The VR Eye-tracking Cognitive Assessment (VECA): A Portable and Efficient Dementia Screening Tool Using Eye-Tracking Technology, Machine Learning, and Virtual Reality

The VR Eye-tracking Cognitive Assessment (VECA): A Portable and Efficient Dementia Screening Tool Using Eye-Tracking Technology, Machine Learning, and Virtual Reality | 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 Article The VR Eye-tracking Cognitive Assessment (VECA): A Portable and Efficient Dementia Screening Tool Using Eye-Tracking Technology, Machine Learning, and Virtual Reality Qing Yuan, Ying Xu, Xu Zhang, Chi Zhang, Baobao Pan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3828765/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 22 Aug, 2024 Read the published version in npj Digital Medicine → Version 1 posted 11 You are reading this latest preprint version Abstract Background: Dementia is a significant global health challenge, and early screening during the preclinical stage is crucial. However, current diagnostic biomarkers for Alzheimer's Disease, the most common cause of dementia, have limitations in terms of cost and invasiveness. Mild cognitive impairment (MCI) is recognized as a transitional stage preceding dementia. Whileneuropsychological tests like the Montreal Cognitive Assessment (MoCA) are effective for identifying MCI, they are not suitable for large-scale dementia screening. Eye-tracking technology has emerged as a promising tool for cognitive assessment by capturing and quantifying eye movements related to cognitive behavior. Subtle changes in eye movements can potentially serve as biomarkers for early MCI identification. However, interpreting the vast amount of eye-tracking data poses challenges. To address this, machine learning methods in computer science can be applied to analyze eye-tracking data and identify patterns or abnormalities indicative of MCI. Machine learning models trained on large datasets can improve the accuracy and efficiency of MCI identification. Additionally, the immersive nature of virtual reality (VR) technology allows for uninterrupted eye-tracking processes, while the portability of VR head-mounted devices enables efficient and large-scale early screening for community cognitive impairment. Objective: Develop a dementia screening tool, VR Eye-tracking Cognitive Assessment (VECA), using eye-tracking technology, machine learning, and virtual reality as an alternative to traditional neuropsychological tests. This tool aims to help physicians detect cognitive impairment, particularly in the early stages of MCI, on a larger scale. Methods: 201 subjects from Shenzhen Baoan Chronic Hospital were administered MoCA test and VECA. Raw gaze data were captured by the eye tracker of the VR headset and filtered as eye movements which would be encoded as features. Machine learning models were established as the predictor of MoCA score and the classifier of cognitive impairment of three education-based groups within which optimal cut-off score was given. The study has been approved by Shenzhen Baoan Chronic Hospital Ethics Committee and all subjects have signed written informed consent for participation. Results: Support vector regression was proposed as the VR-AI model and achieved high correlation of 0.9 with MoCA score, greater than baseline model of 0.58. Optimal cut-off scores (less than 6 years of education: 14/15; 6 to 9 years of education: 18/19; more than 9 years of education: 23/24) can well distinguish normal and cognitive impairment subjects -with overall sensitivity of 88.5% and specificity of 83%. Conclusion: VECA is a portable and efficient dementia screening tool that utilizes eye-tracking technology, machine learning, and virtual reality. It offers a quantitative approach for large-scale early screening of dementia. Health sciences/Neurology/Neurological disorders/Dementia/Alzheimer's disease Biological sciences/Neuroscience/Cognitive neuroscience Cognitive impairment eye tracking early Screening machine learning (ML) virtual reality Dementia - Alzheimer disease Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Dementia refers to a spectrum of neurodegenerative syndromes characterized by progressive disturbances in various cognitive functions severe enough to interfere with the patient’s activities of daily living. The rapid increase in the number of patients with dementia has become a global health challenge, with Alzheimer’s disease (AD) being the most common, accounting for 60-70% [1]. Mild cognitive impairment (MCI) refers to the progressive decline of memory or other cognitive functions but does not affect the ability of daily life as dementia [2]. The increasing prevalence of cognitive impairment has caused enormous economic loss [3, 4]. Accumulating evidence shows that early diagnosis and timely intervention can delay cognitive decline [5–7], so early screening in the preclinical stage of dementia is necessary. Currently, cerebrospinal fluid β -amyloid (amyloid β , A β ) and tau, amyloid positron emission tomography (PET) and AD pathogenic gene carrying are diagnostic markers for AD [8]. Though accurate, these approaches could hardly perform as an early screening tool for high costs and surgical invasiveness. Dementia is essentially cognitive dysfunction, so assessment of cognition is an important part of the diagnosis. Neuropsychological paper-based instruments, such as the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA) are commonly used, which cover orientation, memory, attention, calculation, language ability, visuospatial ability, etc. Though valid and reliable, traditional neuropsychological tests must be administered by trained physicians, and are neither simple nor efficient enough to serve as large-scale dementia screening tools. Moreover, neuropsychological assessment results are affected by both the physician’s subjective judgments and the testing environment. At present, there is no efficient solution for large-scale dementia screening. Cognitive-related uses of eye-tracking technology have been exploded in recent years, enabling online cognitive activity to be recorded, physicians to connect learning outcomes to the cognitive processes of subjects [9], and detection of emotional and cognitive states [10]. Eye tracking is a real-time sensor technology that measures gaze points and eye movements. A large number of studies have shown that detailed and diverse eye movements derived by algorithms from eye trackers made quantitative and workforce-saving cognitive assessment feasible [11–19]. Recent studies have applied VR head-mounted display equipment and VR cognitive assessment tasks to assess cognitive function, which can distinguish patients with mild cognitive impairment from low-risk and high-risk patients with AD [20,21]. The combination of eye-tracking technology and VR, with the portability of VR head-mounted display equipment and VR immersive environment, is expected to be applied to large-scale cognitive impairment screening in multiple scenarios. However, two challenging issues have to be handled. First, these technologies should be integrated in a smart way so that efficiency is guaranteed. Moreover, subtle changes in eye movements can potentially serve as biomarkers for early MCI identification，interpreting the vast amount of eye-tracking data poses challenges. In this study, based on eye-tracking technology, machine learning, and virtual reality, we designed and developed an efficient, portable, and quantitative early screening tool for dementia. This project manages to promote the early detection of people with cognitive impairment, delay the progress of dementia and will help greatly reduce the spiritual, nursing, and economic burdens on both individuals and society caused by dementia. Materials and Methods Participants This study is conducted with the cooperation of Shenzhen Baoan Chronic Hospital，National Engineering laboratory for Big Data System Computing Technology, Shenzhen university, China, and Yiwei Medical Technology Co., Ltd. A total of 201 subjects were evaluated, including 88 from Fuan CRC (Community Rehabilitation Center) and 113 from Wanxiang CRC. Some inclusions of participants were: (1) aged 65.5±5.1 years; (2) 81 males (40.3%), female = 120 (59.7%); (3) years of education 9.4±3.8 (Table 1). The study excluded participants who had psychiatric, ophthalmological, or hearing impairment disorders, as well as those who were unable to sit comfortably due to severe physical illness. Only participants with normal vision or corrected-to-normal vision without color blindness were included. A small number of participants were further excluded from the analysis if they encountered difficulties calibrating to the eye tracker or did not attempt to view the images. Exclusion criteria were applied in cases where the eye tracking equipment faced challenges in achieving proper pupil and corneal reflection due to physiological constraints or visual problems (e.g., droopy eyelid, cataracts, detached retinas, glaucoma, pupils too small), or if participants were unable to complete the eye tracking calibration procedure. Assessment Data acquisition Demographic characteristics including age, gender, and education levels of the participants were collected and each of them underwent paper-based instruments of MoCA Chinese version. Figure 1 depicts the cognitive assessment system based on VR head-mounted display and eye-tracking technology used in the study, collectively referred to as VECA, which takes 5 minutes only. Eye tracking calibration would be fulfilled before participants started the assessment to make sure gaze points captured by the eye tracker are valid. Cognitive tasks include memory (encoding, storage, and recall), visual attention (smooth pursuit), abstraction ability (ability to form abstract concepts), visuospatial working memory, visuospatial and executive function, calculation, language (understanding and execution of commands), and short-term memory binding. Detailed descriptions of each task are given in appendix (A1-A11). In each task, participants were usually given 3 seconds to understand the task and 5 or 8 seconds to fulfill related cognitive task. Multiple images including a correct answer (Area of interest, AOI) and distractors (non-Area of interest, non-AOI) were displayed in the VR scene, and subjects were instructed to identify and fixate on the correct answer. All eye movements were captured by the Tobii eye tracker embedded in Pico neo 3 pro eye, 95% of users can track better than 2° when their eyes are looking straight ahead (https://developer.tobii.com/xr/learn/eye-behavior/hardware-accuracy/). Data processing Biological eye movements such as fixation and saccade were derived out of raw gaze points with I-VT algorithm, which automatically classifies excessive noise such as microsaccades as abnormal movements [22]. Gaze points that extremely deviate from their neighbors may be caused by winks or unexpected shivers and were also eliminated. The percentage of time the subject fixates on the AOI over total fixation time of a certain task is calculated and deemed as features. Previous research claimed that age is the biggest risk factor and made gender also a significant risk factor due to longer lifespan of female [23]. Therefore, demographic information such as age and gender of the subjects were supplemented to the feature set. Education level, as in Figure 2 , demonstrated feature importance of nearly a quarter and would be used as a classifying attribute in the later discussion. Data preprocessing techniques such as one-hot and ordinal encoder were applied to categorical features. Standard scaling was also conducted for scale-sensitive models to eliminate the influence of various scales. MoCA scores were categorized as the target variable so that our system could play as an efficient screening tool. In this way, the target variable better quantified the subjects’ cognitive capability. Modeling Oyama et al. averaged each cognitive task score as the eye movement cognitive assessment score [17] (hereinafter referred to as VR model ) which would be the baseline model. Support Vector Regression (SVR), Multi-layer Perceptron (MLP), Lasso regression (Lasso), and Gradient Boost Regression Tree (GBRT) were adopted. The model performance was evaluated using the following metrics: Median Absolute Error (Median AE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Correlation with MoCA score. The model with the best performance was selected to predict paper-based MoCA score (hereinafter referred to as VR-AI model ) which would be the numerical cognitive indicator of the screening result. Results VR-AI model selection Optimal parameters of candidate paper based MoCA score predictors were found through grid search. Each model underwent 5-fold cross-validation and compared in the following metrics ( Table 2 ). MAE and RMSE are common metrics which are widely used to describe regression performance. However, there was unpredictable noise in the target value due to human factors and lead to extreme bad cases in both train and test set. Instead, median AE could indicate model performance in a more stable way disregard of outliers. Moreover, absolute error shows more intuitive difference between predictions and labels. Therefore, MAE was weighed more as the metrics to compare models. Higher correlation coefficient between predictions and targets indicates better performance of screening ( Figure 3 ). SVR overperformed other models and baseline under most of our metrics, especially under average median absolute error of 2.04 and correlation coefficient as high as 0.9 and was selected as VR-AI model. Screening evaluation Lu et al. proposed a Chinese version of MoCA cutting-off scores which is more adaptive to the Chinese population due to its geographically various languages, education, and local culture, which had long been the gold standard in China [24]. Therefore, we classified participants into three groups according to their education levels. Participants were grouped and labeled by Chinese standards as follows: Group 1: 0-6 years of education: normal cognition (MoCA score 14-30), n=50; cognitive impairment (MoCA score 0-13), n=11. Group 2: 6-9 years of education: normal cognition (MoCA score 20-30), n=44; cognitive impairment (MoCA score 0-19) n=19. Group 3: education level greater than 9 years: normal cognition (MoCA score 25-30), n=47; cognitive impairment (MoCA score 0-24), n=31. Classification performance of VR-AI model was evaluated due to ROC curves. Optimal thresholds of VR- AI score were given by F-β method which, in screening scenario, weighed more on sensitivity. The whole flowchart of the proposed model is shown in Figure 4 . The screening performance was also evaluated in classification aspect. The subjects were grouped based on education levels, and the AUCs of three groups were 0.92(95% CI: 0.85-0.99), 0.95(95% CI: 0.89-1.00), and 0.94(95% CI: 0.89-0.99) respectively ( Figure 5 ). The optimal screening cut-off scores for each group are given by F-β score: less than 6 years of education, 13/14; 6 to 9 years of education, 21/22; more than 9 years of education, 23/24. The VR-AI model showed excellent sensitivities of 81.8%, 94.7%, and 83.8% in screening for cognitive impairment and specificities of 98%, 91.1%, and 73.6%. The sensitivity of the overall screening of cognitive impairment and cognitively normal subjects can reach 88.5% with specificity of 83%. Discussion Eye movements were highly informative indicators of cognitive function. In our study, together with basic information—gender, age and education, encoded eye movements can excellently classify normal and cognitive impairment subjects within education-based groups via machine learning. As a result, we developed an efficient screening tool to identify subjects that would be labeled cognitive impairment by traditional paper-based instruments. The crucial part is regression of MoCA score. SVM regression outperformed and achieved absolute error of 2.04 and strong correlation with target variable. An important application of our screening tool is providing diagnosis suggestions. The AUCs of three education-based groups were 0.92(95% CI: 0.85-0.99), 0.95(95% CI: 0.89-1.00), and 0.94(95% CI: 0.89-0.99) respectively. Proposed optimal cut-off scores for each group achieve sensitivities of 81.8%, 94.7%, and 83.8% in screening for cognitive impairment and specificities of 98%, 91.1%, and 73.6%. To retrospect the sample size of our research, the estimate of the sample size was given according to the general requirements of diagnostic research: 1) screening test sensitivity (Sen)≥ 0.5, sensitivity tolerance error 0.10; 2) The specificity (Spe) of screening test was ≥ 0.5, the tolerance error of specificity was 0.10. 3) Significance test level two-sided α = 0.05. The results are presented in the table below. The sample size of the three groups (Group1 N=61, group2 N=63, group3=78) in this study can meet the requirements of statistics. There are various cognitive assessment and diagnostic methods available, including neuropsychological testing, imaging, and laboratory examinations. Imaging examinations such as CT/MRI/PET-CT provide objective results but expose individuals to radiation, are not suitable for early screening, and are costly. Laboratory examinations such as cerebrospinal fluid testing are invasive but provide objective results. For more than a decade, MoCA has been widely applied in screening and diagnosing dementia either in paper or electronic form. However, besides 30 minutes of duration, cultural and educational influences may make it unsuitable for certain populations. Screening results may be affected by the emotional and mental state of the test subjects as well as how professional psychiatrists are. Moreover, comparison of common diagnosis or screening approaches from the aspect of convenience and efficiency was given in Table 4 . In contrast, the immersive environment of VECA ensures data integrity and reliability, minimizing environmental impact on test subjects. Multiple dimensions of assessment and objective results can be provided. VECA is less time-consumable (5-minute test) and fully self-administered and its portability makes large-scale cognitive screening possible and efficient especially in fundamental clinics and communities. Limitations Despite of satisfying screening performance, there still exist potential bias and limitations. First, participants were from communities of Shenzhen, information on other variables, e.g. comorbidities, functional dependence, eventually drugs, could not be collected. Both will cause bias in participants selection and make our results valid only within certain cohort. Comprehensive data should be collected in future studies to fully consider the influencing factors that affect the evaluation results. Since both train and test cohort were selected within limited samples, lack of external validation should also be considered when applying VECA. Furthermore, the interactions between subjects and VR headset are based on visual and literal capabilities. Lack of either one will cause the system to misinterpret the subjects’ eye movements and produce deviated predictions. On the machine learning level, validity and authenticity of the target variable (MoCA) are affected by the psychiatrists. Excessive noise in MoCA can distort our model's distribution and compromise its accuracy. To improve model performance, other informative features could be extracted from raw eye movements. Their validation and integration into the system should be studied in the future. In practical applications, using multiple methods for screening cognitive impairment can improve diagnostic accuracy and reduce the risk of missed diagnoses and misdiagnoses. Conclusion This research suggests that VECA can be an efficient and portable tool for dementia screening. VECA has the potential to improve early diagnosis and treatment of dementia by helping physicians identify mild cognitive impairment in the early stages. This study highlights the application of eye-tracking technology, machine learning, and virtual reality in cognitive evaluation. Further research with more data is needed to compare the effectiveness of VECA with traditional tests and dementia diagnosis outcomes. Declarations Conflict of interest statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Author contribution statement Qing Yuan: Organizer of the whole project. Responsible for screening scenes, clinical suggestions, and other resources. Propose initial ideas and other clinical details. Xu Zhang: Designing the content for experiments and VR eye-tracking cognitive tests, providing professional digital medical guidance from a perspective of neuroscience science, author of the initial manuscript. Ying Xu: Director of participants, data acquisition, and reviewer of data quality. Chi Zhang: Director of data processing, feature extraction, and modeling. Author of the complete manuscript. Baobao Pan: Provider of algorithm suggestions. Contribution to the field Dementia is a growing global health challenge, and early screening is crucial in identifying patients with mild cognitive impairment (MCI) as a transitional stage preceding dementia. Despite various diagnostic markers available, the current options are limited by cost and invasiveness. Neuropsychological tests have been used, but they are neither simple nor efficient enough to serve as large-scale dementia screening tools. Our study proposes the use of eye-tracking data as a quantitative and multi-dimensional approach to assess cognitive function. We developed a VR-based tool that guarantees the integrity of eye-tracking data and enables efficient large-scale early screening of cognitive impairment. Our proposed machine learning models achieved a high correlation with the MoCA score - a commonly used tool for identifying MCI - and optimal cut-off scores were established for distinguishing normal and cognitive impairment subjects. This method avoids invasiveness, lower workforce, time cost, and may apply to large-scale screening well. Funding information This work was supported by Shenzhen Fundamental Research Program (JCYJ20220531091006014) Ethics statements Studies involving animal subjects. Generated Statement: No animal studies are presented in this manuscript. Studies involving human subjects. Generated Statement: The studies involving human participants were reviewed and approved by Shenzhen Baoan Chronic Hospital. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. Inclusion of identifiable human data Generated Statement: No potentially identifiable human images or data is presented in this study. References WHO. Dementia. 2021; Available from: https://www.who.int/news-room/fact-sheets/detail/dementia Petersen, R.C., Clinical practice. Mild cognitive impairment. The New England journal of medicine, 2011. 364(23): p. 2227-2234. Jia, L., 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. The Lancet Public Health, 2020. 5(12):p. e661-e671. Jia, J., et al., The cost of Alzheimer’s disease in China and re-estimation of costs worldwide. Alzheimers Dement, 2018. 14(4): p. 483-491. Blondell, S.J., R. Hammersley-Mather, and J.L. Veerman, Does physical activity prevent cognitive decline and dementia?: A systematic review and meta-analysis of longitudinal studies. BMC public health, 2014. 14(1): p. 1-12. Ngandu, T., et al., A 2 year multidomain intervention of diet, exercise, cognitive training, and vascu- lar risk monitoring versus control to prevent cognitive decline in at-risk elderly people (FINGER): a randomised controlled trial. The Lancet, 2015. 385(9984): p. 2255-2263. Park, H., et al., Combined Intervention of Physical Activity, Aerobic Exercise, and Cognitive Exercise Intervention to Prevent Cognitive Decline for Patients with Mild Cognitive Impairment: A Randomized Controlled Clinical Study. J Clin Med, 2019. 8(7). Dubois, B., et al., Advancing research diagnostic criteria for Alzheimer’s disease: the IWG-2 criteria. The Lancet Neurology, 2014. 13(6): p. 614-629. Lai, M.-L., et al., A review of using eye-tracking technology in exploring learning from 2000 to 2012. Educational Research Review, 2013. 10: p. 90-115. Skaramagkas, V., et al., Review of eye tracking metrics involved in emotional and cognitive processes. IEEE Rev Biomed Eng, 2021. PP. Lagun, D., et al., Detecting cognitive impairment by eye movement analysis using automatic classifica- tion algorithms. Journal of neuroscience methods, 2011. 201(1): p. 196-203. Nie, J., et al., Early Diagnosis of Mild Cognitive Impairment Based on Eye Movement Parameters in an Aging Chinese Population. Front Aging Neurosci, 2020. 12: p. 221. Haque, R.U., et al., VisMET: a passive, efficient, and sensitive assessment of visuospatial memory in healthy aging, mild cognitive impairment, and Alzheimer’s disease. Learn Mem, 2019. 26(3): p. 93-100. Pereira, M., et al., Visual Search Efficiency in Mild Cognitive Impairment and Alzheimer’s Disease: An Eye Movement Study. J Alzheimers Dis, 2020. 75(1): p. 261-275. Jiang, J., et al., A Novel Detection Tool for Mild Cognitive Impairment Patients Based on Eye Movement and Electroencephalogram. J Alzheimers Dis, 2019. 72(2): p. 389-399. Chehrehnegar, N., et al., Executive function deficits in mild cognitive impairment: evidence from saccade tasks. Aging & mental health, 2021: p. 1-9. Oyama, A., et al., Novel Method for Rapid Assessment of Cognitive Impairment Using High-Performance Eye-Tracking Technology. Sci Rep, 2019. 9(1): p. 12932. Mengoudi, K., et al., Augmenting dementia cognitive assessment with instruction-less eye-tracking tests. IEEE journal of biomedical and health informatics, 2020. 24(11): p. 3066-3075. Parra, M.A., M. Schumacher, and G. Fern´andez, A novel peripheral biomarker for mild cognitive im- pairment and Alzheimer’s disease. Alzheimer’s & Dementia, 2020. 16(S4). Howett, D., et al., Differentiation of mild cognitive impairment using an entorhinal cortex-based test of virtual reality navigation. Brain, 2019. 142(6): p. 1751-1766. Corriveau Lecavalier, N., et al., Use of immersive virtual reality to assess episodic memory: A validation study in older adults. Neuropsychol Rehabil, 2020. 30(3): p. 462-480. Salvucci, D. D., & Goldberg, J. H. (2000, November). Identifying fixations and saccades in eye-tracking protocols. In Proceedings of the 2000 symposium on Eye tracking research & applications (pp. 71-78). Beam CR, Kaneshiro C, Jang JY, Reynolds CA, Pedersen NL, Gatz M. Differences Between Women and Men in Incidence Rates of Dementia and Alzheimer’s Disease. J Alzheimers Dis. 2018;64(4):1077-1083. doi: 10.3233/JAD-180141. PMID: 30010124; PMCID: PMC6226313. Lu J, Li D, Li F, et al. Montreal cognitive assessment in detecting cognitive impairment in Chinese elderly individuals: a population-based study[J]. Journal of geriatric psychiatry and neurology, 2011, 24(4): 184-190. Tables Table 1: Demographic characteristics of participants Attribute Count (N) Pct. Data Source Fuan CRC 88 43.8% Wanxiang CRC 113 56.2% Age 55-65 113 56.2% 65-75 79 39.3% > 75 9 4.5% Gender Male 81 40.3% Female 120 59.3% Education Level Illiterate 8 3.9% Primary School 53 26.4% Junior High School 62 30.8% High School 55 27.4% Bachelor or above 23 11.4% Table 2: Model Performance of Support Vector Regression (SVR), Multi-layer Perceptron (MLP), Lasso regression (Lasso), and Gradient Boost Regression Tree (GBRT)，Median Absolute Error (Median AE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Correlation (Corr) with MoCA score. Model MedianAE MAE RMSE Corr MLP 3.81 ± 0.82 4.13 ± 0.72 5.17 ± 0.84 0.59 ± 0.11 SVR 2.04 ± 0.18 2.84 ± 0.34 3.78 ± 0.57 0.90 ± 0.06 Lasso 2.38 ± 0.07 2.91 ± 0.24 3.76 ± 0.42 0.85 ± 0.05 GBRT 2.31 ± 0.28 2.93 ± 0.29 3.74 ± 0.42 0.77 ± 0.08 Baseline 10.61 ± 0.18 10.71 ± 0.1 11.59 ± 0.10 0.47 ± 0.06 Table 3: Sample size estimation of different sensitivity and specificity，sensitivity (Sen)，specificity (Spe)，number of Sen (Nsen), number of Spe (Nspe). Spe Sen=0.5 Sen=0.6 Sen=0.7 Sen =0.8 Sen=0.9 Nsen Nspe Nsen Nspe Nsen Nspe Nsen Nspe Nsen Nspe Spe=0.5 68 68 65 68 57 68 86 68 86 68 Spe=0.6 68 65 65 65 57 65 82 65 82 65 Spe=0.7 68 57 65 57 57 57 72 57 72 57 Spe=0.8 68 55 65 55 57 55 55 55 55 55 Spe=0.9 68 32 65 32 57 32 32 32 32 32 Table 4: Comparison of common diagnosis or screening approaches Type VECA Neuropsychological tests Imaging Laboratory Content VR eye-tracking cognitive assessment MMSE, MOCA etc. Brain scans Cerebrospinal fluid examination Method Self-administered Assisted CT/MRI/PET-CT Laboratory examinations Pros and Cons Short duration (5 minutes); Non-invasive; Self-administered test; Objective Long duration (30 minutes); Non-invasive; Assisted by others; Subjective Radiation Exposure; Objective; Uncapable for early screening; High cost Invasive; Objective High cost Algorithm 1 Algorithm 1 is available in the Supplementary Files section. Additional Declarations (Not answered) Supplementary Files Appendix.docx Algorithm1.docx Cite Share Download PDF Status: Published Journal Publication published 22 Aug, 2024 Read the published version in npj Digital Medicine → Version 1 posted Editorial decision: revise 22 Mar, 2024 Review # 3 received at journal 14 Mar, 2024 Review # 2 received at journal 11 Mar, 2024 Reviewer # 3 agreed at journal 04 Mar, 2024 Reviewer # 2 agreed at journal 01 Mar, 2024 Review # 1 received at journal 21 Feb, 2024 Reviewer # 1 agreed at journal 14 Feb, 2024 Reviewers invited by journal 23 Jan, 2024 Editor assigned by journal 02 Jan, 2024 Submission checks completed at journal 02 Jan, 2024 First submitted to journal 02 Jan, 2024 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-3828765","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":268865292,"identity":"adfefeec-e5a9-4beb-b8eb-f74858d6a0ae","order_by":0,"name":"Qing Yuan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzElEQVRIiWNgGAWjYBAC+/kHEh8kVPy3sz/eQKQWAwmGxwYfzjAnM5w5QLQWxmeSM9uYGRtuJBCpxVy6OUGa5wwbM+PMxxtvMNTYRBPUYjnnWIIxTwUPH7N0WrEFw7G03AaCeg7kJCTznJFgZpPOMZNgbDhMjJb8D4d52wwYeyTPEKnF4EZCYuPMtgTGGRI8RGqR7DmQzPDhzIFkAx6gXxKI8Qs/e0P6j4SKA3YG7Ic33vhQY0OEX5AdKZFAinKIFlJ1jIJRMApGwcgAAHxaQs5irqldAAAAAElFTkSuQmCC","orcid":"","institution":"Shenzhen Baoan Chronic Hospital, China","correspondingAuthor":true,"prefix":"","firstName":"Qing","middleName":"","lastName":"Yuan","suffix":""},{"id":268865293,"identity":"5899db64-3fce-416f-ab0f-672ec387f197","order_by":1,"name":"Ying Xu","email":"","orcid":"","institution":"Shenzhen Baoan Chronic Hospital, China","correspondingAuthor":false,"prefix":"","firstName":"Ying","middleName":"","lastName":"Xu","suffix":""},{"id":268865294,"identity":"52e7e828-eb44-48f3-a158-5dcc37bb1dc7","order_by":2,"name":"Xu Zhang","email":"","orcid":"https://orcid.org/0000-0002-4803-8628","institution":"National Engineering laboratory for Big Data System Computing Technology, Shenzhen university","correspondingAuthor":false,"prefix":"","firstName":"Xu","middleName":"","lastName":"Zhang","suffix":""},{"id":268865295,"identity":"beb7c4b6-5061-48c0-bff1-9a2f6830d77e","order_by":3,"name":"Chi Zhang","email":"","orcid":"https://orcid.org/0009-0000-3895-7949","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Chi","middleName":"","lastName":"Zhang","suffix":""},{"id":268865296,"identity":"9c7a7413-c060-428b-887f-26c464837421","order_by":4,"name":"Baobao Pan","email":"","orcid":"","institution":"Shenzhen Yiwei Technology, China","correspondingAuthor":false,"prefix":"","firstName":"Baobao","middleName":"","lastName":"Pan","suffix":""}],"badges":[],"createdAt":"2024-01-02 06:55:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3828765/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3828765/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41746-024-01206-5","type":"published","date":"2024-08-22T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":50227220,"identity":"3cbb6b3e-597f-41c8-aaa7-46871a81a91a","added_by":"auto","created_at":"2024-01-26 18:24:48","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":60440,"visible":true,"origin":"","legend":"\u003cp\u003eVR eye-tracking cognitive assessment based on VR head-mounted display and eye-tracking technology. In each task, participants were instructed to understand the task and fulfill related cognitive task. Multiple images including a correct answer (Area of interest, AOI) and distractors (non-Area of interest, non-AOI) were displayed in the VR scene, and subjects were instructed to identify and fixate on the correct answer.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/42f5d0d377b329e8800b2375.jpg"},{"id":50227219,"identity":"3b04de78-c6d5-411a-ad6e-8915bd37e007","added_by":"auto","created_at":"2024-01-26 18:24:48","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":25868,"visible":true,"origin":"","legend":"\u003cp\u003eFeature importance analysis. The graph illustrates the relative importance of different features in the VRCA. The importance of each feature is expressed as a percentage. Gender accounts for 5.7%, age contributes 2.8%, education holds a weight of 23.7%, and task-related factors have the highest importance at 67.8%.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/641aa150782de232f91de0b8.jpg"},{"id":50227667,"identity":"ec18d415-4847-4f4b-b0bf-fd62f7550d33","added_by":"auto","created_at":"2024-01-26 18:32:48","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":64721,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation comparison of VR-AI model and baseline VR model. The coordinates of scatter points consist of model predicted scores (X-axis) and MoCA scores (Y-axis). Linear regression lines were also given to manifest linear relevance of models and MoCA score (VR model VS MoCA, r = 0.58, p \u0026lt; 0.0001; VR-AI model VS MoCA, r = 0.90, p \u0026lt; 0.0001).\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/137b8e0c020bbf593305acb3.jpg"},{"id":50227224,"identity":"9e195a7a-27b3-4211-b204-b36c0f2070a8","added_by":"auto","created_at":"2024-01-26 18:24:49","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":84352,"visible":true,"origin":"","legend":"\u003cp\u003eThe flowchart of machine learning model used for MoCA prediction. The pipeline includes data acquisition, data processing, feature extraction (including memory (Mem), visual attention (Att), abstraction ability (Abs), and calculation (Calc)), machine learning (ML), and optimal threshold score determination.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/95d21f57dee48da9b4cd447a.jpg"},{"id":50227223,"identity":"764e5a4b-f10e-441c-9c82-a292acf79f6b","added_by":"auto","created_at":"2024-01-26 18:24:49","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":75964,"visible":true,"origin":"","legend":"\u003cp\u003eThe ROC curve of the VR-AI model for three education-based groups showed AUCs of 0.92, 0.95, and 0.94 for the groups with less than 6 years, 6 to 9 years, and more than 9 years of education, respectively.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/c18ba18257e9144379a14cc3.jpg"},{"id":63998016,"identity":"922f35d5-8f96-4598-9359-af11e15a0818","added_by":"auto","created_at":"2024-09-04 18:04:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":880178,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/92070ee0-06e8-4251-8bd2-0cbaed907bb2.pdf"},{"id":50227225,"identity":"5dccfc73-5ef5-4aa2-ad0e-1bccc5d42d25","added_by":"auto","created_at":"2024-01-26 18:24:49","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2441855,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/fbd664584e7aabc48e0d44b6.docx"},{"id":50227222,"identity":"56ac86ff-c489-4e0e-89f0-84f52fcba279","added_by":"auto","created_at":"2024-01-26 18:24:48","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":18307,"visible":true,"origin":"","legend":"","description":"","filename":"Algorithm1.docx","url":"https://assets-eu.researchsquare.com/files/rs-3828765/v1/8740443240ab934208056e1f.docx"}],"financialInterests":"(Not answered)","formattedTitle":"The VR Eye-tracking Cognitive Assessment (VECA): A Portable and Efficient Dementia Screening Tool Using Eye-Tracking Technology, Machine Learning, and Virtual Reality","fulltext":[{"header":"Introduction","content":"\u003cp\u003eDementia refers to a spectrum of neurodegenerative syndromes characterized by progressive disturbances in various cognitive functions severe enough to interfere with the patient\u0026rsquo;s activities of daily living. The rapid increase in the number of patients with dementia has become a global health challenge, with Alzheimer\u0026rsquo;s disease (AD) being the most common, accounting for 60-70% [1]. Mild cognitive impairment (MCI) refers to the progressive decline of memory or other cognitive functions but does not affect the ability of daily life as dementia [2]. The increasing prevalence of cognitive impairment has caused enormous economic loss [3, 4]. Accumulating evidence shows that early diagnosis and timely intervention can delay cognitive decline [5\u0026ndash;7], so early screening in the preclinical stage of dementia is necessary. \u003c/p\u003e\n\u003cp\u003eCurrently, cerebrospinal fluid \u003cem\u003e\u0026beta;\u003c/em\u003e-amyloid (amyloid \u003cem\u003e\u0026beta;\u003c/em\u003e, A\u003cem\u003e\u0026beta;\u003c/em\u003e) and tau, amyloid positron emission tomography (PET) and AD pathogenic gene carrying are diagnostic markers for AD [8]. Though accurate, these approaches could hardly perform as an early screening tool for high costs and surgical invasiveness. Dementia is essentially cognitive dysfunction, so assessment of cognition is an important part of the diagnosis. Neuropsychological paper-based instruments, such as the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA) are commonly used, which cover orientation, memory, attention, calculation, language ability, visuospatial ability, etc. Though valid and reliable, traditional neuropsychological tests must be administered by trained physicians, and are neither simple nor efficient enough to serve as large-scale dementia screening tools. Moreover, neuropsychological assessment results are affected by both the physician\u0026rsquo;s subjective judgments and the testing environment. At present, there is no efficient solution for large-scale dementia screening.\u003c/p\u003e\n\u003cp\u003eCognitive-related uses of eye-tracking technology have been exploded in recent years, enabling online cognitive activity to be recorded, physicians to connect learning outcomes to the cognitive processes of subjects [9], and detection of emotional and cognitive states [10]. Eye tracking is a real-time sensor technology that measures gaze points and eye movements. A large number of studies have shown that detailed and diverse eye movements derived by algorithms from eye trackers made quantitative and workforce-saving cognitive assessment feasible [11\u0026ndash;19]. Recent studies have applied VR head-mounted display equipment and VR cognitive assessment tasks to assess cognitive function, which can distinguish patients with mild cognitive impairment from low-risk and high-risk patients with AD [20,21]. The combination of eye-tracking technology and VR, with the portability of VR head-mounted display equipment and VR immersive environment, is expected to be applied to large-scale cognitive impairment screening in multiple scenarios. However, two challenging issues have to be handled. First, these technologies should be integrated in a smart way so that efficiency is guaranteed. Moreover, subtle changes in eye movements can potentially serve as biomarkers for early MCI identification，interpreting the vast amount of eye-tracking data poses challenges. \u003c/p\u003e\n\u003cp\u003eIn this study, based on eye-tracking technology, machine learning, and virtual reality, we designed and developed an efficient, portable, and quantitative early screening tool for dementia. This project manages to promote the early detection of people with cognitive impairment, delay the progress of dementia and will help greatly reduce the spiritual, nursing, and economic burdens on both individuals and society caused by dementia.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003ch3\u003eParticipants\u003c/h3\u003e\n\u003cp\u003eThis study is conducted with the cooperation of Shenzhen Baoan Chronic Hospital，National Engineering laboratory for Big Data System Computing Technology, Shenzhen university, China, and Yiwei Medical Technology Co., Ltd. A total of 201 subjects were evaluated, including 88 from Fuan CRC (Community Rehabilitation Center) and 113 from Wanxiang CRC. Some inclusions of participants were: (1) aged 65.5\u0026plusmn;5.1 years; (2) 81 males (40.3%), female = 120 (59.7%); (3) years of education 9.4\u0026plusmn;3.8 (Table 1). The study excluded participants who had psychiatric, ophthalmological, or hearing impairment disorders, as well as those who were unable to sit comfortably due to severe physical illness. Only participants with normal vision or corrected-to-normal vision without color blindness were included. A small number of participants were further excluded from the analysis if they encountered difficulties calibrating to the eye tracker or did not attempt to view the images. Exclusion criteria were applied in cases where the eye tracking equipment faced challenges in achieving proper pupil and corneal reflection due to physiological constraints or visual problems (e.g., droopy eyelid, cataracts, detached retinas, glaucoma, pupils too small), or if participants were unable to complete the eye tracking calibration procedure.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAssessment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData acquisition\u0026nbsp;\u003c/strong\u003eDemographic characteristics including age, gender, and education levels of the participants were collected and each of them underwent paper-based instruments of MoCA Chinese version. \u003cstrong\u003eFigure 1\u0026nbsp;\u003c/strong\u003edepicts the cognitive assessment system based on VR head-mounted display and eye-tracking technology used in the study, collectively referred to as VECA, which takes 5 minutes only. Eye tracking calibration would be fulfilled before participants started the assessment to make sure gaze points captured by the eye tracker are valid. Cognitive tasks include memory (encoding, storage, and recall), visual attention (smooth pursuit), abstraction ability (ability to form abstract concepts), visuospatial working memory, visuospatial and executive function, calculation, language (understanding and execution of commands), and short-term memory binding. Detailed descriptions of each task are given in \u003cstrong\u003eappendix\u003c/strong\u003e (A1-A11). In each task, participants were usually given 3 seconds to understand the task and 5 or 8 seconds to fulfill related cognitive task. Multiple images including a correct answer (Area of interest, AOI) and distractors (non-Area of interest, non-AOI) were displayed in the VR scene, and subjects were instructed to identify and fixate on the correct answer. All eye movements were captured by the Tobii eye tracker embedded in Pico neo 3 pro eye, 95% of users can track better than 2\u0026deg; when their eyes are looking straight ahead (https://developer.tobii.com/xr/learn/eye-behavior/hardware-accuracy/).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData processing\u003c/strong\u003e Biological eye movements such as fixation and saccade were derived out of raw gaze points with I-VT algorithm, which automatically classifies excessive noise such as microsaccades as abnormal movements [22]. Gaze points that extremely deviate from their neighbors may be caused by winks or unexpected shivers and were also eliminated. The percentage of time the subject fixates on the AOI over total fixation time of a certain task is calculated and deemed as features. Previous research claimed that age is the biggest risk factor and made gender also a significant risk factor due to longer lifespan of female [23]. Therefore, demographic information such as age and gender of the subjects were supplemented to the feature set. Education level, as in \u003cstrong\u003eFigure 2\u003c/strong\u003e, demonstrated feature importance of nearly a quarter and would be used as a classifying attribute in the later discussion. Data preprocessing techniques such as one-hot and ordinal encoder were applied to categorical features. Standard scaling was also conducted for scale-sensitive models to eliminate the influence of various scales. MoCA scores were categorized as the target variable so that our system could play as an efficient screening tool. In this way, the target variable better quantified the subjects\u0026rsquo; cognitive capability.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModeling\u003c/strong\u003e Oyama et al. averaged each cognitive task score as the eye movement cognitive assessment score [17] (hereinafter referred to as \u003cstrong\u003eVR model\u003c/strong\u003e) which would be the baseline model. Support Vector Regression (SVR), Multi-layer Perceptron (MLP), Lasso regression (Lasso), and Gradient Boost Regression Tree (GBRT) were adopted. The model performance was evaluated using the following metrics: Median Absolute Error (Median AE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Correlation with MoCA score. The model with the best performance was selected to predict paper-based MoCA score (hereinafter referred to as \u003cstrong\u003eVR-AI model\u003c/strong\u003e) which would be the numerical cognitive indicator of the screening result.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eVR-AI model selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOptimal parameters of candidate paper based MoCA score predictors were found through grid search. Each model underwent 5-fold cross-validation and compared in the following metrics (\u003cstrong\u003eTable 2\u003c/strong\u003e). MAE and RMSE are common metrics which are widely used to describe regression performance. However, there was unpredictable noise in the target value due to human factors and lead to extreme bad cases in both train and test set. Instead, median AE could indicate model performance in a more stable way disregard of outliers. Moreover, absolute error shows more intuitive difference between predictions and labels. Therefore, MAE was weighed more as the metrics to compare models. Higher correlation coefficient between predictions and targets indicates better performance of screening (\u003cstrong\u003eFigure 3\u003c/strong\u003e). SVR overperformed other models and baseline under most of our metrics, especially under average median absolute error of 2.04 and correlation coefficient as high as 0.9 and was selected as VR-AI model. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eScreening evaluation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLu et al. proposed a Chinese version of MoCA cutting-off scores which is more adaptive to the Chinese population due to its geographically various languages, education, and local culture, which had long been the gold standard in China [24]. Therefore, we classified participants into three groups according to their education levels. Participants were grouped and labeled by Chinese standards as follows:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGroup 1:\u003c/strong\u003e 0-6 years of education: normal cognition (MoCA score 14-30), n=50; cognitive impairment (MoCA score 0-13), n=11.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGroup 2:\u003c/strong\u003e 6-9 years of education: normal cognition (MoCA score 20-30), n=44; cognitive impairment (MoCA score 0-19) n=19.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGroup 3:\u003c/strong\u003e education level greater than 9 years: normal cognition (MoCA score 25-30), n=47; cognitive impairment (MoCA score 0-24), n=31.\u003c/p\u003e\n\u003cp\u003eClassification performance of VR-AI model was evaluated due to ROC curves. Optimal thresholds of VR- AI score were given by F-\u0026beta; method which, in screening scenario, weighed more on sensitivity. The whole flowchart of the proposed model is shown in \u003cstrong\u003eFigure 4\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eThe screening performance was also evaluated in classification aspect. The subjects were grouped based on education levels, and the AUCs of three groups were 0.92(95% CI: 0.85-0.99), 0.95(95% CI: 0.89-1.00), and 0.94(95% CI: 0.89-0.99) respectively (\u003cstrong\u003eFigure 5\u003c/strong\u003e). The optimal screening cut-off scores for each group are given by F-\u0026beta; score: less than 6 years of education, 13/14; 6 to 9 years of education, 21/22; more than 9 years of education, 23/24. The VR-AI model showed excellent sensitivities of 81.8%, 94.7%, and 83.8% in screening for cognitive impairment and specificities of 98%, 91.1%, and 73.6%. The sensitivity of the overall screening of cognitive impairment and cognitively normal subjects can reach 88.5% with specificity of 83%.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eEye movements were highly informative indicators of cognitive function. In our study, together with basic information\u0026mdash;gender, age and education, encoded eye movements can excellently classify normal and cognitive impairment subjects within education-based groups via machine learning. As a result, we developed an efficient screening tool to identify subjects that would be labeled cognitive impairment by traditional paper-based instruments. The crucial part is regression of MoCA score. SVM regression outperformed and achieved absolute error of 2.04 and strong correlation with target variable. An important application of our screening tool is providing diagnosis suggestions. The AUCs of three education-based groups were 0.92(95% CI: 0.85-0.99), 0.95(95% CI: 0.89-1.00), and 0.94(95% CI: 0.89-0.99) respectively. Proposed optimal cut-off scores for each group achieve sensitivities of 81.8%, 94.7%, and 83.8% in screening for cognitive impairment and specificities of 98%, 91.1%, and 73.6%.\u003c/p\u003e\n\u003cp\u003eTo retrospect the sample size of our research, the estimate of the sample size was given according to the general requirements of diagnostic research: 1) screening test sensitivity (Sen)\u0026ge; 0.5, sensitivity tolerance error 0.10; 2) The specificity (Spe) of screening test was \u0026ge; 0.5, the tolerance error of specificity was 0.10.\u003c/p\u003e\n\u003cp\u003e3) Significance test level two-sided \u0026alpha; = 0.05. The results are presented in the table below. The sample size of the three groups (Group1 N=61, group2 N=63, group3=78) in this study can meet the requirements of statistics.\u003c/p\u003e\n\u003cp\u003eThere are various cognitive assessment and diagnostic methods available, including neuropsychological testing, imaging, and laboratory examinations. Imaging examinations such as CT/MRI/PET-CT provide objective results but expose individuals to radiation, are not suitable for early screening, and are costly. Laboratory examinations such as cerebrospinal fluid testing are invasive but provide objective results. For more than a decade, MoCA has been widely applied in screening and diagnosing dementia either in paper or electronic form. However, besides 30 minutes of duration, cultural and educational influences may make it unsuitable for certain populations. Screening results may be affected by the emotional and mental state of the test subjects as well as how professional psychiatrists are. Moreover, comparison of common diagnosis or screening approaches from the aspect of convenience and efficiency was given in \u003cstrong\u003eTable 4\u003c/strong\u003e. In contrast, the immersive environment of VECA ensures data integrity and reliability, minimizing environmental impact on test subjects. Multiple dimensions of assessment and objective results can be provided. VECA is less time-consumable (5-minute test) and fully self-administered and its portability makes large-scale cognitive screening possible and efficient especially in fundamental clinics and communities.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLimitations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDespite of satisfying screening performance, there still exist potential bias and limitations. First, participants were from communities of Shenzhen, information on other variables, e.g. comorbidities, functional dependence, eventually drugs, could not be collected. Both will cause bias in participants selection and make our results valid only within certain cohort. Comprehensive data should be collected in future studies to fully consider the influencing factors that affect the evaluation results. Since both train and test cohort were selected within limited samples, lack of external validation should also be considered when applying VECA.\u003c/p\u003e\n\u003cp\u003eFurthermore, the interactions between subjects and VR headset are based on visual and literal capabilities. Lack of either one will cause the system to misinterpret the subjects\u0026rsquo; eye movements and produce deviated predictions. On the machine learning level, validity and authenticity of the target variable (MoCA) are affected by the psychiatrists. Excessive noise in MoCA can distort our model\u0026apos;s distribution and compromise its accuracy. To improve model performance, other informative features could be extracted from raw eye movements. Their validation and integration into the system should be studied in the future.\u003c/p\u003e\n\u003cp\u003eIn practical applications, using multiple methods for screening cognitive impairment can improve diagnostic accuracy and reduce the risk of missed diagnoses and misdiagnoses.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis research suggests that VECA can be an efficient and portable tool for dementia screening. VECA has the potential to improve early diagnosis and treatment of dementia by helping physicians identify mild cognitive impairment in the early stages. This study highlights the application of eye-tracking technology, machine learning, and virtual reality in cognitive evaluation. Further research with more data is needed to compare the effectiveness of VECA with traditional tests and dementia diagnosis outcomes.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflict of interest statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contribution statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eQing Yuan: Organizer of the whole project. Responsible for screening scenes, clinical suggestions, and other resources. Propose initial ideas and other clinical details.\u003c/p\u003e\n\u003cp\u003eXu Zhang: Designing the content for experiments and VR eye-tracking cognitive tests, providing professional digital medical guidance from a perspective of neuroscience science, author of the initial manuscript.\u003c/p\u003e\n\u003cp\u003eYing Xu: Director of participants, data acquisition, and reviewer of data quality.\u003c/p\u003e\n\u003cp\u003eChi Zhang: Director of data processing, feature extraction, and modeling. Author of the complete manuscript. Baobao Pan: Provider of algorithm suggestions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContribution to the field\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDementia is a growing global health challenge, and early screening is crucial in identifying patients with mild cognitive impairment (MCI) as a transitional stage preceding dementia. Despite various diagnostic markers available, the current options are limited by cost and invasiveness. Neuropsychological tests have been used, but they are neither simple nor efficient enough to serve as\u003c/p\u003e\n\u003cp\u003elarge-scale dementia screening tools. Our study proposes the use of eye-tracking data as a quantitative and multi-dimensional approach to assess cognitive function. We developed a VR-based tool that guarantees the integrity of eye-tracking data and enables efficient large-scale early screening of cognitive impairment. Our proposed machine learning models achieved a high correlation with the MoCA score - a commonly used tool for identifying MCI - and optimal cut-off scores were established for distinguishing normal and cognitive impairment subjects. This method avoids invasiveness, lower workforce, time cost, and may apply to large-scale screening well.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by Shenzhen Fundamental Research Program (JCYJ20220531091006014)\u003c/p\u003e\n\u003cp\u003eEthics statements\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudies involving animal subjects.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGenerated Statement: No animal studies are presented in this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudies involving human subjects.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGenerated Statement: The studies involving human participants were reviewed and approved by Shenzhen Baoan Chronic Hospital. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInclusion of identifiable human data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGenerated Statement: No potentially identifiable human images or data is presented in this study.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWHO. Dementia. 2021; Available from: https://www.who.int/news-room/fact-sheets/detail/dementia\u003c/li\u003e\n\u003cli\u003ePetersen, R.C., Clinical practice. Mild cognitive impairment. The New England journal of medicine, 2011. 364(23): p. 2227-2234.\u003c/li\u003e\n\u003cli\u003eJia, L., 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. The Lancet Public Health, 2020. 5(12):p. e661-e671.\u003c/li\u003e\n\u003cli\u003eJia, J., et al., The cost of Alzheimer\u0026rsquo;s disease in China and re-estimation of costs worldwide. Alzheimers Dement, 2018. 14(4): p. 483-491.\u003c/li\u003e\n\u003cli\u003eBlondell, S.J., R. Hammersley-Mather, and J.L. Veerman, Does physical activity prevent cognitive decline and dementia?: A systematic review and meta-analysis of longitudinal studies. BMC public health, 2014. 14(1): p. 1-12.\u003c/li\u003e\n\u003cli\u003eNgandu, T., et al., A 2 year multidomain intervention of diet, exercise, cognitive training, and vascu- lar risk monitoring versus control to prevent cognitive decline in at-risk elderly people (FINGER): a randomised controlled trial. The Lancet, 2015. 385(9984): p. 2255-2263.\u003c/li\u003e\n\u003cli\u003ePark, H., et al., Combined Intervention of Physical Activity, Aerobic Exercise, and Cognitive Exercise Intervention to Prevent Cognitive Decline for Patients with Mild Cognitive Impairment: A Randomized Controlled Clinical Study. J Clin Med, 2019. 8(7).\u003c/li\u003e\n\u003cli\u003eDubois, B., et al., Advancing research diagnostic criteria for Alzheimer\u0026rsquo;s disease: the IWG-2 criteria. The Lancet Neurology, 2014. 13(6): p. 614-629.\u003c/li\u003e\n\u003cli\u003eLai, M.-L., et al., A review of using eye-tracking technology in exploring learning from 2000 to 2012. Educational Research Review, 2013. 10: p. 90-115.\u003c/li\u003e\n\u003cli\u003eSkaramagkas, V., et al., Review of eye tracking metrics involved in emotional and cognitive processes. IEEE Rev Biomed Eng, 2021. PP.\u003c/li\u003e\n\u003cli\u003eLagun, D., et al., Detecting cognitive impairment by eye movement analysis using automatic classifica- tion algorithms. Journal of neuroscience methods, 2011. 201(1): p. 196-203.\u003c/li\u003e\n\u003cli\u003eNie, J., et al., Early Diagnosis of Mild Cognitive Impairment Based on Eye Movement Parameters in an Aging Chinese Population. Front Aging Neurosci, 2020. 12: p. 221.\u003c/li\u003e\n\u003cli\u003eHaque, R.U., et al., VisMET: a passive, efficient, and sensitive assessment of visuospatial memory in healthy aging, mild cognitive impairment, and Alzheimer\u0026rsquo;s disease. Learn Mem, 2019. 26(3): p. 93-100.\u003c/li\u003e\n\u003cli\u003ePereira, M., et al., Visual Search Efficiency in Mild Cognitive Impairment and Alzheimer\u0026rsquo;s Disease: An Eye Movement Study. J Alzheimers Dis, 2020. 75(1): p. 261-275.\u003c/li\u003e\n\u003cli\u003eJiang, J., et al., A Novel Detection Tool for Mild Cognitive Impairment Patients Based on Eye Movement and Electroencephalogram. J Alzheimers Dis, 2019. 72(2): p. 389-399.\u003c/li\u003e\n\u003cli\u003eChehrehnegar, N., et al., Executive function deficits in mild cognitive impairment: evidence from saccade tasks. Aging \u0026amp; mental health, 2021: p. 1-9.\u003c/li\u003e\n\u003cli\u003eOyama, A., et al., Novel Method for Rapid Assessment of Cognitive Impairment Using High-Performance Eye-Tracking Technology. Sci Rep, 2019. 9(1): p. 12932.\u003c/li\u003e\n\u003cli\u003eMengoudi, K., et al., Augmenting dementia cognitive assessment with instruction-less eye-tracking tests. IEEE journal of biomedical and health informatics, 2020. 24(11): p. 3066-3075.\u003c/li\u003e\n\u003cli\u003eParra, M.A., M. Schumacher, and G. Fern\u0026acute;andez, A novel peripheral biomarker for mild cognitive im- pairment and Alzheimer\u0026rsquo;s disease. Alzheimer\u0026rsquo;s \u0026amp; Dementia, 2020. 16(S4).\u003c/li\u003e\n\u003cli\u003eHowett, D., et al., Differentiation of mild cognitive impairment using an entorhinal cortex-based test of virtual reality navigation. Brain, 2019. 142(6): p. 1751-1766.\u003c/li\u003e\n\u003cli\u003eCorriveau Lecavalier, N., et al., Use of immersive virtual reality to assess episodic memory: A validation study in older adults. Neuropsychol Rehabil, 2020. 30(3): p. 462-480.\u003c/li\u003e\n\u003cli\u003eSalvucci, D. D., \u0026amp; Goldberg, J. H. (2000, November). Identifying fixations and saccades in eye-tracking protocols. In Proceedings of the 2000 symposium on Eye tracking research \u0026amp; applications (pp. 71-78).\u003c/li\u003e\n\u003cli\u003eBeam CR, Kaneshiro C, Jang JY, Reynolds CA, Pedersen NL, Gatz M. Differences Between Women and Men in Incidence Rates of Dementia and Alzheimer\u0026rsquo;s Disease. J Alzheimers Dis. 2018;64(4):1077-1083. doi: 10.3233/JAD-180141. PMID: 30010124; PMCID: PMC6226313.\u003c/li\u003e\n\u003cli\u003eLu J, Li D, Li F, et al. Montreal cognitive assessment in detecting cognitive impairment in Chinese elderly individuals: a population-based study[J]. Journal of geriatric psychiatry and neurology, 2011, 24(4): 184-190.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1:\u0026nbsp;\u003c/strong\u003eDemographic characteristics of participants\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAttribute\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eCount\u0026nbsp;(N)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePct.\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eData\u0026nbsp;Source\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eFuan\u0026nbsp;CRC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e43.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eWanxiang\u0026nbsp;CRC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e113\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e56.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e55-65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e113\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e56.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e65-75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e39.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026gt;\u0026nbsp;\u003c/em\u003e75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e4.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eGender\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e40.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e120\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e59.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eEducation\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eLevel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eIlliterate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e3.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003ePrimary\u0026nbsp;School\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e26.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eJunior\u0026nbsp;High School\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e30.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eHigh\u0026nbsp;School\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e27.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"40.4639175257732%\" valign=\"top\"\u003e\n \u003cp\u003eBachelor\u0026nbsp;or\u0026nbsp;above\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"30.412371134020617%\" valign=\"top\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.123711340206185%\" valign=\"top\"\u003e\n \u003cp\u003e11.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2:\u0026nbsp;\u003c/strong\u003eModel Performance of Support Vector Regression (SVR), Multi-layer Perceptron (MLP), Lasso regression (Lasso), and Gradient Boost Regression Tree (GBRT)，Median Absolute Error (Median AE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Correlation (Corr) with MoCA score.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"477\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"19.07756813417191%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.270440251572328%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMedianAE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.238993710691823%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMAE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.91614255765199%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eRMSE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.49685534591195%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eCorr\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"19.07756813417191%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMLP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.270440251572328%\" valign=\"top\"\u003e\n \u003cp\u003e3.81\u0026nbsp;\u0026plusmn;\u0026nbsp;0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.238993710691823%\" valign=\"top\"\u003e\n \u003cp\u003e4.13\u0026nbsp;\u0026plusmn;\u0026nbsp;0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.91614255765199%\" valign=\"top\"\u003e\n \u003cp\u003e5.17\u0026nbsp;\u0026plusmn;\u0026nbsp;0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.49685534591195%\" valign=\"top\"\u003e\n \u003cp\u003e0.59\u0026nbsp;\u0026plusmn;\u0026nbsp;0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"19.07756813417191%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.270440251572328%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e2.04\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e\u0026plusmn;\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.238993710691823%\" valign=\"top\"\u003e\n \u003cp\u003e2.84\u0026nbsp;\u0026plusmn;\u0026nbsp;0.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.91614255765199%\" valign=\"top\"\u003e\n \u003cp\u003e3.78\u0026nbsp;\u0026plusmn;\u0026nbsp;0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.49685534591195%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.90\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e\u0026plusmn;\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e0.06\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"19.07756813417191%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eLasso\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.270440251572328%\" valign=\"top\"\u003e\n \u003cp\u003e2.38\u0026nbsp;\u0026plusmn;\u0026nbsp;0.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.238993710691823%\" valign=\"top\"\u003e\n \u003cp\u003e2.91\u0026nbsp;\u0026plusmn;\u0026nbsp;0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.91614255765199%\" valign=\"top\"\u003e\n \u003cp\u003e3.76\u0026nbsp;\u0026plusmn;\u0026nbsp;0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.49685534591195%\" valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u0026plusmn;\u0026nbsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"19.07756813417191%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eGBRT\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.270440251572328%\" valign=\"top\"\u003e\n \u003cp\u003e2.31\u0026nbsp;\u0026plusmn;\u0026nbsp;0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.238993710691823%\" valign=\"top\"\u003e\n \u003cp\u003e2.93\u0026nbsp;\u0026plusmn;\u0026nbsp;0.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.91614255765199%\" valign=\"top\"\u003e\n \u003cp\u003e3.74\u0026nbsp;\u0026plusmn;\u0026nbsp;0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.49685534591195%\" valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u0026plusmn;\u0026nbsp;0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"19.07756813417191%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eBaseline\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.270440251572328%\" valign=\"top\"\u003e\n \u003cp\u003e10.61\u0026nbsp;\u0026plusmn;\u0026nbsp;0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.238993710691823%\" valign=\"top\"\u003e\n \u003cp\u003e10.71\u0026nbsp;\u0026plusmn;\u0026nbsp;0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.91614255765199%\" valign=\"top\"\u003e\n \u003cp\u003e11.59\u0026nbsp;\u0026plusmn;\u0026nbsp;0.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.49685534591195%\" valign=\"top\"\u003e\n \u003cp\u003e0.47 \u0026plusmn; 0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3:\u003c/strong\u003e Sample size estimation of different sensitivity and specificity，sensitivity (Sen)，specificity (Spe)，number of Sen (Nsen), number of Spe (Nspe).\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.573673870333987%\" valign=\"top\"\u003e\n \u003cp\u003eSpe\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"17.4852652259332%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eSen=0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"17.4852652259332%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eSen=0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"17.4852652259332%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eSen=0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.644400785854616%\" valign=\"top\"\u003e\n \u003cp\u003eSen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.840864440078585%\" valign=\"top\"\u003e\n \u003cp\u003e=0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"17.4852652259332%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eSen=0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.549019607843137%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003eNsen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003eNspe\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003eNsen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003eNspe\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003eNsen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003eNspe\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003eNsen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003eNspe\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003eNsen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003eNspe\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.549019607843137%\" valign=\"top\"\u003e\n \u003cp\u003eSpe=0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.549019607843137%\" valign=\"top\"\u003e\n \u003cp\u003eSpe=0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.549019607843137%\" valign=\"top\"\u003e\n \u003cp\u003eSpe=0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.549019607843137%\" valign=\"top\"\u003e\n \u003cp\u003eSpe=0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.549019607843137%\" valign=\"top\"\u003e\n \u003cp\u003eSpe=0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.627450980392156%\" valign=\"top\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.823529411764707%\" valign=\"top\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4:\u003c/strong\u003e Comparison of common diagnosis or screening approaches\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"654\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.431192660550458%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eType\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.464831804281346%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eVECA\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.01834862385321%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNeuropsychological\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003etests\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.972477064220183%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eImaging\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.1131498470948%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eLaboratory\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.431192660550458%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eContent\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.464831804281346%\" valign=\"top\"\u003e\n \u003cp\u003eVR eye-tracking\u003c/p\u003e\n \u003cp\u003ecognitive assessment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.01834862385321%\" valign=\"top\"\u003e\n \u003cp\u003eMMSE, MOCA etc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.972477064220183%\" valign=\"top\"\u003e\n \u003cp\u003eBrain scans\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.1131498470948%\" valign=\"top\"\u003e\n \u003cp\u003eCerebrospinal\u003c/p\u003e\n \u003cp\u003efluid examination\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.431192660550458%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMethod\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.464831804281346%\" valign=\"top\"\u003e\n \u003cp\u003eSelf-administered\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.01834862385321%\" valign=\"top\"\u003e\n \u003cp\u003eAssisted\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.972477064220183%\" valign=\"top\"\u003e\n \u003cp\u003eCT/MRI/PET-CT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.1131498470948%\" valign=\"top\"\u003e\n \u003cp\u003eLaboratory examinations\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.431192660550458%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003ePros and Cons\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.464831804281346%\" valign=\"top\"\u003e\n \u003cp\u003eShort duration (5 minutes);\u003c/p\u003e\n \u003cp\u003eNon-invasive;\u003c/p\u003e\n \u003cp\u003eSelf-administered test; Objective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.01834862385321%\" valign=\"top\"\u003e\n \u003cp\u003eLong duration (30 minutes);\u003c/p\u003e\n \u003cp\u003eNon-invasive;\u003c/p\u003e\n \u003cp\u003eAssisted by others; Subjective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.972477064220183%\" valign=\"top\"\u003e\n \u003cp\u003eRadiation\u003c/p\u003e\n \u003cp\u003eExposure;\u003c/p\u003e\n \u003cp\u003eObjective; Uncapable for early screening;\u003c/p\u003e\n \u003cp\u003eHigh cost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.1131498470948%\" valign=\"top\"\u003e\n \u003cp\u003eInvasive; Objective\u003c/p\u003e\n \u003cp\u003eHigh cost\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Algorithm 1","content":"\u003cp\u003eAlgorithm 1 is available in the Supplementary Files section.\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":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Cognitive impairment, eye tracking, early Screening, machine learning (ML), virtual reality, Dementia - Alzheimer disease","lastPublishedDoi":"10.21203/rs.3.rs-3828765/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3828765/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003eDementia is a significant global health challenge, and early screening during the preclinical stage is crucial. However, current diagnostic biomarkers for Alzheimer's Disease, the most common cause of dementia, have limitations in terms of cost and invasiveness. Mild cognitive impairment (MCI) is recognized as a transitional stage preceding dementia. Whileneuropsychological tests like the Montreal Cognitive Assessment (MoCA) are effective for identifying MCI, they are not suitable for large-scale dementia screening.\u003c/p\u003e\n\u003cp\u003eEye-tracking technology has emerged as a promising tool for cognitive assessment by capturing and quantifying eye movements related to cognitive behavior. Subtle changes in eye movements can potentially serve as biomarkers for early MCI identification. However, interpreting the vast amount of eye-tracking data poses challenges. To address this, machine learning methods in computer science can be applied to analyze eye-tracking data and identify patterns or abnormalities indicative of MCI. Machine learning models trained on large datasets can improve the accuracy and efficiency of MCI identification. Additionally, the immersive nature of virtual reality (VR) technology allows for uninterrupted eye-tracking processes, while the portability of VR head-mounted devices enables efficient and large-scale early screening for community cognitive impairment.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eObjective: \u003c/strong\u003eDevelop a dementia screening tool, VR Eye-tracking Cognitive Assessment (VECA), using eye-tracking technology, machine learning, and virtual reality as an alternative to traditional neuropsychological tests. This tool aims to help physicians detect cognitive impairment, particularly in the early stages of MCI, on a larger scale.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003e201 subjects from Shenzhen Baoan Chronic Hospital were administered MoCA test and VECA. Raw gaze data were captured by the eye tracker of the VR headset and filtered as eye movements which would be encoded as features. Machine learning models were established as the predictor of MoCA score and the classifier of cognitive impairment of three education-based groups within which optimal cut-off score was given. The study has been approved by Shenzhen Baoan Chronic Hospital Ethics Committee and all subjects have signed written informed consent for participation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eSupport vector regression was proposed as the VR-AI model and achieved high correlation of 0.9 with MoCA score, greater than baseline model of 0.58. Optimal cut-off scores (less than 6 years of education: 14/15; 6 to 9 years of education: 18/19; more than 9 years of education: 23/24) can well distinguish normal and cognitive impairment subjects -with overall sensitivity of 88.5% and specificity of 83%.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003eVECA is a portable and efficient dementia screening tool that utilizes eye-tracking technology, machine learning, and virtual reality. It offers a quantitative approach for large-scale early screening of dementia.\u003c/p\u003e","manuscriptTitle":"The VR Eye-tracking Cognitive Assessment (VECA): A Portable and Efficient Dementia Screening Tool Using Eye-Tracking Technology, Machine Learning, and Virtual Reality","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-26 18:24:44","doi":"10.21203/rs.3.rs-3828765/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"revise","date":"2024-03-22T04:11:10+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"This content is not available.","date":"2024-03-14T22:34:48+00:00","index":3,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2024-03-11T07:30:59+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2024-03-04T08:11:29+00:00","index":3,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2024-03-01T07:34:47+00:00","index":2,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2024-02-21T05:58:50+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2024-02-14T22:43:19+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewersInvited","content":"","date":"2024-01-23T19:02:32+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-01-03T01:09:35+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-01-02T09:35:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2024-01-02T06:50:32+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bebbe7f1-49b3-4b23-922c-f4b354cd30ea","owner":[],"postedDate":"January 26th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":28327681,"name":"Health sciences/Neurology/Neurological disorders/Dementia/Alzheimer's disease"},{"id":28327682,"name":"Biological sciences/Neuroscience/Cognitive neuroscience"}],"tags":[],"updatedAt":"2024-09-04T17:12:15+00:00","versionOfRecord":{"articleIdentity":"rs-3828765","link":"https://doi.org/10.1038/s41746-024-01206-5","journal":{"identity":"npj-digital-medicine","isVorOnly":false,"title":"npj Digital Medicine"},"publishedOn":"2024-08-22 04:00:00","publishedOnDateReadable":"August 22nd, 2024"},"versionCreatedAt":"2024-01-26 18:24:44","video":"","vorDoi":"10.1038/s41746-024-01206-5","vorDoiUrl":"https://doi.org/10.1038/s41746-024-01206-5","workflowStages":[]},"version":"v1","identity":"rs-3828765","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3828765","identity":"rs-3828765","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}