one of the leading causes of long-term disability and mortality worldwide, having a substantial burden on global health care systems and socioeconomic improvements. More than 55 million people worldwide currently living with dementia, a figure expected to nearly triple to 139 million by 2050. [ 1 ] Notably, low- and middle-income countries (LMICs) shoulder a disproportionate burden of dementia, accounting for two-thirds of global cases, while China alone contributes nearly 25% of the worldwide dementia prevalence. [ 2 , 3 ] Cognitive impairment (CI) is a precursor of dementia, during which early detection and prompt treatment can slow the progression or even delay the onset of dementia. However, detection of CI remains challenging in China. Currently, community‑based screening still depends on neuropsychological tests. Notwithstanding their clinical utility, these tests are labor-intensive, time-consuming, and require highly trained clinicians, greatly restricting their wider adoption. This challenge is particularly pronounced in LMICs like China, where high CI prevalence contrasts with limited resources of public health. [ 4 , 5 ] Another important issue worth considering is that these neuropsychological tests, despite being administered by trained professionals, are still susceptible to subjective bias. Additionally, educational barriers exacerbate this issue. In China, over 50% of adults aged over 60 have limited formal education (less than 6 years), a rate that exceeds 60% in rural regions. [ 6 , 7 ] These educational disparities have impeded the implementation of certain cognitive screening tools, such as the Montreal Cognitive Assessment (MoCA). [ 8 ] Therefore, there is an urgent need to develop an time-saving, objective and easy-to-use tool for large-scale CI screening in resource‑limited settings impractical. Given the rapid development of wearable devices, digital biomarkers—due to their objective and non-invasive nature—are gaining increasing popularity in screening research. [ 9 ] Accumulating evidence demonstrates that oculomotor patterns reflect distinct pathophysiological mechanisms in neurodegenerative diseases. The Alzheimer's disease continuum exhibits significantly prolonged saccadic latency, reflecting tau pathology-induced functional disconnection within the frontoparietal network. [ 10 ] In addition, Alzheimer’s patients displayed specific abnormalities in some eye-tracking tests, including elevated pursuit error, reduced target fixation during visual memory tasks, and prolonged search time with increased saccadic frequency in visual search tasks. These deficits are consistent with the known vulnerability of posterior cortical networks mediating visuospatial processing and attention. [ 11 , 12 ] Emerging research further underscores the clinical validity of oculometric patterns as promising digital biomarkers, demonstrating robust utility in identifying individuals with cognitive impairment through real-time tracking capabilities. [ 13 , 14 ] However, existing eye-tracking devices remain too bulky and expensive for widespread deployment, especially not being easily accessible to rural elders from China and other LMICs. Addressing this critical gap requires the development of portable, user-friendly, and cost-effective eye-tracking devices, which are essential for achieving equitable, population-level screening of cognitive impairment. In this study, we developed and validated a user-friendly, tablet-compatible eye-tracking application, the Mobile Eye-Tracking Application (m-ETA), to detect CI in community settings. m-ETA demonstrated high accuracy in identifying individuals with impaired cognition across diverse settings, spanning both clinical and community populations. Furthermore, the study provided insights into the biological plausibility linking m-ETA to Alzheimer’s biomarkers, and positioned m-ETA as a scalable digital biomarker for CI, facilitating large-scale cognitive screening applications. Materials and methods Study design Using a three-step approach (Fig. 1 ), the discovery phase utilized a distinct discovery dataset(N = 204). 145 dementia patients and 159 normal cognition (NC) individuals were recruited from the memory clinic or medical examination center at Huashan hospital. After performing 1:1 propensity score matching (PSM) to account for age and sex, 112 case-control pairs were matched. And 20 outlier samples failed to quality control were removed. Finally, 106 dementia cases and 98 normal controls were identified. This phase aimed to identify key oculometric features associated with cognitive decline and developed the m-ETA. In the exploratory phase, we included an exploratory dataset comprising 101 Alzheimer’s disease (AD) patients to further validate the biological feasibility of m-ETA through explore its links to cognitive function scores, Magnetic Resonance (MR) -derived structural brain changes, and pathological protein deposition via Tau-positron emission tomography (PET) imaging. All patients were clinically assessed at the Department of Neurology, Huashan Hospital. In the validation phase, we externally assessed the clinically established m-ETA using an external validation cohort enrolled form Taizhou imaging study (TIS), which included 433 participants with complete eye-tracking data. This community-based elderly cohort, enriched with comprehensive cognitive data, allowed assessment of the model’s utility in screening for CI in community populations. To further validate m-ETA’s generalizability, we conducted multicenter validation across distinct demographic populations: 806 urban residents from the Hongmei community and 2,577 rural residents from the Chongming community. After quality control of eye-movement data, 660 urban and 2,025 rural participants were retained, providing balanced representation of Shanghai’s urban and rural elderly populations. Neuropsychological assessment and diagnosis All participants across the discovery dataset, exploratory dataset and external validation cohort underwent comprehensive neuropsychological evaluations. [ 15 – 20 ] General cognitive function was assessed with the Mini-Mental Status Examination (MMSE) and the Beijing version Montreal Cognitive Assessment (MoCA). Memory was assessed using the Auditory Verbal Learning Test (AVLT; for participants from the primary care dataset, exploratory dataset, or those with > 6 years of education in external validation cohort) or the Modified Fuld Object Memory Evaluation (FOME; for participants with ≤ 6 years of education in external validation cohort). Attention/executive function was measured through the Trail-Making Test A&B and the Conflicting Instructions Task (Go/No Go Task). Language abilities were assessed via Animal Fluency Test (AFT) and Boston Naming Test (BNT). Visuospatial function was examined using the Clock Drawing Test (CDT). All cognitive test scores were standardized into Z scores for comparative analysis. Participants from Chongming and Hongmei cohorts underwent rapid CI screening using the Ascertain Dementia 8 (AD-8) questionnaire, [ 21 ] a brief, validated informant-based tool developed for dementia detection in community settings. For the discovery dataset, a consensus diagnosis of dementia was established by a panel of neurologists and neuropsychologists through integrated multimodal data (medical history, neurological examination, and neuropsychological assessment), strictly adhering to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V) criteria. [ 22 ] For the participants of external validation cohort, CI was diagnosed by the same panel primarily using the MMSE criteria from a previous Chinese population-based study. [ 16 ] For the AD patients of exploratory dataset, diagnoses were based on the ATN (Amyloid/Tau/Neurodegeneration) biomarker framework proposed by the National Institute on Aging-Alzheimer's Association(NIA-AA) in 2018. [ 23 ] Eye-tracking procedure m-ETA comprised an Android-based application deployed on a tablet (equipped with a 30 Hz front-facing camera) for video acquisition. Participants underwent a comprehensive pre-task calibration procedure to verify optimal testing conditions and standardized model calibration for each individual’s head position including appropriate ambient illumination, correct participant-device distance (approximately 50–60 cm), stable head positioning, and removal of potential visual obstructions (e.g., glasses). Following calibration, participants completed two standardized eye-tracking tasks associated with cognitive function: an anti-saccade task and a visual paired comparison (VPC) task ( Fig. 2 ) . Recordings from the two eye-tracking tasks were securely transmitted to a local server, where oculometric features such as Accuracy (the percentage of trials with correct gaze shifts) and Central (mean) (the ratio of dwelling on novel images to old images) were extracted. m-ETA then automatically processed and analyzed these oculometric features to calculate personalized risk probabilities for CI. These scores were presented as a diagnostic report within the tablet’s user interface. Detailed descriptions of m-ETA, quality control criteria, oculometric features, and their clinical interpretations are outlined in the Supplementary Material Methods , Fig S1 and Table S1 . Operated by personnel with standardized 15-minute training (no specialized expertise required), the application completes testing in approximately 6 minutes while enabling automated data acquisition and real-time analytical feedback. To ensure model reliability, we implemented rigorous quality control: (1) excluded frames with off-screen gaze points; (2) removed videos with < 6 valid saccades; (3) retained only qualified data for analysis. This protocol ensured data integrity. Tau-PET and MRI assessment In the exploratory dataset, 82 patients underwent amyloid and 18F-APN-1607 tau PET scans using a uMI 780 PET/CT system (United Imaging Healthcare), while the remaining participants provided cerebrospinal fluid (CSF) biomarker data. All participants across both the exploratory dataset and external validation cohort underwent structural MRI examinations on a Siemens MAGNETOM Prisma 3T scanner. Image processing protocols are detailed in the Supplementary Materials. Visual assessments of brain atrophy were performed using established scales: the Global Cortical Atrophy (GCA) scale and Medial Temporal Lobe Atrophy (MTA) scale. [ 24 , 25 ] The MRI imaging protocol included T1-weighted, T2-weighted, and fluid-attenuated inversion recovery (FLAIR) sequences using a Siemens MAGNETOM Prisma 3T scanner (Siemens Healthineers, Erlangen, Germany) equipped with a 64-channel head coil. Two independent raters conducted standardized evaluations of the Global Cortical Atrophy (GCA) scale and Medial Temporal Lobe Atrophy (MTA) scale, demonstrating excellent inter-rater reliability (weighted κ: MTA = 0.80, GCA = 0.62; regional atrophy assessments on GCA scale: κ = 0.77–0.88). Statistical analysis Descriptive statistics were presented as means (± standard deviations, SD), median (interquartile range, IQR) or frequencies (percentages). Categorical variables were compared using chi-square tests, and continuous variables with Mann-Whitney U tests. For feature selection and model construction, Spearman correlation analysis was used to eliminate redundant features and prevent collinearity. Multiple stepwise regression analysis was then employed to identify features that significantly contributed to the model. With the dataset randomly split into training (75%) and test (25%) sets, the machine learning classification model, Logistic Regression (LR), is used to distinguish between NC and dementia, NC and CI individuals. (Detailed in Supplementary Material methods ). Model interpretability was visualized using Shapley Additive Explanations (SHAP) analysis. Model performance was evaluated using sensitivity, specificity, negative predictive value (NPV), and area under the curve (AUC). To explore the clinical associations, regression analyses were used to evaluate the association between m-ETA-derived features and regional tau-PET standardized uptake value ratio (SUVR; linear regression), atrophy scales (GCA/MTA/neuropsychological assessments; beta regression). [ 26 ] Statistical significance was defined as P < 0.05, with False Discovery Rate (FDR) correction applied for multiple comparisons where applicable. All analyses were conducted using Python (v3.10). Data sharing statement The datasets generated and analysed during the current study are not publicly available due to data sensitivity but are available from the corresponding author on reasonable request. Results Baseline characteristics This study incorporated two independent datasets and three distinctive cohorts. The discovery dataset comprised the NC and dementia groups with age and sex matched (Table 1) . The exploratory dataset enrolled AD patients with a mean age of 62.2 ± 9.3 years, of whom 52.5% were females (Table S2a) . The external validation consisted of the NC and the CI groups with a mean age of 68.1 ± 3.5 years and females accounted for 52.4% (Table S2b) , and the CI group exhibited a higher mean age and a greater proportion of females than the NC group (Table 1) . Additionally, two community-based cohorts from Shanghai were analyzed: the urban Hongmei cohort, with a mean age of 69.5 ± 8.0 years and 71.2% females, among whom 41.1% had attained at least a senior high school education; and the rural Chongming cohort, with a mean age of 69.5 ± 7.7 years and 65.1% females, of whom only 7.4% finished senior high school education or higher (Table S2b) . Significant disparities in MMSE and MoCA scores, as well as m-ETA-derived features, were observed between the NC and both the dementia patients and the CI group (Table 1, Table S3) . Developing an m-ETA in clinical settings Through correlation analysis and stepwise regression, six oculometric features were finally identified as m-ETA-derived signatures ( Fig S2 A-B ), including Central (mean), Fix duration valid (mean), Accuracy, Correction latency (mean), Peak velocity (mean) and Final ratio (mean). Employing the LR approach, which achieved optimal performance among various algorithms (Fig S3) , m-ETA demonstrated exceptional classification accuracy with an AUC of 0.99 (95% CI: 0.96,1.00), sensitivity of 91.3%, specificity of 96.4%, and NPV of 93.1%. m-ETA significantly outperformed in comparison with both the conventional demographic model (age, sex, and education level; AUC = 0.74, 95% CI: 0.59–0.86) and the combined model integrating m-ETA with demographic data (AUC = 0.94, 95% CI: 0.86–0.99) ( Fig. 3 A ) . Diagnostic performance was further quantified using receiver operating characteristic (ROC) curves (Fig. 3 A), with detailed classification metrics and performance statistics presented in Fig. 3 B. Associations between m-ETA and AD biomarkers To elucidate the potential biological plausibility for the discriminative capacity of m-ETA in characterizing dementia, we investigated its links to AD biomarkers. Accuracy and Central (mean), which ranked first and third, respectively, in SHAP analysis for m-ETA ( Fig S2 C ), were associated with cognitive function in dementia patients (Table S4) and maintained positive correlations with global and domain-specific cognitive scores in exploratory dataset (Fig. 4 A). Accuracy demonstrated a robust negative association with GCA scores (β= -1.662, FDR-corrected P = 0.007) in structural MRI, particularly in frontal (β= -0.875, FDR-corrected P = 0.050) (Fig. 4 B). Additionally, Central (mean) showed negative association with medial temporal lobe atrophy (MTA/GCA) in structural brain changes (MTA, β= -0.143, P = 0.048; GCA, β= -2.021, P = 0.020, Table S5 ). Particularly, there was a significant association between Central (mean) and tau deposition in temporal lobe detected by molecular imaging (β= -1.004, FDR-corrected P = 0.050), suggesting that tau pathology may contribute to visual processing deficits in AD (Fig. 4 C, Table S6 ). Performance of m-ETA in Community Settings 1) External validation in community for CI detection Next, we investigated the value of m-ETA in detecting CI in communities by assessing its performance in an actual community with comprehensive cognitive data. m-ETA yielded a robust AUC of 0.80 (95% CI: 0.75, 0.85), with sensitivity of 71.2%, specificity of 73.3% and NPV of 87.2%. This performance substantially exceeded the demographic model (AUC = 0.59, 95% CI: 0.53, 0.66) and matched the combined model (AUC = 0.80, 95% CI: 0.75, 0.85), which demonstrated the standalone utility of m-ETA for community-based CI screening. The classification accuracy was systematically evaluated through ROC curve analysis ( Fig. 3 C ) , with detailed performance metrics visualized in a confusion matrix ( Fig. 3 D ) . Participants with prediction probability predicted by m-ETA below the cutoff were categorized as low probability (Low-Prob) of CI, whereas those exceeding the cutoff were designated as high probability (High-Prob). The High-Prob group exhibited significantly lower education levels and poorer performance across global and domain-specific cognitive measures than the Low-Prob group ( P < 0.001) (Fig. 5 A-F, Table S7 ). No significant difference was observed in APOE-ε4 allele distribution between the two groups ( Table S7 ). 2) Multicenter validation in real-world applications Furthermore, we investigated whether our findings could be replicated in other two communities. The High-Prob group showed remarkably higher AD-8 scores than the Low-Prob group in Hongmei cohort (median 4 [IQR 7] vs. 2 [IQR 8], P = 0.002), as well as in Chongming cohort (median 0 [IQR 3] vs. 0 [IQR 1], P < 0.001) (Fig. 5 G-H). Additionally, the High-Prob group exhibited lower education levels and a higher mean age in both cohorts (all P < 0.001), aligning with traditional risk factors for cognitive decline ( Table S8-9 ), which further supports the classification’s accuracy. Discussion In this study, we developed a m-ETA, verified its biological relevance and then confirmed its accuracy and generalizability across community-dwelling elderly populations. Remarkably, m-ETA achieved a high AUC of 0.99 in dementia discrimination and showed significant correlations with AD-related neuropathological changes, including regional brain atrophy and tau accumulation. m-ETA demonstrated robust detection accuracy for cognitive impairment risk (AUC = 0.80) in community-dwelling populations, underscoring its potential as a non-invasive, scalable screening tool, particularly valuable in resource-limited settings. m-ETA, which uses the deep learning-based method for eye gaze tracking with regular cameras, eliminating the need for near-infrared (NIR) illumination by leveraging advanced Artificial Intelligence algorithms to facilitate gaze estimation. We utilized the public GazeCapture and domestic databases(1450 participants worldwide and over 2.5 million frames, [ 27 ] 550 participants and over 3.5 million frames, respectively) to develop the gaze model. Participants of databases were required to gaze at specific dots on the screen of a mobile device, while the device’s regular camera captured images of their faces and eye features as they looked at these points. Then, we trained the model that, by inputting these photos containing facial and eye features, could output the predicted gaze points. The performance of the gaze model was assessed by comparing the positions of the predicted gaze points with the actual screen coordinates. This process was achieved through the deep learning of a Convolutional Neural Network (CNN) algorithm, and the efficacy and robustness of our gaze model have been rigorously validated across multiple benchmark datasets, including Eyediap, MPIIGaze, and UT-multiview. Notably, the gaze model achieved an error rate of 3.80 degrees on the MPIIGaze dataset. [ 28 ] In this study, we recorded videos of individuals performing eye-tracking tasks using regular cameras and extracted m-ETA-derived oculometric features, which were acknowledged association with cognition. Eye movements, which are fundamental to cognitive processing and mediated by widely distributed neural circuits, [ 29 ] serve as sensitive digital biomarkers for detecting subtle cognitive alterations. These movements—including saccades, smooth pursuit, fixation stability, and pupillary responses—are governed by cortical, subcortical, and cerebellar circuits, regulating visual attention and reflecting complicated higher-order cognitive capabilities. [ 30 ] Extensive evidence links eye movement abnormalities to cognitive deficits caused by neural circuit disruptions. For instance, schizophrenia patients exhibit impaired fixation stability and impaired saccades. [ 31 ],[ 32 ] In early-stage dementia, subtle cognitive decline—often undetectable by conventional neuropsychological tests—can be identified through simplified eye-tracking paradigms. While traditional NIR-based systems have validated these associations, their cost and complexity limit scalability. Our AI-based m-ETA helps to bridge this gap by offering a cost-effective, scalable alternative. We focused on two tasks with specific relevance to early dementia: the anti-saccade task, which assesses frontal lobe-mediated inhibitory control by requiring suppression of reflexive eye movements, [ 33 ] and the visual paired comparison task, which evaluates temporal lobe-mediated memory encoding. [ 34 ] These tasks target inhibitory deficits and memory impairments observed in early dementia, mirroring findings that identified these domains as the earliest markers of decline. [ 35 , 36 ] Our findings align with existing literature, demonstrating significant associations between m-ETA-derived features and cognition decline across multiple datasets (discovery, exploratory, and external validation), supporting their validity as digital biomarkers. It is noteworthy that anti-saccade Accuracy and Central (mean), being m-ETA’s most contributory features, showed prominent associations with clinical and pathological phenotypes of AD. Specifically, Accuracy exhibited a robust negative association with frontal lobe atrophy and executive dysfunction, consistent with known prefrontal inhibitory control mechanisms [ 37 ] while Central (mean) positively associated with memory performance and inversely with temporal lobe tau deposition, suggesting disrupted temporo-parietal integration during memory processing. For other features in m-ETA, previous studies have shown their association with higher cortical function. Fixed duration (mean), representing the total number of fixations in the VPC task, was acknowledged as an indicator of attentional control during visual processing. [ 38 , 39 ] Correction latency (mean) reflects error-monitoring efficiency, encompassing both error detection and correction evaluation. [ 40 ] Additionally, peak velocity (mean) and final ratio (mean) were considered, with lower values interpreted as indicative of worse oculometric performance mediated by diverse neural networks. [ 41 ] Collectively, these oculometric patterns mirror neural network dynamics across cognitive domains, positioning them as promising digital biomarkers for detecting CI and enhancing diagnostic workflows. Our tablet-based m-ETA offers several technological strengths for population-based screening. First of all, its non-invasiveness design and automated algorithm enable accessible, objective assessments with no requirement of specialized medical personnel, reducing cultural biases during the measurement. m-ETA operates without the need for dedicated facilities and offers real-time reports, providing significant economic benefits compared to traditional methods like blood biomarker tests or advanced imaging, particularly promising for primary care and community settings in rural and developing areas, where cost-effectiveness is crucial for equitable access to medical resources. Moreover, simple eye-tracking tasks remain unaffected by linguistic complexity, making it universally accessible across diverse populations. Additionally, beyond its notable AUC performance, m-ETA’s high NPV (87.2% in communities) is critical for screening efficiency. [ 42 ] High NPV minimizes unnecessary follow-up evaluations for low-risk individuals, alleviating healthcare burdens while also minimizing the stigma associated with cognitive testing, which is particularly important in culturally sensitive regions. In summary, our AI-based eye-tracking tool holds great promise for CI screening and exhibits greater accessibility compared to currently available tools. Despite these advances, several limitations warrant consideration. First, the study is currently limited to stratifying the probabilities of CI in the population. In the future, longitudinal follow-up can be conducted to evaluate m-ETA's potential as an early screening tool. Second, broader validation among the elderly in different regions of China and other LMICs, and across different mobile devices, like phones, is needed to provide more real-world applicability evidence for the implementation of m-ETA in large-scale cognitive screening. Finally, while the Chongming and Hongmei cohorts offered representative real-world validation, incomplete diagnostic data limited generalizability assessment. Subsequent research would benefit from community-based validation with complete neuropsychological and biomarker characterization. In summary, our study demonstrates that m-ETA represents an efficient and accessible tool for large-scale CI screening. This evidence-based application has the potential to provide a practical solution for advancing the CI detection and optimizing use of healthcare resource in resource-limited settings, such as primary care and rural communities in large populations like China. Declarations Acknowledgements The authors gratefully acknowledge the patients who made this study possible through their participation and the physicians who collected the data. Funding We acknowledge financial supports from the Ministry of Science and Technology of China (STI2030-Major Projects 2030 2021ZD0201806), National Key R&D Program of China (grant number: 2023YFC3606300), National Natural Science Foundation of China (grant numbers: 82373658, 82271221), Natural Science Foundation of Shanghai, China (grant number: 22ZR1405300), Shanghai Municipal Health Commission (grant number: 20234Z0013), Shanghai Municipal Health Commission Clinical Research Special Program for the Health Industry (grant number: 20244Y0103), Shanghai Oriental Talent Program, Shanghai Medical New Star Program. Contributors MC, YFJ and YZW contributed to the study’s conception, design, data acquisition, data analysis, data interpretation, and writing of the manuscript. MXW and JCL equally contributed to study design, data interpretation, drafting, and critically reviewing the manuscript. TYY, YY, JYL, SZL, ZSJ, QH, CTZ, and XDC contributed to the interpretation of data and revision of the manuscript. JFY, JTY, and QD contributed to the data acquisition and data analysis. All authors read and approved the final manuscript. Ethics declarations Ethics approval and consent to participate The studies were approved by Ethics Committee of Huashan Hospital, Fudan University, Shanghai, China (KY2024-1144) and the Ethics Committee of the Fudan University Taizhou Institute of Health Sciences (institutional review board approval number: B017). Written informed consent was obtained from all participants. Competing interest The authors declare no competing interests. References International AD. World Alzheimer Report 2023: Reducing Dementia Risk: Never too early, never too late. 2023 [cited 2025 Apr 29]; Available from: https://www.alzint.org/resource/world-alzheimer-report-2023/ Liu Y, Wu Y, Chen Y, Lobanov-Rostovsky S, Liu Y, Zeng M, et al. Projection for dementia burden in China to 2050: a macro-simulation study by scenarios of dementia incidence trends. Lancet Reg Health West Pac. 2024; 50:101158. Wimo A, Seeher K, Cataldi R, Cyhlarova E, Dielemann JL, Frisell O, et al. The worldwide costs of dementia in 2019. Alzheimers Dement. 2023; 19:2865–73. Nasreddine ZS, Phillips NA, Bédirian V, Charbonneau S, Whitehead V, Collin I, et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc. 2005; 53:695–9. Folstein MF, Folstein SE, McHugh PR. “Mini-mental state”. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975; 12:189–98. Liu X, Song F, Liu F, Mao Z, Qu S. Multiple chronic conditions among older adults in China: differences in socio-demographic characteristics. Heliyon. 2022;8:e11129. Jia L, Du Y, Chu L, Zhang Z, Li F, Lyu D, et al. Prevalence, risk factors, and management of dementia and mild cognitive impairment in adults aged 60 years or older in China: a cross-sectional study. Lancet Public Health. 2020;5: e661–71. Allegri RF, Butman J, Arizaga RL, Machnicki G, Serrano C, Taragano FE, et al. Economic impact of dementia in developing countries: an evaluation of costs of Alzheimer-type dementia in Argentina. Int Psychogeriatr. 2007; 19:705–18. Kourtis LC, Regele OB, Wright JM, Jones GB. Digital biomarkers for Alzheimer’s disease: the mobile/ wearable devices opportunity. NPJ Digit Med. 2019; 2:9. Boxer AL, Garbutt S, Seeley WW, Jafari A, Heuer HW, Mirsky J, et al. Saccade abnormalities in autopsy-confirmed frontotemporal lobar degeneration and Alzheimer disease. Arch Neurol. 2012; 69:509–17. Lage C, López-García S, Bejanin A, Kazimierczak M, Aracil-Bolaños I, Calvo-Córdoba A, et al. Distinctive Oculomotor Behaviors in Alzheimer’s Disease and Frontotemporal Dementia. Front Aging Neurosci. 2020; 12:603790. Tokushige S-I, Matsumoto H, Matsuda S-I, Inomata-Terada S, Kotsuki N, Hamada M, et al. Early detection of cognitive decline in Alzheimer’s disease using eye tracking. Front Aging Neurosci. 2023; 15:1123456. Xu Y, Zhang C, Pan B, Yuan Q, Zhang X. A portable and efficient dementia screening tool using eye tracking machine learning and virtual reality. npj Digit Med. 2024; 7:219. Chehrehnegar N, Shati M, Esmaeili M, Foroughan M. Executive function deficits in mild cognitive impairment: evidence from saccade tasks. Aging Ment Health. 2022; 26:1001–9. Jiang Y, Cui M, Tian W, Zhu S, Chen J, Suo C, et al. Lifestyle, multi‐omics features, and preclinical dementia among Chinese: The Taizhou Imaging Study. Alzheimer’s Dement. 2021; 17:18–28. Katzman R, Zhang MY, Ouang-Ya-Qu, Wang ZY, Liu WT, Yu E, et al. A Chinese version of the Mini-Mental State Examination; impact of illiteracy in a Shanghai dementia survey. J Clin Epidemiol. 1988; 41:971–8. Yu J, Li J, Huang X. The Beijing version of the Montreal Cognitive Assessment as a brief screening tool for mild cognitive impairment: a community-based study. BMC Psychiatry. 2012; 12:156. Zhao Q, Lv Y, Zhou Y, Hong Z, Guo Q. Short-term delayed recall of auditory verbal learning test is equivalent to long-term delayed recall for identifying amnestic mild cognitive impairment. PLoS One. 2012;7: e51157. Zhao Q, Guo Q, Li F, Zhou Y, Wang B, Hong Z. The Shape Trail Test: application of a new variant of the Trail making test. PLoS One. 2013;8: e57333. Zhao Q, Guo Q, Hong Z. Clustering and switching during a semantic verbal fluency test contribute to differential diagnosis of cognitive impairment. Neurosci Bull. 2013; 29:75–82. Zhang Y, Xu J, Zhang C, Zhang X, Yuan X, Ni W, et al. Community screening for dementia among older adults in China: a machine learning-based strategy. BMC Public Health. 2024; 24:1206. First MB. Diagnostic and Statistical Manual of Mental Disorders, 5th Edition, and Clinical Utility. J Nerv Ment Dis. 2013; 201:727–9. Jack CR, Bennett DA, Blennow K, Carrillo MC, Dunn B, Haeberlein SB, et al. NIA-AA Research Framework: Toward a biological definition of Alzheimer’s disease. Alzheimer’s Dement. 2018; 14:535–62. Hobden G, Colbourne E, Pendlebury ST, Demeyere N. Reliability of the global cortical atrophy visual rating scale applied to computed tomography versus magnetic resonance imaging scans in acute stroke. Neurol Sci. 2024; 45:1549–56. Zhang D, Zhang W, Ming C, Gao X, Yuan H, Lin X, et al. P-tau217 correlates with neurodegeneration in Alzheimer’s disease, and targeting p-tau217 with immunotherapy ameliorates murine tauopathy. Neuron. 2024; 112:1676-1693.e12. Lu M, Hedin LO. Global plant-symbiont organization and emergence of biogeochemical cycles resolved by evolution-based trait modelling. Nat Ecol Evol. 2019; 3:239–50. Krafka K, Khosla A, Kellnhofer P, Kannan H, Bhandarkar S, Matusik W, et al. Eye Tracking for Everyone. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) [Internet]. Las Vegas, NV, USA: IEEE; 2016 [cited 2025 Feb 18]. p. 2176–84. Available from: http://ieeexplore.ieee.org/document/7780608/ Liu G, Yu Y, Mora KAF, Odobez J-M. A Differential Approach for Gaze Estimation. IEEE Trans Pattern Anal Mach Intell. 2021; 43:1092–9. Kragel JE, Voss JL. Looking for the neural basis of memory. Trends Cogn Sci. 2022; 26:53–65. Liu X, Cheng Z, Lin H, Tan J, Chen W, Bao Y, et al. Decoding effects of psychoactive drugs in a high-dimensional space of eye movements in monkeys. Natl Sci Rev. 2023;10: 255. Kissler J, Clementz BA. Fixation stability among schizophrenia patients. Neuropsychobiology. 1998; 38:57–62. Camchong J, Dyckman KA, Chapman CE, Yanasak NE, McDowell JE. Basal ganglia-thalamocortical circuitry disruptions in schizophrenia during delayed response tasks. Biol Psychiatry. 2006; 60:235–41. Heuer HW, Mirsky JB, Kong EL, Dickerson BC, Miller BL, Kramer JH, et al. Antisaccade task reflects cortical involvement in mild cognitive impairment. Neurology. 2013; 81:1235–43. Zola SM, Squire LR, Teng E, Stefanacci L, Buffalo EA, Clark RE. Impaired recognition memory in monkeys after damage limited to the hippocampal region. J Neurosci. 2000; 20:451–63. Mueller S, Chao L, Berman B, Weiner M. Evidence for functional specialization of hippocampal subfields detected by MR subfield volumetry on high resolution images at 4T. Neuroimage. 2011; 56:851–7. Wilcockson TDW, Mardanbegi D, Xia B, Taylor S, Sawyer P, Gellersen HW, et al. Abnormalities of saccadic eye movements in dementia due to Alzheimer’s disease and mild cognitive impairment. Aging. 2019; 11:5389–98. Averbeck BB, O’Sullivan SS, Djamshidian A. Impulsive and Compulsive Behaviours in Parkinson’s disease. Annu Rev Clin Psycho. 2014; 10:553–80. Beckner AG, Arnold CD, Bragg MG, Caswell BL, Chen Z, Cox K, et al. Examining infants’ visual paired comparison performance in the US and rural Malawi. Dev Sci. 2024;27: e13439. Boujelbane MA, Trabelsi K, Salem A, Ammar A, Glenn JM, Boukhris O, et al. Eye Tracking During Visual Paired-Comparison Tasks: A Systematic Review and Meta-Analysis of the Diagnostic Test Accuracy for Detecting Cognitive Decline. Journal of Alzheimer’s Disease. 2024; 99:207–21. Opwonya J, Wang C, Jang K-M, Lee K, Kim JI, Kim JU. Inhibitory Control of Saccadic Eye Movements and Cognitive Impairment in Mild Cognitive Impairment. Front Aging Neurosci. 2022; 14:871432. Nij Bijvank JA, Strijbis EMM, Nauta IM, Kulik SD, Balk LJ, Stam CJ, et al. Impaired saccadic eye movements in multiple sclerosis are related to altered functional connectivity of the oculomotor brain network. Neuroimage Clin. 2021; 32:102848. O’Bryant Sid E, Zhang Fan, Petersen Melissa, Hall James R, Johnson Leigh A, Yaffe Kristine, et al. A blood screening tool for detecting mild cognitive impairment and Alzheimer’s disease among community-dwelling Mexican Americans and non-Hispanic Whites: A method for increasing representation of diverse populations in clinical research. Alzheimer’s dement : j Alzheimer’s Assoc [Internet]. 2021;18. Available from: https://kns.cnki.net/kcms2/article/abstract?v=WnCf0VAm38qAhkKNKxCqX8bWBamJbVUxXi6cNJbtJmViVZc4-oEpWno6w5mcdcAEQhxdqvVJoucrw-8Lc5qlJZ5mLuuA7TpgQy0CZpdHIYev-jkirU9WV36kctMpECePrI6Ao8oxSsma-nBAr_5AXA==&uniplatform=NZKPT&language=gb Table 1 Table 1. Characteristics of the discovery dataset and the external validation cohort Discovery dataset (n=204) External validation cohort (n=433) NC (n=106) Dementia (n=98) P -value NC (n=315) CI (n=118) P -value Age, years, mean (SD) 66.8(7.6) 68.2(8.50) 0.112 67.9(3.7) 68.6(3.0) 0.080 Female, No. (%) 64(60.4) 55(56) 0.538 153(48.6) 74(62.7) 0.009 Education, No. (%) <0.001 0.004 Illiteracy and primary 6(5.7) 14(14) 182(57.8) 90(76.3) Junior 34(32.1) 51(52) 91(28.9) 22(18.6) Senior and above 66(62.3) 33(34) 42(13.3) 6(5.1) MMSE, median (IQR) 28(2) 18(6) <0.001 25(5) 17(10) <0.001 MoCA, median (IQR) 25(5) 13(6) <0.001 17(8) 10(7) <0.001 Central (mean), mean (SD), % 0.75(0.11) 0.54(0.10) <0.001 0.64(0.12) 0.54(0.14) <0.001 Fix duration valid (mean), mean (SD), ms 4517.58(1254.00) 4925.16(963.63) 0.017 5749.16(746.10) 5788.13(679.33) 0.907 Accuracy, mean (SD), % 0.53(0.21) 0.38(0.23) <0.001 0.53(0.21) 0.39(0.22) <0.001 Correction latency (mean), mean (SD), ms 241.23(162.96) 337.07(167.61) <0.001 307.66(116.50) 400.87(143.37) <0.001 Peak velocity (mean), mean (SD), pixel/ms 4.47(1.75) 4.77(1.58) 0.137 4.28(1.46) 3.90(1.31) 0.013 Final ratio (mean), mean (SD), % 1.51(5.52) 1.40(2.82) <0.001 6.04(46.03) 1.33(2.11) <0.001 Abbreviations: CI, cognitive impairment; IQR, interquartile range; MMSE, Mini Mental State Examination; MoCA, Montreal Cognitive Assessment; NC, normal cognition; SD, standard deviation. Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial.docx Cite Share Download PDF Status: Published Journal Publication published 17 Nov, 2025 Read the published version in Alzheimer's Research & Therapy → Version 1 posted Editorial decision: Revision requested 11 Sep, 2025 Reviews received at journal 29 Aug, 2025 Reviewers agreed at journal 26 Aug, 2025 Reviewers agreed at journal 25 Aug, 2025 Reviews received at journal 24 Jul, 2025 Reviewers agreed at journal 24 Jul, 2025 Reviewers agreed at journal 24 Jul, 2025 Reviewers invited by journal 24 Jul, 2025 Editor assigned by journal 16 Jul, 2025 Submission checks completed at journal 16 Jul, 2025 First submitted to journal 16 Jul, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7141899","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":491382763,"identity":"11fd4daa-8c0e-4cda-85c4-d90f1ff22ea0","order_by":0,"name":"Mingxia Wei","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Mingxia","middleName":"","lastName":"Wei","suffix":""},{"id":491382764,"identity":"7a01af85-e87c-4b9f-ba4f-e1d957df5274","order_by":1,"name":"Jincheng Li","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Jincheng","middleName":"","lastName":"Li","suffix":""},{"id":491382765,"identity":"be25319a-a971-4b34-bf43-e089e05fbd56","order_by":2,"name":"Tongyao You","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Tongyao","middleName":"","lastName":"You","suffix":""},{"id":491382766,"identity":"308d2331-40c2-4ce1-bed1-d5d31781ed4f","order_by":3,"name":"Yu Yu","email":"","orcid":"","institution":"Department of R\u0026D, NeuroWeave, Co.","correspondingAuthor":false,"prefix":"","firstName":"Yu","middleName":"","lastName":"Yu","suffix":""},{"id":491382767,"identity":"69ce8188-e846-4a19-aaac-f67a9d657d4b","order_by":4,"name":"Jiaying Lu","email":"","orcid":"","institution":"Department of Nuclear Medicine and PET Center, National Center for Neurological Disorders, and National Clinical Research Center for Aging and Medicine, Huashan Hospital, Fudan University,","correspondingAuthor":false,"prefix":"","firstName":"Jiaying","middleName":"","lastName":"Lu","suffix":""},{"id":491382768,"identity":"aaa73cff-3a4c-4b11-bf8c-35a5685e13f6","order_by":5,"name":"Suzhen Liang","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Suzhen","middleName":"","lastName":"Liang","suffix":""},{"id":491382769,"identity":"28944081-1d39-41a7-a313-afdc0f1c7388","order_by":6,"name":"Zishuo Jin","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Zishuo","middleName":"","lastName":"Jin","suffix":""},{"id":491382770,"identity":"5e468763-1c15-40ce-9ee1-306632a8b727","order_by":7,"name":"Qi Han","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Han","suffix":""},{"id":491382773,"identity":"d5d9b99c-9f8b-45ac-a78f-38b307ec31fa","order_by":8,"name":"Chuantao Zuo","email":"","orcid":"","institution":"Department of Nuclear Medicine and PET Center, National Center for Neurological Disorders, and National Clinical Research Center for Aging and Medicine, Huashan Hospital, Fudan University,","correspondingAuthor":false,"prefix":"","firstName":"Chuantao","middleName":"","lastName":"Zuo","suffix":""},{"id":491382775,"identity":"f54714ad-dbd4-4c80-84fb-e8b0ecdf5368","order_by":9,"name":"Jianfeng Ye","email":"","orcid":"","institution":"Department of R\u0026D, NeuroWeave, Co.","correspondingAuthor":false,"prefix":"","firstName":"Jianfeng","middleName":"","lastName":"Ye","suffix":""},{"id":491382777,"identity":"ab7eeb8d-62fe-4a58-a928-0fa10eae65b4","order_by":10,"name":"Jintai Yu","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Jintai","middleName":"","lastName":"Yu","suffix":""},{"id":491382778,"identity":"2ce27f12-c9b6-405f-b077-e7df5a3992b1","order_by":11,"name":"Xingdong Chen","email":"","orcid":"","institution":"Human Phenome Institute, Shanghai Pudong Hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Xingdong","middleName":"","lastName":"Chen","suffix":""},{"id":491382779,"identity":"629fb0d3-bac4-4ffc-b264-04e21cae0822","order_by":12,"name":"Qiang Dong","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Qiang","middleName":"","lastName":"Dong","suffix":""},{"id":491382780,"identity":"b42b4a6d-0cc3-4925-a7ba-3960034a1dd5","order_by":13,"name":"Yingzhe Wang","email":"","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Yingzhe","middleName":"","lastName":"Wang","suffix":""},{"id":491382781,"identity":"4d15183d-e642-4ac3-95b9-b4d9dc4f49de","order_by":14,"name":"Yanfeng Jiang","email":"","orcid":"","institution":"Human Phenome Institute, Shanghai Pudong Hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Yanfeng","middleName":"","lastName":"Jiang","suffix":""},{"id":491382782,"identity":"61a9d62d-905b-4505-a87f-c7944a4a468d","order_by":15,"name":"Mei Cui","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAz0lEQVRIiWNgGAWjYFACxgYGBgMbfgaGBBCPmWgtaZINJGgBg8MkaOGfkdy64UPBeQmD48nPHjBUWCc2sJ89gFeLxI3EtpszDG5LGJx5Zm7AcCY9sYEnLwGvFgOJxLbbPAa36wxuJJhJMLYdTmyQ4DEgrOWPwTkJgxvp3yQY/xGrhcHgAFBLDtCWBiK0SJx52HazxyBZQvLMmzKJhGPpxm08Ofi18LenP7vx44+dBN/x9G0SH2qsZfvZz+DXggoSgJiNBPWjYBSMglEwCnAAABULRsYzabjcAAAAAElFTkSuQmCC","orcid":"","institution":"Department of Neurology, National Center for Neurological Disorders, Huashan hospital, Fudan University","correspondingAuthor":true,"prefix":"","firstName":"Mei","middleName":"","lastName":"Cui","suffix":""}],"badges":[],"createdAt":"2025-07-16 16:08:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7141899/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7141899/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s13195-025-01884-7","type":"published","date":"2025-11-17T15:56:50+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":87661860,"identity":"aad764fa-d543-42b1-9723-2b455074ebed","added_by":"auto","created_at":"2025-07-27 10:21:19","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":895296,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart showing the overall study design.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAbbreviations: AD, Alzheimer’s disease; ATN, Amyloid/Tau/Neurodegeneration; CI, cognitive impairment; D, dementia; LR, logistic regression; m-ETA, Mobile Eye-Tracking Application; NC, normal cognition.\u003c/p\u003e\n\u003cp\u003eNote: *All AD patients were diagnosed according to the ATN biomarker framework.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7141899/v1/c4509d5eb09bce44aad36ede.jpg"},{"id":87662143,"identity":"a08a7baf-7811-474d-9c9a-28523953986d","added_by":"auto","created_at":"2025-07-27 10:29:19","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":551931,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe Eye-tracking paradigm.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNote: The eye-tracking paradigm incorporates the visual paired comparison (VPC) task and the anti-saccade task. The anti-saccade task requires participants to shift gaze away from a red dot stimulus to the opposite screen location. The VPC task requires novel image viewing. Latency (mean), Correct latency (mean), Correction latency (mean), Velocity (mean), Peak velocity (mean), Final ratio (mean), Initial amplitude (mean), and Accuracy were recorded in anti-saccade task. Central (mean), Central duration (mean), Old duration (mean), Fix duration valid (mean) were recorded in VPC task.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7141899/v1/b683f4be2734ec70d2eb15eb.jpg"},{"id":87661863,"identity":"721d759f-85b2-434b-be90-6902851bf7a5","added_by":"auto","created_at":"2025-07-27 10:21:19","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":625626,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003em-ETA performance in dementia discrimination and community cognitive impairment screening.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAbbreviations: CI, cognitive impairment; m-ETA, Mobile Eye-Tracking Application; NC, normal cognition; Prob, probability.\u003c/p\u003e\n\u003cp\u003eNote: Receiver operating characteristic (ROC) curves and cross-tabulations of internal validation in discovery dataset (A-B) and external validation in external validation cohort (C-D) are presented. Panels B and D present cross-tabulations comparing m-ETA classifications with cognitive diagnoses, along with corresponding model performance metrics. Area under the curve (AUC) metrics and 95% Confidence Intervals (CIs) are reported on each ROC plot in panels A and C. The blue-line ROC includes demographic characteristics (age, sex, and education level), the orange-line ROC includes six key mES-derived features: Central (mean), Fix duration valid (mean), Accuracy, Correction latency (mean), Peak velocity (mean) and Final ratio (mean), the red line ROC includes demographic characteristics and key mES-derived features.\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7141899/v1/0381b7a92d30b26a0e56ad95.jpg"},{"id":87661867,"identity":"4f3031ad-f234-4f32-bd5d-384948b9d8df","added_by":"auto","created_at":"2025-07-27 10:21:19","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":358062,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAssociation between m-ETA and biomarkers in Alzheimer's Disease.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAbbreviations: GCA, Global Cortical Atrophy; m-ETA, Mobile Eye-Tracking Application; MMSE, Mini Mental State Examination; MoCA, Montreal Cognitive Assessment; MTA, Medial Temporal lobe Atrophy.\u003c/p\u003e\n\u003cp\u003eNote: The associations between mES-derived features and cognitive assessments (A), brain atrophy score (B), and the standardized uptake value ratios of tau deposition in brain regions (C). The associations were analyzed using beta regression (A-B) and linear regression (C), with age, sex, education level as covariates. Refer to eTable 5-6 for more detailed analysis results.\u003c/p\u003e\n\u003cp\u003e* \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.01, *** \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7141899/v1/40bb225d8bcaf10e5412d511.jpg"},{"id":87662184,"identity":"cdaf4071-80da-4a16-b0f7-aef8d20fbfd8","added_by":"auto","created_at":"2025-07-27 10:37:19","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":487745,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe performance of m-ETA in real-world applications.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAbbreviations: AD-8, ascertain Dementia 8; m-ETA, Mobile Eye-Tracking Application; MMSE, Mini Mental State Examination; MoCA, Montreal Cognitive Assessment; Prob, probability; TIS, Taizhou Imaging Study.\u003c/p\u003e\n\u003cp\u003eNote: Participants with predicted probabilities below the cutoff were categorized as low probability of cognitive impairment (Low-Prob), whereas those exceeding the cutoff were designated as high probability (High-Prob). The differences in cognitive function between two groups of TIS (A-B), Hongmei cohort (C) and Chongming cohort (D) are shown in the figure.\u003c/p\u003e\n\u003cp\u003e* \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.01, ***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001, ****\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.0001\u003c/p\u003e","description":"","filename":"Figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7141899/v1/2613a2f06e64e384a22389d2.jpg"},{"id":96649956,"identity":"8dbfb669-e95f-4b4b-8ba9-58e855c4f249","added_by":"auto","created_at":"2025-11-24 16:01:21","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3965210,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7141899/v1/7788b0ba-8913-482e-985b-cb11637b8b20.pdf"},{"id":87661868,"identity":"843cfe8f-3891-426e-ad38-1c4254ff23fc","added_by":"auto","created_at":"2025-07-27 10:21:19","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5435311,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-7141899/v1/f9cc337ad2951a3db6627216.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"An accessible and efficient mobile eye-tracking application for community-based cognitive impairment screening in China","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWith an aging population, dementia has become one of the leading causes of long-term disability and mortality worldwide, having a substantial burden on global health care systems and socioeconomic improvements. More than 55\u0026nbsp;million people worldwide currently living with dementia, a figure expected to nearly triple to 139\u0026nbsp;million by 2050.\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e Notably, low- and middle-income countries (LMICs) shoulder a disproportionate burden of dementia, accounting for two-thirds of global cases, while China alone contributes nearly 25% of the worldwide dementia prevalence.\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e Cognitive impairment (CI) is a precursor of dementia, during which early detection and prompt treatment can slow the progression or even delay the onset of dementia.\u003c/p\u003e\u003cp\u003eHowever, detection of CI remains challenging in China. Currently, community‑based screening still depends on neuropsychological tests. Notwithstanding their clinical utility, these tests are labor-intensive, time-consuming, and require highly trained clinicians, greatly restricting their wider adoption. This challenge is particularly pronounced in LMICs like China, where high CI prevalence contrasts with limited resources of public health.\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e Another important issue worth considering is that these neuropsychological tests, despite being administered by trained professionals, are still susceptible to subjective bias. Additionally, educational barriers exacerbate this issue. In China, over 50% of adults aged over 60 have limited formal education (less than 6 years), a rate that exceeds 60% in rural regions.\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e These educational disparities have impeded the implementation of certain cognitive screening tools, such as the Montreal Cognitive Assessment (MoCA).\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e Therefore, there is an urgent need to develop an time-saving, objective and easy-to-use tool for large-scale CI screening in resource‑limited settings impractical.\u003c/p\u003e\u003cp\u003eGiven the rapid development of wearable devices, digital biomarkers\u0026mdash;due to their objective and non-invasive nature\u0026mdash;are gaining increasing popularity in screening research.\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e Accumulating evidence demonstrates that oculomotor patterns reflect distinct pathophysiological mechanisms in neurodegenerative diseases. The Alzheimer's disease continuum exhibits significantly prolonged saccadic latency, reflecting tau pathology-induced functional disconnection within the frontoparietal network.\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e In addition, Alzheimer\u0026rsquo;s patients displayed specific abnormalities in some eye-tracking tests, including elevated pursuit error, reduced target fixation during visual memory tasks, and prolonged search time with increased saccadic frequency in visual search tasks. These deficits are consistent with the known vulnerability of posterior cortical networks mediating visuospatial processing and attention.\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e Emerging research further underscores the clinical validity of oculometric patterns as promising digital biomarkers, demonstrating robust utility in identifying individuals with cognitive impairment through real-time tracking capabilities.\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e However, existing eye-tracking devices remain too bulky and expensive for widespread deployment, especially not being easily accessible to rural elders from China and other LMICs. Addressing this critical gap requires the development of portable, user-friendly, and cost-effective eye-tracking devices, which are essential for achieving equitable, population-level screening of cognitive impairment.\u003c/p\u003e\u003cp\u003eIn this study, we developed and validated a user-friendly, tablet-compatible eye-tracking application, the Mobile Eye-Tracking Application (m-ETA), to detect CI in community settings. m-ETA demonstrated high accuracy in identifying individuals with impaired cognition across diverse settings, spanning both clinical and community populations. Furthermore, the study provided insights into the biological plausibility linking m-ETA to Alzheimer\u0026rsquo;s biomarkers, and positioned m-ETA as a scalable digital biomarker for CI, facilitating large-scale cognitive screening applications.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003e\u003cb\u003eStudy design\u003c/b\u003e\u003c/p\u003e\u003cp\u003eUsing a three-step approach (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), the discovery phase utilized a distinct discovery dataset(N\u0026thinsp;=\u0026thinsp;204). 145 dementia patients and 159 normal cognition (NC) individuals were recruited from the memory clinic or medical examination center at Huashan hospital. After performing 1:1 propensity score matching (PSM) to account for age and sex, 112 case-control pairs were matched. And 20 outlier samples failed to quality control were removed. Finally, 106 dementia cases and 98 normal controls were identified. This phase aimed to identify key oculometric features associated with cognitive decline and developed the m-ETA.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eIn the exploratory phase, we included an exploratory dataset comprising 101 Alzheimer\u0026rsquo;s disease (AD) patients to further validate the biological feasibility of m-ETA through explore its links to cognitive function scores, Magnetic Resonance (MR) -derived structural brain changes, and pathological protein deposition via Tau-positron emission tomography (PET) imaging. All patients were clinically assessed at the Department of Neurology, Huashan Hospital.\u003c/p\u003e\u003cp\u003eIn the validation phase, we externally assessed the clinically established m-ETA using an external validation cohort enrolled form Taizhou imaging study (TIS), which included 433 participants with complete eye-tracking data. This community-based elderly cohort, enriched with comprehensive cognitive data, allowed assessment of the model\u0026rsquo;s utility in screening for CI in community populations. To further validate m-ETA\u0026rsquo;s generalizability, we conducted multicenter validation across distinct demographic populations: 806 urban residents from the Hongmei community and 2,577 rural residents from the Chongming community. After quality control of eye-movement data, 660 urban and 2,025 rural participants were retained, providing balanced representation of Shanghai\u0026rsquo;s urban and rural elderly populations.\u003c/p\u003e\u003cp\u003e\u003cb\u003eNeuropsychological assessment and diagnosis\u003c/b\u003e\u003c/p\u003e\u003cp\u003eAll participants across the discovery dataset, exploratory dataset and external validation cohort underwent comprehensive neuropsychological evaluations.\u003csup\u003e[\u003cspan additionalcitationids=\"CR16 CR17 CR18 CR19\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e General cognitive function was assessed with the Mini-Mental Status Examination (MMSE) and the Beijing version Montreal Cognitive Assessment (MoCA). Memory was assessed using the Auditory Verbal Learning Test (AVLT; for participants from the primary care dataset, exploratory dataset, or those with \u0026gt;\u0026thinsp;6 years of education in external validation cohort) or the Modified Fuld Object Memory Evaluation (FOME; for participants with \u0026le;\u0026thinsp;6 years of education in external validation cohort). Attention/executive function was measured through the Trail-Making Test A\u0026amp;B and the Conflicting Instructions Task (Go/No Go Task). Language abilities were assessed via Animal Fluency Test (AFT) and Boston Naming Test (BNT). Visuospatial function was examined using the Clock Drawing Test (CDT). All cognitive test scores were standardized into Z scores for comparative analysis.\u003c/p\u003e\u003cp\u003eParticipants from Chongming and Hongmei cohorts underwent rapid CI screening using the Ascertain Dementia 8 (AD-8) questionnaire,\u003csup\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e a brief, validated informant-based tool developed for dementia detection in community settings.\u003c/p\u003e\u003cp\u003eFor the discovery dataset, a consensus diagnosis of dementia was established by a panel of neurologists and neuropsychologists through integrated multimodal data (medical history, neurological examination, and neuropsychological assessment), strictly adhering to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V) criteria.\u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e For the participants of external validation cohort, CI was diagnosed by the same panel primarily using the MMSE criteria from a previous Chinese population-based study.\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e For the AD patients of exploratory dataset, diagnoses were based on the ATN (Amyloid/Tau/Neurodegeneration) biomarker framework proposed by the National Institute on Aging-Alzheimer's Association(NIA-AA) in 2018.\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eEye-tracking procedure\u003c/b\u003e\u003c/p\u003e\u003cp\u003em-ETA comprised an Android-based application deployed on a tablet (equipped with a 30 Hz front-facing camera) for video acquisition. Participants underwent a comprehensive pre-task calibration procedure to verify optimal testing conditions and standardized model calibration for each individual\u0026rsquo;s head position including appropriate ambient illumination, correct participant-device distance (approximately 50\u0026ndash;60 cm), stable head positioning, and removal of potential visual obstructions (e.g., glasses). Following calibration, participants completed two standardized eye-tracking tasks associated with cognitive function: an anti-saccade task and a visual paired comparison (VPC) task \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e. Recordings from the two eye-tracking tasks were securely transmitted to a local server, where oculometric features such as Accuracy (the percentage of trials with correct gaze shifts) and Central (mean) (the ratio of dwelling on novel images to old images) were extracted. m-ETA then automatically processed and analyzed these oculometric features to calculate personalized risk probabilities for CI. These scores were presented as a diagnostic report within the tablet\u0026rsquo;s user interface. Detailed descriptions of m-ETA, quality control criteria, oculometric features, and their clinical interpretations are outlined in the \u003cb\u003eSupplementary Material Methods\u003c/b\u003e, \u003cb\u003eFig \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e and \u003cb\u003eTable \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eOperated by personnel with standardized 15-minute training (no specialized expertise required), the application completes testing in approximately 6 minutes while enabling automated data acquisition and real-time analytical feedback.\u003c/p\u003e\u003cp\u003eTo ensure model reliability, we implemented rigorous quality control: (1) excluded frames with off-screen gaze points; (2) removed videos with \u0026lt;\u0026thinsp;6 valid saccades; (3) retained only qualified data for analysis. This protocol ensured data integrity.\u003c/p\u003e\u003cp\u003e\u003cb\u003eTau-PET and MRI assessment\u003c/b\u003e\u003c/p\u003e\u003cp\u003eIn the exploratory dataset, 82 patients underwent amyloid and 18F-APN-1607 tau PET scans using a uMI 780 PET/CT system (United Imaging Healthcare), while the remaining participants provided cerebrospinal fluid (CSF) biomarker data. All participants across both the exploratory dataset and external validation cohort underwent structural MRI examinations on a Siemens MAGNETOM Prisma 3T scanner. Image processing protocols are detailed in the Supplementary Materials.\u003c/p\u003e\u003cp\u003eVisual assessments of brain atrophy were performed using established scales: the Global Cortical Atrophy (GCA) scale and Medial Temporal Lobe Atrophy (MTA) scale.\u003csup\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e The MRI imaging protocol included T1-weighted, T2-weighted, and fluid-attenuated inversion recovery (FLAIR) sequences using a Siemens MAGNETOM Prisma 3T scanner (Siemens Healthineers, Erlangen, Germany) equipped with a 64-channel head coil. Two independent raters conducted standardized evaluations of the Global Cortical Atrophy (GCA) scale and Medial Temporal Lobe Atrophy (MTA) scale, demonstrating excellent inter-rater reliability (weighted κ: MTA\u0026thinsp;=\u0026thinsp;0.80, GCA\u0026thinsp;=\u0026thinsp;0.62; regional atrophy assessments on GCA scale: κ\u0026thinsp;=\u0026thinsp;0.77\u0026ndash;0.88).\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eDescriptive statistics were presented as means (\u0026plusmn;\u0026thinsp;standard deviations, SD), median (interquartile range, IQR) or frequencies (percentages). Categorical variables were compared using chi-square tests, and continuous variables with Mann-Whitney U tests.\u003c/p\u003e\u003cp\u003eFor feature selection and model construction, Spearman correlation analysis was used to eliminate redundant features and prevent collinearity. Multiple stepwise regression analysis was then employed to identify features that significantly contributed to the model. With the dataset randomly split into training (75%) and test (25%) sets, the machine learning classification model, Logistic Regression (LR), is used to distinguish between NC and dementia, NC and CI individuals. (Detailed in \u003cb\u003eSupplementary Material methods\u003c/b\u003e). Model interpretability was visualized using Shapley Additive Explanations (SHAP) analysis. Model performance was evaluated using sensitivity, specificity, negative predictive value (NPV), and area under the curve (AUC).\u003c/p\u003e\u003cp\u003eTo explore the clinical associations, regression analyses were used to evaluate the association between m-ETA-derived features and regional tau-PET standardized uptake value ratio (SUVR; linear regression), atrophy scales (GCA/MTA/neuropsychological assessments; beta regression).\u003csup\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e Statistical significance was defined as \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05, with False Discovery Rate (FDR) correction applied for multiple comparisons where applicable.\u003c/p\u003e\u003cp\u003eAll analyses were conducted using Python (v3.10).\u003c/p\u003e\u003cp\u003e\u003cb\u003eData sharing statement\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe datasets generated and analysed during the current study are not publicly available due to data sensitivity but are available from the corresponding author on reasonable request.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cb\u003eBaseline characteristics\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThis study incorporated two independent datasets and three distinctive cohorts. The discovery dataset comprised the NC and dementia groups with age and sex matched \u003cb\u003e(Table\u0026nbsp;1)\u003c/b\u003e. The exploratory dataset enrolled AD patients with a mean age of 62.2\u0026thinsp;\u0026plusmn;\u0026thinsp;9.3 years, of whom 52.5% were females \u003cb\u003e(Table S2a)\u003c/b\u003e. The external validation consisted of the NC and the CI groups with a mean age of 68.1\u0026thinsp;\u0026plusmn;\u0026thinsp;3.5 years and females accounted for 52.4% \u003cb\u003e(Table S2b)\u003c/b\u003e, and the CI group exhibited a higher mean age and a greater proportion of females than the NC group \u003cb\u003e(Table\u0026nbsp;1)\u003c/b\u003e. Additionally, two community-based cohorts from Shanghai were analyzed: the urban Hongmei cohort, with a mean age of 69.5\u0026thinsp;\u0026plusmn;\u0026thinsp;8.0 years and 71.2% females, among whom 41.1% had attained at least a senior high school education; and the rural Chongming cohort, with a mean age of 69.5\u0026thinsp;\u0026plusmn;\u0026thinsp;7.7 years and 65.1% females, of whom only 7.4% finished senior high school education or higher \u003cb\u003e(Table S2b)\u003c/b\u003e. Significant disparities in MMSE and MoCA scores, as well as m-ETA-derived features, were observed between the NC and both the dementia patients and the CI group \u003cb\u003e(Table\u0026nbsp;1, Table S3)\u003c/b\u003e.\u003c/p\u003e\u003cp\u003e\u003cb\u003eDeveloping an m-ETA in clinical settings\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThrough correlation analysis and stepwise regression, six oculometric features were finally identified as m-ETA-derived signatures (\u003cb\u003eFig S2 A-B\u003c/b\u003e), including Central (mean), Fix duration valid (mean), Accuracy, Correction latency (mean), Peak velocity (mean) and Final ratio (mean). Employing the LR approach, which achieved optimal performance among various algorithms \u003cb\u003e(Fig S3)\u003c/b\u003e, m-ETA demonstrated exceptional classification accuracy with an AUC of 0.99 (95% CI: 0.96,1.00), sensitivity of 91.3%, specificity of 96.4%, and NPV of 93.1%. m-ETA significantly outperformed in comparison with both the conventional demographic model (age, sex, and education level; AUC\u0026thinsp;=\u0026thinsp;0.74, 95% CI: 0.59\u0026ndash;0.86) and the combined model integrating m-ETA with demographic data (AUC\u0026thinsp;=\u0026thinsp;0.94, 95% CI: 0.86\u0026ndash;0.99) \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA\u003cb\u003e)\u003c/b\u003e. Diagnostic performance was further quantified using receiver operating characteristic (ROC) curves (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA), with detailed classification metrics and performance statistics presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eAssociations between m-ETA and AD biomarkers\u003c/b\u003e\u003c/p\u003e\u003cp\u003eTo elucidate the potential biological plausibility for the discriminative capacity of m-ETA in characterizing dementia, we investigated its links to AD biomarkers. Accuracy and Central (mean), which ranked first and third, respectively, in SHAP analysis for m-ETA (\u003cb\u003eFig S2 C\u003c/b\u003e), were associated with cognitive function in dementia patients \u003cb\u003e(Table S4)\u003c/b\u003e and maintained positive correlations with global and domain-specific cognitive scores in exploratory dataset (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). Accuracy demonstrated a robust negative association with GCA scores (β= -1.662, FDR-corrected \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.007) in structural MRI, particularly in frontal (β= -0.875, FDR-corrected \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.050) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). Additionally, Central (mean) showed negative association with medial temporal lobe atrophy (MTA/GCA) in structural brain changes (MTA, β= -0.143, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.048; GCA, β= -2.021, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.020, \u003cb\u003eTable S5\u003c/b\u003e). Particularly, there was a significant association between Central (mean) and tau deposition in temporal lobe detected by molecular imaging (β= -1.004, FDR-corrected \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.050), suggesting that tau pathology may contribute to visual processing deficits in AD (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC, \u003cb\u003eTable S6\u003c/b\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003ePerformance of m-ETA in Community Settings\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003e1) External validation in community for CI detection\u003c/b\u003e\u003c/p\u003e\u003cp\u003eNext, we investigated the value of m-ETA in detecting CI in communities by assessing its performance in an actual community with comprehensive cognitive data. m-ETA yielded a robust AUC of 0.80 (95% CI: 0.75, 0.85), with sensitivity of 71.2%, specificity of 73.3% and NPV of 87.2%. This performance substantially exceeded the demographic model (AUC\u0026thinsp;=\u0026thinsp;0.59, 95% CI: 0.53, 0.66) and matched the combined model (AUC\u0026thinsp;=\u0026thinsp;0.80, 95% CI: 0.75, 0.85), which demonstrated the standalone utility of m-ETA for community-based CI screening. The classification accuracy was systematically evaluated through ROC curve analysis \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC\u003cb\u003e)\u003c/b\u003e, with detailed performance metrics visualized in a confusion matrix \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD\u003cb\u003e)\u003c/b\u003e.\u003c/p\u003e\u003cp\u003eParticipants with prediction probability predicted by m-ETA below the cutoff were categorized as low probability (Low-Prob) of CI, whereas those exceeding the cutoff were designated as high probability (High-Prob). The High-Prob group exhibited significantly lower education levels and poorer performance across global and domain-specific cognitive measures than the Low-Prob group (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA-F, \u003cb\u003eTable S7\u003c/b\u003e). No significant difference was observed in APOE-ε4 allele distribution between the two groups (\u003cb\u003eTable S7\u003c/b\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003e2) Multicenter validation in real-world applications\u003c/b\u003e\u003c/p\u003e\u003cp\u003eFurthermore, we investigated whether our findings could be replicated in other two communities. The High-Prob group showed remarkably higher AD-8 scores than the Low-Prob group in Hongmei cohort (median 4 [IQR 7] vs. 2 [IQR 8], \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.002), as well as in Chongming cohort (median 0 [IQR 3] vs. 0 [IQR 1], \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eG-H). Additionally, the High-Prob group exhibited lower education levels and a higher mean age in both cohorts (all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), aligning with traditional risk factors for cognitive decline (\u003cb\u003eTable S8-9\u003c/b\u003e), which further supports the classification\u0026rsquo;s accuracy.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we developed a m-ETA, verified its biological relevance and then confirmed its accuracy and generalizability across community-dwelling elderly populations. Remarkably, m-ETA achieved a high AUC of 0.99 in dementia discrimination and showed significant correlations with AD-related neuropathological changes, including regional brain atrophy and tau accumulation. m-ETA demonstrated robust detection accuracy for cognitive impairment risk (AUC\u0026thinsp;=\u0026thinsp;0.80) in community-dwelling populations, underscoring its potential as a non-invasive, scalable screening tool, particularly valuable in resource-limited settings.\u003c/p\u003e\u003cp\u003em-ETA, which uses the deep learning-based method for eye gaze tracking with regular cameras, eliminating the need for near-infrared (NIR) illumination by leveraging advanced Artificial Intelligence algorithms to facilitate gaze estimation. We utilized the public GazeCapture and domestic databases(1450 participants worldwide and over 2.5\u0026nbsp;million frames,\u003csup\u003e[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e 550 participants and over 3.5\u0026nbsp;million frames, respectively) to develop the gaze model. Participants of databases were required to gaze at specific dots on the screen of a mobile device, while the device\u0026rsquo;s regular camera captured images of their faces and eye features as they looked at these points. Then, we trained the model that, by inputting these photos containing facial and eye features, could output the predicted gaze points. The performance of the gaze model was assessed by comparing the positions of the predicted gaze points with the actual screen coordinates. This process was achieved through the deep learning of a Convolutional Neural Network (CNN) algorithm, and the efficacy and robustness of our gaze model have been rigorously validated across multiple benchmark datasets, including Eyediap, MPIIGaze, and UT-multiview. Notably, the gaze model achieved an error rate of 3.80 degrees on the MPIIGaze dataset.\u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e In this study, we recorded videos of individuals performing eye-tracking tasks using regular cameras and extracted m-ETA-derived oculometric features, which were acknowledged association with cognition.\u003c/p\u003e\u003cp\u003eEye movements, which are fundamental to cognitive processing and mediated by widely distributed neural circuits,\u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e serve as sensitive digital biomarkers for detecting subtle cognitive alterations. These movements\u0026mdash;including saccades, smooth pursuit, fixation stability, and pupillary responses\u0026mdash;are governed by cortical, subcortical, and cerebellar circuits, regulating visual attention and reflecting complicated higher-order cognitive capabilities.\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e Extensive evidence links eye movement abnormalities to cognitive deficits caused by neural circuit disruptions. For instance, schizophrenia patients exhibit impaired fixation stability and impaired saccades.\u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e],[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e In early-stage dementia, subtle cognitive decline\u0026mdash;often undetectable by conventional neuropsychological tests\u0026mdash;can be identified through simplified eye-tracking paradigms. While traditional NIR-based systems have validated these associations, their cost and complexity limit scalability. Our AI-based m-ETA helps to bridge this gap by offering a cost-effective, scalable alternative. We focused on two tasks with specific relevance to early dementia: the anti-saccade task, which assesses frontal lobe-mediated inhibitory control by requiring suppression of reflexive eye movements,\u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e and the visual paired comparison task, which evaluates temporal lobe-mediated memory encoding.\u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e These tasks target inhibitory deficits and memory impairments observed in early dementia, mirroring findings that identified these domains as the earliest markers of decline.\u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eOur findings align with existing literature, demonstrating significant associations between m-ETA-derived features and cognition decline across multiple datasets (discovery, exploratory, and external validation), supporting their validity as digital biomarkers. It is noteworthy that anti-saccade Accuracy and Central (mean), being m-ETA\u0026rsquo;s most contributory features, showed prominent associations with clinical and pathological phenotypes of AD. Specifically, Accuracy exhibited a robust negative association with frontal lobe atrophy and executive dysfunction, consistent with known prefrontal inhibitory control mechanisms\u003csup\u003e[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/sup\u003e while Central (mean) positively associated with memory performance and inversely with temporal lobe tau deposition, suggesting disrupted temporo-parietal integration during memory processing. For other features in m-ETA, previous studies have shown their association with higher cortical function. Fixed duration (mean), representing the total number of fixations in the VPC task, was acknowledged as an indicator of attentional control during visual processing.\u003csup\u003e[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e Correction latency (mean) reflects error-monitoring efficiency, encompassing both error detection and correction evaluation.\u003csup\u003e[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/sup\u003e Additionally, peak velocity (mean) and final ratio (mean) were considered, with lower values interpreted as indicative of worse oculometric performance mediated by diverse neural networks.\u003csup\u003e[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/sup\u003e Collectively, these oculometric patterns mirror neural network dynamics across cognitive domains, positioning them as promising digital biomarkers for detecting CI and enhancing diagnostic workflows.\u003c/p\u003e\u003cp\u003eOur tablet-based m-ETA offers several technological strengths for population-based screening. First of all, its non-invasiveness design and automated algorithm enable accessible, objective assessments with no requirement of specialized medical personnel, reducing cultural biases during the measurement. m-ETA operates without the need for dedicated facilities and offers real-time reports, providing significant economic benefits compared to traditional methods like blood biomarker tests or advanced imaging, particularly promising for primary care and community settings in rural and developing areas, where cost-effectiveness is crucial for equitable access to medical resources. Moreover, simple eye-tracking tasks remain unaffected by linguistic complexity, making it universally accessible across diverse populations. Additionally, beyond its notable AUC performance, m-ETA\u0026rsquo;s high NPV (87.2% in communities) is critical for screening efficiency.\u003csup\u003e[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/sup\u003e High NPV minimizes unnecessary follow-up evaluations for low-risk individuals, alleviating healthcare burdens while also minimizing the stigma associated with cognitive testing, which is particularly important in culturally sensitive regions. In summary, our AI-based eye-tracking tool holds great promise for CI screening and exhibits greater accessibility compared to currently available tools.\u003c/p\u003e\u003cp\u003eDespite these advances, several limitations warrant consideration. First, the study is currently limited to stratifying the probabilities of CI in the population. In the future, longitudinal follow-up can be conducted to evaluate m-ETA's potential as an early screening tool. Second, broader validation among the elderly in different regions of China and other LMICs, and across different mobile devices, like phones, is needed to provide more real-world applicability evidence for the implementation of m-ETA in large-scale cognitive screening. Finally, while the Chongming and Hongmei cohorts offered representative real-world validation, incomplete diagnostic data limited generalizability assessment. Subsequent research would benefit from community-based validation with complete neuropsychological and biomarker characterization.\u003c/p\u003e\u003cp\u003eIn summary, our study demonstrates that m-ETA represents an efficient and accessible tool for large-scale CI screening. This evidence-based application has the potential to provide a practical solution for advancing the CI detection and optimizing use of healthcare resource in resource-limited settings, such as primary care and rural communities in large populations like China.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors gratefully acknowledge the patients who made this study possible through their participation and the physicians who collected the data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe acknowledge financial supports from the Ministry of Science and Technology of China (STI2030-Major Projects 2030 2021ZD0201806), National Key R\u0026amp;D Program of China (grant number: 2023YFC3606300), National Natural Science Foundation of China (grant numbers: 82373658, 82271221), Natural Science Foundation of Shanghai, China (grant number: 22ZR1405300), Shanghai Municipal Health Commission (grant number: 20234Z0013), Shanghai Municipal Health Commission Clinical Research Special Program for the Health Industry (grant number: 20244Y0103), Shanghai Oriental Talent Program, Shanghai Medical New Star Program.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContributors\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMC, YFJ and YZW contributed to the study’s conception, design, data acquisition, data analysis, data interpretation, and writing of the manuscript. MXW and JCL equally contributed to study design, data interpretation, drafting, and critically reviewing the manuscript. TYY, YY, JYL, SZL, ZSJ, QH, CTZ, and XDC contributed to the interpretation of data and revision of the manuscript. JFY, JTY, and QD contributed to the data acquisition and data analysis. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics declarations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe studies were approved by Ethics Committee of Huashan Hospital, Fudan University, Shanghai, China (KY2024-1144) and the Ethics Committee of the Fudan University Taizhou Institute of Health Sciences (institutional review board approval number: B017). Written informed consent was obtained from all participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eInternational AD. World Alzheimer Report 2023: Reducing Dementia Risk: Never too early, never too late. 2023 [cited 2025 Apr 29]; Available from: https://www.alzint.org/resource/world-alzheimer-report-2023/\u003c/li\u003e\n\u003cli\u003eLiu Y, Wu Y, Chen Y, Lobanov-Rostovsky S, Liu Y, Zeng M, et al. Projection for dementia burden in China to 2050: a macro-simulation study by scenarios of dementia incidence trends. Lancet Reg Health West Pac. 2024; 50:101158. \u003c/li\u003e\n\u003cli\u003eWimo A, Seeher K, Cataldi R, Cyhlarova E, Dielemann JL, Frisell O, et al. The worldwide costs of dementia in 2019. Alzheimers Dement. 2023; 19:2865\u0026ndash;73. \u003c/li\u003e\n\u003cli\u003eNasreddine ZS, Phillips NA, B\u0026eacute;dirian V, Charbonneau S, Whitehead V, Collin I, et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc. 2005; 53:695\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eFolstein MF, Folstein SE, McHugh PR. \u0026ldquo;Mini-mental state\u0026rdquo;. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975; 12:189\u0026ndash;98. \u003c/li\u003e\n\u003cli\u003eLiu X, Song F, Liu F, Mao Z, Qu S. Multiple chronic conditions among older adults in China: differences in socio-demographic characteristics. Heliyon. 2022;8:e11129. \u003c/li\u003e\n\u003cli\u003eJia L, Du Y, Chu L, Zhang Z, Li F, Lyu D, et al. Prevalence, risk factors, and management of dementia and mild cognitive impairment in adults aged 60 years or older in China: a cross-sectional study. Lancet Public Health. 2020;5: e661\u0026ndash;71. \u003c/li\u003e\n\u003cli\u003eAllegri RF, Butman J, Arizaga RL, Machnicki G, Serrano C, Taragano FE, et al. Economic impact of dementia in developing countries: an evaluation of costs of Alzheimer-type dementia in Argentina. Int Psychogeriatr. 2007; 19:705\u0026ndash;18. \u003c/li\u003e\n\u003cli\u003eKourtis LC, Regele OB, Wright JM, Jones GB. Digital biomarkers for Alzheimer\u0026rsquo;s disease: the mobile/ wearable devices opportunity. NPJ Digit Med. 2019; 2:9. \u003c/li\u003e\n\u003cli\u003eBoxer AL, Garbutt S, Seeley WW, Jafari A, Heuer HW, Mirsky J, et al. Saccade abnormalities in autopsy-confirmed frontotemporal lobar degeneration and Alzheimer disease. Arch Neurol. 2012; 69:509\u0026ndash;17. \u003c/li\u003e\n\u003cli\u003eLage C, L\u0026oacute;pez-Garc\u0026iacute;a S, Bejanin A, Kazimierczak M, Aracil-Bola\u0026ntilde;os I, Calvo-C\u0026oacute;rdoba A, et al. Distinctive Oculomotor Behaviors in Alzheimer\u0026rsquo;s Disease and Frontotemporal Dementia. Front Aging Neurosci. 2020; 12:603790. \u003c/li\u003e\n\u003cli\u003eTokushige S-I, Matsumoto H, Matsuda S-I, Inomata-Terada S, Kotsuki N, Hamada M, et al. Early detection of cognitive decline in Alzheimer\u0026rsquo;s disease using eye tracking. Front Aging Neurosci. 2023; 15:1123456. \u003c/li\u003e\n\u003cli\u003eXu Y, Zhang C, Pan B, Yuan Q, Zhang X. A portable and efficient dementia screening tool using eye tracking machine learning and virtual reality. npj Digit Med. 2024; 7:219. \u003c/li\u003e\n\u003cli\u003eChehrehnegar N, Shati M, Esmaeili M, Foroughan M. Executive function deficits in mild cognitive impairment: evidence from saccade tasks. Aging Ment Health. 2022; 26:1001\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eJiang Y, Cui M, Tian W, Zhu S, Chen J, Suo C, et al. Lifestyle, multi‐omics features, and preclinical dementia among Chinese: The Taizhou Imaging Study. Alzheimer\u0026rsquo;s Dement. 2021; 17:18\u0026ndash;28. \u003c/li\u003e\n\u003cli\u003eKatzman R, Zhang MY, Ouang-Ya-Qu, Wang ZY, Liu WT, Yu E, et al. A Chinese version of the Mini-Mental State Examination; impact of illiteracy in a Shanghai dementia survey. J Clin Epidemiol. 1988; 41:971\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eYu J, Li J, Huang X. The Beijing version of the Montreal Cognitive Assessment as a brief screening tool for mild cognitive impairment: a community-based study. BMC Psychiatry. 2012; 12:156. \u003c/li\u003e\n\u003cli\u003eZhao Q, Lv Y, Zhou Y, Hong Z, Guo Q. Short-term delayed recall of auditory verbal learning test is equivalent to long-term delayed recall for identifying amnestic mild cognitive impairment. PLoS One. 2012;7: e51157. \u003c/li\u003e\n\u003cli\u003eZhao Q, Guo Q, Li F, Zhou Y, Wang B, Hong Z. The Shape Trail Test: application of a new variant of the Trail making test. PLoS One. 2013;8: e57333. \u003c/li\u003e\n\u003cli\u003eZhao Q, Guo Q, Hong Z. Clustering and switching during a semantic verbal fluency test contribute to differential diagnosis of cognitive impairment. Neurosci Bull. 2013; 29:75\u0026ndash;82. \u003c/li\u003e\n\u003cli\u003eZhang Y, Xu J, Zhang C, Zhang X, Yuan X, Ni W, et al. Community screening for dementia among older adults in China: a machine learning-based strategy. BMC Public Health. 2024; 24:1206. \u003c/li\u003e\n\u003cli\u003eFirst MB. Diagnostic and Statistical Manual of Mental Disorders, 5th Edition, and Clinical Utility. J Nerv Ment Dis. 2013; 201:727\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eJack CR, Bennett DA, Blennow K, Carrillo MC, Dunn B, Haeberlein SB, et al. NIA-AA Research Framework: Toward a biological definition of Alzheimer\u0026rsquo;s disease. Alzheimer\u0026rsquo;s Dement. 2018; 14:535\u0026ndash;62. \u003c/li\u003e\n\u003cli\u003eHobden G, Colbourne E, Pendlebury ST, Demeyere N. Reliability of the global cortical atrophy visual rating scale applied to computed tomography versus magnetic resonance imaging scans in acute stroke. Neurol Sci. 2024; 45:1549\u0026ndash;56. \u003c/li\u003e\n\u003cli\u003eZhang D, Zhang W, Ming C, Gao X, Yuan H, Lin X, et al. P-tau217 correlates with neurodegeneration in Alzheimer\u0026rsquo;s disease, and targeting p-tau217 with immunotherapy ameliorates murine tauopathy. Neuron. 2024; 112:1676-1693.e12. \u003c/li\u003e\n\u003cli\u003eLu M, Hedin LO. Global plant-symbiont organization and emergence of biogeochemical cycles resolved by evolution-based trait modelling. Nat Ecol Evol. 2019; 3:239\u0026ndash;50. \u003c/li\u003e\n\u003cli\u003eKrafka K, Khosla A, Kellnhofer P, Kannan H, Bhandarkar S, Matusik W, et al. Eye Tracking for Everyone. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) [Internet]. Las Vegas, NV, USA: IEEE; 2016 [cited 2025 Feb 18]. p. 2176\u0026ndash;84. Available from: http://ieeexplore.ieee.org/document/7780608/\u003c/li\u003e\n\u003cli\u003eLiu G, Yu Y, Mora KAF, Odobez J-M. A Differential Approach for Gaze Estimation. IEEE Trans Pattern Anal Mach Intell. 2021; 43:1092\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eKragel JE, Voss JL. Looking for the neural basis of memory. Trends Cogn Sci. 2022; 26:53\u0026ndash;65. \u003c/li\u003e\n\u003cli\u003eLiu X, Cheng Z, Lin H, Tan J, Chen W, Bao Y, et al. Decoding effects of psychoactive drugs in a high-dimensional space of eye movements in monkeys. Natl Sci Rev. 2023;10: 255. \u003c/li\u003e\n\u003cli\u003eKissler J, Clementz BA. Fixation stability among schizophrenia patients. Neuropsychobiology. 1998; 38:57\u0026ndash;62. \u003c/li\u003e\n\u003cli\u003eCamchong J, Dyckman KA, Chapman CE, Yanasak NE, McDowell JE. Basal ganglia-thalamocortical circuitry disruptions in schizophrenia during delayed response tasks. Biol Psychiatry. 2006; 60:235\u0026ndash;41. \u003c/li\u003e\n\u003cli\u003eHeuer HW, Mirsky JB, Kong EL, Dickerson BC, Miller BL, Kramer JH, et al. Antisaccade task reflects cortical involvement in mild cognitive impairment. Neurology. 2013; 81:1235\u0026ndash;43. \u003c/li\u003e\n\u003cli\u003eZola SM, Squire LR, Teng E, Stefanacci L, Buffalo EA, Clark RE. Impaired recognition memory in monkeys after damage limited to the hippocampal region. J Neurosci. 2000; 20:451\u0026ndash;63. \u003c/li\u003e\n\u003cli\u003eMueller S, Chao L, Berman B, Weiner M. Evidence for functional specialization of hippocampal subfields detected by MR subfield volumetry on high resolution images at 4T. Neuroimage. 2011; 56:851\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eWilcockson TDW, Mardanbegi D, Xia B, Taylor S, Sawyer P, Gellersen HW, et al. Abnormalities of saccadic eye movements in dementia due to Alzheimer\u0026rsquo;s disease and mild cognitive impairment. Aging. 2019; 11:5389\u0026ndash;98. \u003c/li\u003e\n\u003cli\u003eAverbeck BB, O\u0026rsquo;Sullivan SS, Djamshidian A. Impulsive and Compulsive Behaviours in Parkinson\u0026rsquo;s disease. Annu Rev Clin Psycho. 2014; 10:553\u0026ndash;80. \u003c/li\u003e\n\u003cli\u003eBeckner AG, Arnold CD, Bragg MG, Caswell BL, Chen Z, Cox K, et al. Examining infants\u0026rsquo; visual paired comparison performance in the US and rural Malawi. Dev Sci. 2024;27: e13439. \u003c/li\u003e\n\u003cli\u003eBoujelbane MA, Trabelsi K, Salem A, Ammar A, Glenn JM, Boukhris O, et al. Eye Tracking During Visual Paired-Comparison Tasks: A Systematic Review and Meta-Analysis of the Diagnostic Test Accuracy for Detecting Cognitive Decline. Journal of Alzheimer\u0026rsquo;s Disease. 2024; 99:207\u0026ndash;21. \u003c/li\u003e\n\u003cli\u003eOpwonya J, Wang C, Jang K-M, Lee K, Kim JI, Kim JU. Inhibitory Control of Saccadic Eye Movements and Cognitive Impairment in Mild Cognitive Impairment. Front Aging Neurosci. 2022; 14:871432. \u003c/li\u003e\n\u003cli\u003eNij Bijvank JA, Strijbis EMM, Nauta IM, Kulik SD, Balk LJ, Stam CJ, et al. Impaired saccadic eye movements in multiple sclerosis are related to altered functional connectivity of the oculomotor brain network. Neuroimage Clin. 2021; 32:102848. \u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Bryant Sid E, Zhang Fan, Petersen Melissa, Hall James R, Johnson Leigh A, Yaffe Kristine, et al. A blood screening tool for detecting mild cognitive impairment and Alzheimer\u0026rsquo;s disease among community-dwelling Mexican Americans and non-Hispanic Whites: A method for increasing representation of diverse populations in clinical research. Alzheimer\u0026rsquo;s dement : j Alzheimer\u0026rsquo;s Assoc [Internet]. 2021;18. Available from: https://kns.cnki.net/kcms2/article/abstract?v=WnCf0VAm38qAhkKNKxCqX8bWBamJbVUxXi6cNJbtJmViVZc4-oEpWno6w5mcdcAEQhxdqvVJoucrw-8Lc5qlJZ5mLuuA7TpgQy0CZpdHIYev-jkirU9WV36kctMpECePrI6Ao8oxSsma-nBAr_5AXA==\u0026amp;uniplatform=NZKPT\u0026amp;language=gb\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table 1","content":"\u003cp\u003e\u003cstrong\u003eTable 1. Characteristics\u0026nbsp;of the discovery dataset and the external validation cohort\u003c/strong\u003e\u003c/p\u003e \n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 217px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 359px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDiscovery dataset (n=204)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 353px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExternal validation cohort (n=433)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 217px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003eNC (n=106)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003eDementia (n=98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003eNC (n=315)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003eCI (n=118)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e-value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eAge, years, mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e66.8(7.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e68.2(8.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e0.112\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e67.9(3.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e68.6(3.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e0.080\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eFemale, No. (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e64(60.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e55(56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e0.538\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e153(48.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e74(62.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e0.009\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eEducation, No. (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eIlliteracy and primary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e6(5.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e14(14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e182(57.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e90(76.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eJunior\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e34(32.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e51(52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e91(28.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e22(18.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eSenior and above\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e66(62.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e33(34)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e42(13.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e6(5.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eMMSE, median (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e28(2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e18(6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e25(5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e17(10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eMoCA, median (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e25(5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e13(6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e17(8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e10(7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eCentral (mean),\u0026nbsp;mean (SD), %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e0.75(0.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e0.54(0.10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e0.64(0.12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e0.54(0.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eFix duration valid (mean),\u0026nbsp;mean (SD),\u0026nbsp;ms\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e4517.58(1254.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e4925.16(963.63)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e0.017\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e5749.16(746.10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e5788.13(679.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e0.907\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eAccuracy, mean (SD), %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e0.53(0.21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e0.38(0.23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e0.53(0.21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e0.39(0.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eCorrection latency (mean),\u0026nbsp;mean (SD),\u0026nbsp;ms\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e241.23(162.96)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e337.07(167.61)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e307.66(116.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e400.87(143.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ePeak velocity (mean),\u0026nbsp;mean (SD),\u0026nbsp;pixel/ms\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e4.47(1.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e4.77(1.58)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e0.137\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e4.28(1.46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e3.90(1.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e0.013\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eFinal ratio (mean), mean (SD), %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e1.51(5.52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\n \u003cp\u003e1.40(2.82)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003e6.04(46.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e1.33(2.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAbbreviations: CI, cognitive impairment; IQR, interquartile range; MMSE, Mini Mental State Examination; MoCA, Montreal Cognitive Assessment; NC, normal cognition; SD, standard deviation.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"alzheimers-research-and-therapy","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"azrt","sideBox":"Learn more about [Alzheimer's Research and Therapy](http://alzres.biomedcentral.com/)","snPcode":"13195","submissionUrl":"https://submission.nature.com/new-submission/13195/3","title":"Alzheimer's Research \u0026 Therapy","twitterHandle":"@AlzheimersRes","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7141899/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7141899/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eCognitive impairment (CI) screening remains critically inaccessible in resource-limited Chinese communities.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eWe developed a tablet-based Mobile Eye-Tracking Application (m-ETA) using a three-step approach. A dementia discrimination model based on six oculometric features was trained in a hospital cohort (N\u0026thinsp;=\u0026thinsp;204) and validated for biological relevance with Alzheimer's biomarkers (N\u0026thinsp;=\u0026thinsp;101). Generalizability and accuracy were further assessed in a community cohort (N\u0026thinsp;=\u0026thinsp;433) and two real-world populations (N\u0026thinsp;=\u0026thinsp;2,685).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003em-ETA achieved high diagnostic accuracy for dementia (AUC\u0026thinsp;=\u0026thinsp;0.99). The oculometric features were significantly associated with cognitive performance, brain atrophy, and tau deposition (all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05). m-ETA accurately detected CI (AUC\u0026thinsp;=\u0026thinsp;0.80), with excellent negative predictive value for ruling out CI, and identified individuals with lower cognition performance across diverse communities.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003em-ETA offers a low-cost, non-invasive, and efficient tool for large-scale CI screening, particularly suited to underserved and low-literacy communities in China.\u003c/p\u003e","manuscriptTitle":"An accessible and efficient mobile eye-tracking application for community-based cognitive impairment screening in China","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-27 10:21:14","doi":"10.21203/rs.3.rs-7141899/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-12T03:50:25+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-29T09:09:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"218255686350995088700896368323013682636","date":"2025-08-26T17:50:47+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"167883703855968250078437475002504731087","date":"2025-08-25T06:43:10+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-24T17:35:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"283887594864095762290223897421172765575","date":"2025-07-24T15:29:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"91388937236043014147571766322950279587","date":"2025-07-24T07:42:18+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-07-24T05:28:36+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-17T03:27:49+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-07-17T03:26:21+00:00","index":"","fulltext":""},{"type":"submitted","content":"Alzheimer's Research \u0026 Therapy","date":"2025-07-16T16:00:54+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"alzheimers-research-and-therapy","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"azrt","sideBox":"Learn more about [Alzheimer's Research and Therapy](http://alzres.biomedcentral.com/)","snPcode":"13195","submissionUrl":"https://submission.nature.com/new-submission/13195/3","title":"Alzheimer's Research \u0026 Therapy","twitterHandle":"@AlzheimersRes","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"59173859-d810-457d-ab2b-71209a3f68ed","owner":[],"postedDate":"July 27th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-24T15:59:30+00:00","versionOfRecord":{"articleIdentity":"rs-7141899","link":"https://doi.org/10.1186/s13195-025-01884-7","journal":{"identity":"alzheimers-research-and-therapy","isVorOnly":false,"title":"Alzheimer's Research \u0026 Therapy"},"publishedOn":"2025-11-17 15:56:50","publishedOnDateReadable":"November 17th, 2025"},"versionCreatedAt":"2025-07-27 10:21:14","video":"","vorDoi":"10.1186/s13195-025-01884-7","vorDoiUrl":"https://doi.org/10.1186/s13195-025-01884-7","workflowStages":[]},"version":"v1","identity":"rs-7141899","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7141899","identity":"rs-7141899","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}