Machine Learning Prediction Models for Cognitive Impairment in Cerebral Small Vessel Disease

Machine Learning Prediction Models for Cognitive Impairment in Cerebral Small Vessel Disease | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine Learning Prediction Models for Cognitive Impairment in Cerebral Small Vessel Disease Qi Wu#, Jupeng Zhang#, Peng Lei, Xiqi Zhu, Changhui Huang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5365831/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Early identification of cerebral small vessel disease (CSVD) patients with a higher risk of developing cognitive impairment is essential for timely intervention and improvement of patient prognosis. The advancement of medical imaging and computing capabilities provides new methods for early detection of cognitive disorders. Machine learning (ML) has emerged as a promising technique for cognitive impairment in CSVD. This study aims to conduct a thorough meta-analysis and comparison of published ML prediction models for cognitive impairment in patients with CSVD. Methods In September 2024, relevant studies were retrieved from four databases: PubMed, Embase, Web of Science, and the Cochrane Library. The Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to assess the risk of bias of the ML models. The random effects model was used for meta-analysis of C-index, while a bivariate mixed-effects model was used to calculate the pooled sensitivity and specificity with their 95% confidence intervals (CIs). In addition, to limit the influence of heterogeneity, we also performed sensitivity analyses, a meta-regression, and subgroup analysis. Results Twenty-one prediction models from thirteen studies, involving 3444 patients met criteria for inclusion. The reported C-index ranged from 0.708 to 0.952. The pooled C-index, sensitivity, and specificity were 0.85 (95% CI 0.82–0.87), 0.82 (95% CI 0.77–0.87), and 0.81 (95% CI 0.73–0.87). As one of the most commonly used ML methods, logistic regression achieved a total merged C-index of 0.81, while non logistic regression models performed better with a total merged C-index of 0.86. Conclusions ML models holds significant promise in forecasting the risk of cognitive impairment in patients with CSVD. However, future high-quality research that externally validates the algorithm through prospective studies with larger, more diverse cohorts is needed before it can be introduced into clinical practice. cognitive impairment machine learning prediction model cerebral small vessel disease meta-analysis Figures Figure 1 Figure 2 Figure 3 1. Introduction Cerebral small vessel disease (CSVD) represents a heterogeneous group of disorders affecting the small blood vessels in the brain, including arterioles, capillaries, and venules (Bos et al., 2018 ; Li et al., 2022 ). These conditions are increasingly recognized as significant contributors to vascular cognitive impairment and dementia, posing a major public health challenge as populations age (Pantoni, 2010 ; Zanon Zotin et al., 2021 ). Despite its clinical importance, accurately detecting and quantifying cognitive impairment attributable to CSVD remains a complex and challenging task. There is an urgent need for effective strategies to identify individuals at risk of developing cognitive impairment (Das et al., 2019 ). The diagnosis of CSVD-related cognitive impairment requires a combination of clinical manifestations, neuropsychological evaluation, and neuroimaging examinations (Rosenberg et al., 2016 ). Traditional clinical assessments are subject to variability and may fail to capture the nuanced, yet critical changes in cognitive function that may occur in the early stages of disease (Li et al., 2024 ). Neuropsychological assessment takes a long time and is susceptible to subjective influences from both the subjects and evaluators. Consequently, there is a growing interest in the application of machine learning (ML) approaches that can leverage large datasets to uncover patterns and predict cognitive decline more effectively than conventional methods. The application of neuroimaging machine learning in CSVD mainly includes assisting diagnosis and disease prediction. Several meta-analyses have demonstrated the efficacy of ML models in predicting cognitive decline across various conditions. For instance, Odusami et al ( 2024 ) and Grueso et al (2021) provided comprehensive analysis of ML predicting progression from mild cognitive impairment to Alzheimer's disease, highlighting their potential for early detection. Moreover, Li et al. ( 2023 ) illustrated the feasibility of ML in the prediction of post-stroke cognitive impairment. However, ML has limitations and may lead to inaccurate predictions in certain clinical situations. Some researchers found that ML survival or classification models brought little improvement over traditional statistical methods, and the benefits of its assessment should be approached with caution, especially considering the limited sample size and features (R. Li et al., 2023 ; Noroozi et al., 2024 ). Recent work has shown that ML algorithms can classify CSVD and normal patients with high accuracy. Unfortunately, few evidence-based studies of ML models for CSVD-related cognitive impairment are currently available. As a result, this study is aim to analyze ML prediction models applied to neuroimaging data combined with other variables to predict cognitive impairment in CSVD and provide a useful reference for clinical practice and future research. 2. Methods This study was conducted according to the Preferred Reporting Items for a Systematic Review and Meta-analysis (PRISMA) 2020 guidelines (Page et al., 2021 ). It was registered on the PROSPERO website (CRD42024601473). 2.1 Search strategy Two researchers (QW and JZ) independently conducted a systematic search of the electronic databases MEDLINE (PubMed), Cochrane Library, Embase, and Web of Science, and the retrieval was as of 21 September 2024. The retrieval strategy was as follows: (“cerebral small vessel diseases” OR “subcortical ischemic vascular disease”) AND (“cognitive Dysfunction” OR “cognitive impairment”) AND (“artificial intelligence” OR “machine learning” OR “neural network” OR “deep learning”). The detailed strategies are available in the Supplementary Table S1 . In case of disagreement between assessors, consensus was reached through discussion and negotiation. We also identified additional relevant studies by reviewing the reference lists of the retrieved studies and review articles. This study followed the PICOTS system, recommended by the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) checklist (Palazón-Bru et al., 2020 ). The key items of our systematic review are described below: P (Population): Patients with CSVD. C (Comparator): ML methods as interventions. O (Outcome): the performance for the prediction and diagnosis of ML models, including model discrimination or concordance index (C-index), specificity, sensitivity, and area under the curve (AUC). T (Timing): The outcome was predicted after evaluating basic information at admission, clinical scoring scale results, and laboratory indicators. S (Setting): The use of the ML prediction model is to individualize the prediction of cognitive impairment in patients with CSVD, facilitating the implementation of preventive measures to prevent adverse events. 2.2. Inclusion and exclusion criteria All studies included had to meet the following criteria: (1) patients diagnosed with CSVD; (2) studies published in English; (3) ML was applied to predict CSVD-related cognitive impairment prediction, with a clear description of the ML models; (4) at least one measure of model performance (discrimination or calibration) was reported. Exclusion criteria were as follows: (1) only analysis of risk factors was conducted, without building complete ML models; (2) publication types such as review articles, case reports, editorials, conference abstracts and animal studies; (3) studies on the accuracy of single-factor prediction models; (4) studies that prediction models were developed, but not validated; (5) the full text could not be retrieved despite contacting the authors via email. 2.3. Literature screening EndNote 20 software (Clarivate Analytics, Philadelphia, PA, USA) were employed to manage the studies and remove duplicate items. Three investigators (QW JZ and PL) conducted literature screening and ensured that all studies met the inclusion criteria Each selected article has been screened at least twice. Any disagreement was dissolved by consulting a third reviewer (XZ). 2.4. Data extraction The data extracted from each article was categorized into two groups: (1) Basic information: name of the first author, publication year, country, study design, sample size, age, diagnostic criteria for cognitive impairment, number of model variables and modeling variables. (2) Model information: variable selection method, handling of missing value, ML algorithms, model validation method, model performance measures. The extraction of information was performed by one reviewer, then checked by another reviewer to ensure accuracy and consistency. 2.5. Quality and Bias Assessments The Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to assess the risk of bias and applicability of the included study. The PROBAST consists of 20 signaling questions in four distinct domains, namely participants, predictors, outcome, and analysis (Moons et al., 2019 ). Each signaling question can be answered as “yes”, “probably yes”, “no”, “probably no” or “no information”. If at least one signaling question in a domain is answered as “no” or “probably no”, that domain should be considered at high risk of bias. Only when all domains are judged as low risk of bias, the overall bias can be considered low risk (Fu et al., 2024 ). Two independent researchers (QW and JZ) evaluated the quality and the bias risk of the included studies. Any disagreements will be resolved by consensus by a third researcher (CH). 2.6. Outcome measures We extracted the C-index as the primary outcome measure, which can be used to reflect the overall accuracy of ML models. However, this indicator alone may not fully reflect the predictive accuracy of ML models. Therefore, sensitivity and specificity were included as complementary outcome measures to evaluate the predictive accuracy of ML in CSVD-related cognitive impairment. 2.7. Data synthesis and statistical analysis Stata version 15.1 (StataCorp LP, College Station, TX, USA) were used to conduct the meta-analyses. Given the differences in modeling variables and parameters, the C-index was preferred to be pooled using a random effects model using a random effects model while a bivariate mixed-effects model was used to calculate the pooled sensitivity and specificity. If the C-index did not report 95% CIs and standard error (SEs), we estimated the SEs using the methods by Debray (Debray et al., 2019 ). C-index is similar to the AUC (Nezic, 2020 ), indicates its diagnostic or prognostic discrimination ability as low (C-index ≤ 0.5), modest (C-index > 0.6 to 0.7), good (C-index > 0.7 to 0.8), or strong (C-index > 0.8) (Snell et al., 2018 ). Cochrane Q-test and I 2 statistics were performed to examine heterogeneity. The I 2 statistics provides a measure of heterogeneity, with values of 25%, 50%, and 75% indicating low, moderate, and high heterogeneity respectively (Higgins et al., 2003 ). Subgroup analyses, meta-regression and sensitivity analysis were also performed to gain insight into potential sources of heterogeneity. The C-index of the different ML algorithms for predicting cognitive impairment in patients with CSVD are discussed in the subgroup Analysis section. Sensitivity analysis was conducted by sequentially excluding each individual study and subsequently recalculating the pooled effect size for the remaining dataset. Publication bias was evaluated using Egger’s test, with p > 0.05 indicating a low publication bias (Egger et al., 1997 ). 3. Results 3.1. Study selection The study search process is illustrated in the PRISMA flowchart (Fig. 1 ). In total, 5603 records were obtained by performing electronic and manual searches. After removing duplicates and screening titles and abstracts, 89 articles remained. On the basis of full-text review, thirteen articles were included in the study. 3.2. Study characteristics Table 1 summarizes the design and participant characteristics of the included studies. The publication years of the article ranged from 2016 to 2024; 5 out of 13 (38%) were published in 2024 (Chen et al., 2024 ; Huang et al., 2024 ; Li et al., 2024 ; Zhang et al., 2024 ; Zhu et al., 2024 ). Of the thirteen eligible studies, ten were conducted in China (H. F. Chen et al., 2020 ; Chen et al., 2023 ; Huang et al., 2024 ; Li et al., 2024 ; M. Liu et al., 2022 ; Qin et al., 2023 ; Wang et al., 2021 ; L. Zhang et al., 2022 ; Zhang et al., 2024 ; Zhu et al., 2024 ), one in Italy (Ciulli et al., 2016 ), and two in the UK (Chen et al., 2024 ; R. Li et al., 2023 ). In the studies we included, eight were prospective (including three multicenter study) (Li et al., 2024 ; M. Liu et al., 2022 ; L. Zhang et al., 2022 ; Zhang et al., 2024 ; Zhu et al., 2024 ), and five were retrospective (H. F. Chen et al., 2020 ; Chen et al., 2023 ; Chen et al., 2024 ; Ciulli et al., 2016 ; Huang et al., 2024 ; R. Li et al., 2023 ; Qin et al., 2023 ; Wang et al., 2021 ), all conducted in single centers. The source data for ML training were mostly obtained from clinical hospitals; some also included public database. Among the 13 studies, the most commonly used diagnostic criteria for cognitive impairment were Montreal Cognitive Assessment (MoCA) (n = 6), followed by Mini-Mental State Examination (MMSE) (n = 2) and Diagnostic and Statistical Manual of Mental Disorders (DSM) (n = 2). Table 1 Overview of basic data of the included studies Author (year) Country Data source Study design Age, years Sample size, n Diagnostic criteria for cognitive impairment Included variables Model evaluation metrics Zhang, L (2022) China Neurology department of a hospital Retrospective study 71.06 ± 9.88 159 MoCA score < 26 (an additional 1 point for education < 12 years) hypertension, homocysteine, total CSVD MRI burden, and years of schooling C-index, AUC Huang (2024) China Clinical hospital Prospective study 57.28 ± 1.37 304 mild-CSVD: MoCA score 18–26; sever-CSVD: MoCA score < 17 Hippocampal textures AUC, Sensitivity Specificity, Accuracy, PPV, NPV Zhang, Y (2024) China Neurology department of a hospital Retrospective study 65.91 ± 7.90 71 MoCA score < 26 BEN, BEN/ ALFF, and BEN/ReHo AUC, Sensitivity Specificity, Accuracy Li, N (2024) China Neurology department of a hospital Retrospective cross-sectional study NR 377 MoCA score < 25 Age, hypertension, total CSVD burden, ApoA1 AUC, Sensitivity Specificity, Accuracy, PPV, NPV Zhu (2024) China Neurology department of a hospital Retrospective study 66.7 ± 6.99 227 MoCA score < 26 (an additional 1 point for education < 12 years) Age, homocysteine, hypertension, lacunar infarct score, total CSVD burden C-index Chen (2024) UK Multicentre (UKB, SCANS, RUN DMC) Prospective cohort study NR 761 NR Demographic factors (n = 3), Cognitive scores (n = 3) C-index Table 1 (continued) Author (year) Country Data source Study design Age, years Sample size, n Diagnostic criteria for cognitive impairment Included variables Model evaluation metrics Liu, M (2022) China Neurology department of a hospital Retrospective cohort study 50–80 197 neuropsychological tests scores fallen beyond ± 1.5 standard deviations (SDs) from the mean; multiscale features of the brain functional interactions AUC, Sensitivity Specificity, Accuracy Li (2023) UK, Netherland, Singapore Multicentre (SCANS, RUN DMC, HARMONISATION) Prospective cohort study NR 889 DSM Demographic features (n = 7), Imaging features (n = 5 Cognitive features (n = 3) AUC, Sensitivity Specificity, Accuracy, Precision Wang (2021) China Neurology department of a hospital Prospective study 64.71 ± 6.94 113 MoCA score, MMSE score, neuropsychological tests scores: 1.5 standard deviations below the normative mean DTI features (n = 8), CBF features (n = 5) AUC, Sensitivity Specificity, Accuracy Qin (2023) China Multicentre (3 hospitals) Prospective cohort study 64.8 ± 7.8 136 DSM-IV; CDR of ≥ 0.5 on at least one domain and a global score ≥ 0.5 DTI features, rsfMRI features, clinical features Neuropsychological measures Sensitivity, Precision Specificity, Accuracy Chen, X (2023) China Neurology department of a hospital Prospective study 70.25 100 Shanghai Manual of Cohort Studies on Cerebral Small Vessel Diseases HBP, DM, HLP, Smoke, Drink AUC, Sensitivity, Specificity, Accuracy PPV, NPV Chen, H (2020) China Neurology department of a hospital Prospective cross-sectional study 64.8 ± 1.57 70 Beijing version of the MoCA (MoCA-BJ), MMSE DTI features: AD, FA, MD, RD Sensitivity, Specificity, Accuracy Ciulli (2016) Italy Clinical hospital Prospective, randomized, single-blind 75.3 ± 6.8 40 TMT-B MD, FA Sensitivity, Specificity, Accuracy Abbreviations: NR: not reported, RF: random forests; AUC: area under the receiver operating characteristic curve; PPV: positive predictive value; NPV: negative predictive value; MMSE: Mini-Mental State Examination; MoCA: Montreal Cognitive Assessment; CDR, clinical dementia rating; DSM: Diagnostic and Statistical Manual of Mental Disorders; BEN: brain entropy, ALFF: amplitude of low frequency fluctuation, ReHo: regional homogeneity; SCANS: St George’s Cognition and Neuroimaging in Stroke; RUN DMC: Radboud University Nijmegen Diffusion Tensor and MRI Cohort; HARMONISATION: A memory clinic study in Singapore; DTI: diffusion tensor imaging; rsfMRI: resting-state fMRI CBF: cerebral blood flow; HBP: high blood pressure; DM: diabetes mellitus; HLP: hypercholesterolemia; TMT: Trail Making Test Part B Table 2 provides the model information in the included studies. There were 21 models in the included studies, among which 17 models reported AUC or C-statistic values, ranging from 0.708 to 0.952, and 10 models reported sensitivity and specificity. In the 13 studies, the logistic regression (LR) (38.4%; n = 5) was the most universally used algorithm, followed by the support-vector machine (SVM) (23%; n = 3). Besides, seven studies also assessed the performance of other ML algorithms: random forest (RF), k-nearest neighbors (KNN), decision trees (DT), connectome-based prediction model (CPM), DS-GAN (diffusion scalar generative adversarial network), Reg-Logistic, generalised matrix learning vector quantisation (GMLVQ), generalised relevance learning vector quantisation (GRLVQ), and unsupervised machine learning model. For measuring deep learning performance, AUC and the Youden index were most commonly used. Table 2 Methodological characteristics of the included machine learning models Author (year) Missing value processing Validation set generation method [internal validation/external validation] Variable screening/ feature selection method ML algorithms Model performance (C-index/ Sensitivity, Specificity) Zhang, L (2022) NR Bootstrapping Univariate significance level, multivariate logistic regression with stepwise regression (Forward: LR) LR C-index: 0.806 (0.735–0.877) Huang (2024) Median interpolation 5-fold cross-validation Spearman correlation, LASSO regression KNN C-index: 0.818 (0.673–0.96) Sensitivity: 0.538 Specificity: 0.947 Zhang, Y (2024) ML method leave-one-out cross validation SVM classifier SVM C-index: 0.824(0.750–0.899), Sensitivity: 0.865, Specificity: 0.618 Li, N (2024) NR 10-fold cross validation multivariate logistic regression analysis, LASSO regression LR C-index: 0.852 (0.781–0.923) Sensitivity: 0.769, Specificity: 0.74 Zhu (2024) Median interpolation Bootstrapping Univariate analysis, binary logistic regression LR C-index: 0.867(0.788–0.947) Chen (2024) NR External validation Information value DS-GAN (consist of 2 deep learning models: a generator and a discriminator) SCANS: C-index:0.828 (0.733, 0.923),0.903(0.828–0.978) RUN DMC: C-index:0.845(0.800–0.890),0.858(0.815–0.901) Liu, M (2022) NR 5-fold cross-validation SVM classifier, GCN CPM (with SVM as the classifier) and GCN C-index: 0.821(0.747–0.895), Sensitivity: 0.818, Specificity: 0.823 Table 2 (continued) Author (year) Missing value processing Validation set generation method [internal validation/external validation] Variable screening/ feature selection method ML algorithms Model performance (C-index/ Sensitivity, Specificity) Li (2023) Complete case analysis 5-fold cross-validation External validation Sparse logistic regression classifier, LASSO regression, feature ranking LR, SVM, Reg-Logistic, GMLVQ, GRLVQ C-index: LR 0.860 (0.826–0.894), Reg-Logistic 0.870(0.837–0.903) SVM 0.858(0.824–0.892) GMLVQ 0.817 (0.779–0.856) GRLVQ 0.868 (0.835–0.901) Wang (2021) NR leave-one-out cross validation Sparse logistic regression classifier, LASSO regression SLR C-index: 0.708 (0.667–0.740), Sensitivity: 0.77, Specificity: 0.641 Qin (2024) Remove highly missing variables, fill missings with model Internal validation External validation LASSO regression Unsupervised machine learning model Sensitivity: 0.795, Specificity: 0.962 Sensitivity: 0.711, Specificity: 0.938 Chen, X (2023) Complete case analysis 1,000 times cross-validation Statistical analysis, k-means clustering DT C-index: 0.952 (0.911–0.993) Sensitivity: 0. 981, Specificity: 0.896 Chen, H (2020) imputation Bootstrapping Spearman correlation, RF RF Sensitivity: 0.818, Specificity: 0.79 Ciulli (2016) NR leave-one-out cross validation Averaging Mean Diffusivity (MD) and Fractional Anisotropy (FA) maps within 50 Regions of Interest (ROIs) SVM Sensitivity: 0.895, Specificity: 0.714 Abbreviations: NR: not reported, ML: machine learning; LASSO: Least Absolute Shrinkage and Selection Operator; RF: random forest; KNN: k-nearest neighbors; CPM: connectome-based prediction model; GCN: graph convolutional network; LR: logistic regression; SVM: support-vector machine; Reg-Logistic: regularised logistic regression with elastic net penalty; GRLVQ: generalised relevance learning vector quantisation; GMLVQ: generalised matrix learning vector quantisation; DT: decision tree; SLR: sparse logistic regression DS-GAN: diffusion scalar generative adversarial network Table 3 Results of subgroup analyses of C-index by study design, validation set generation method and machine learning type Subgroups Models Pooled C-index (95%CI) P Value I 2 (%) Overall 17 0.85 (0.82, 0.87) P < 0.001 81.6 Study design Retrospective 5 0.83 (0.80, 0.87) P = 0.792 0 Prospective 12 0.85 (0.82, 0.88) P < 0.001 86.9 Validation method Internal validation 8 0.83 (0.76, 0.91) P < 0.001 91 External validation 9 0.86 (0.84, 0.87) P = 0.515 0 Model type Logistic regression 5 0.81 (0.74, 0.89) P < 0.001 89.8 Non–logistic regression 12 0.86 (0.84, 0.89) P < 0.001 63.5 Feature selection is an important step for ML training. Some researchers favored features that are statistically correlated and easily obtainable with CSVD (i.e., structural MRI, demographic, and cognitive results), whereas others incorporated factors based on established knowledge from previous models. In the included studies, 4 studies used DTI (diffusion tensor imaging) data as input features (H. F. Chen et al., 2020 ; Ciulli et al., 2016 ; Qin et al., 2023 ; Wang et al., 2021 ), 3 studies used rs-fMRI (resting-state functional MRI) data as input features (M. Liu et al., 2022 ; Qin et al., 2023 ; Zhang et al., 2024 ), and one used a radiomics model with hippocampal texture features (Huang et al., 2024 ). 3.3. Model validation In the included studies, all ML models were internally or externally validated. Among them, internal validation was conducted in ten studies using random split, k-fold cross-validation or bootstrapping, while the remaining three conducted external validation. Models that only developed without validation were not included. 3.4. Quality assessment Five studies had a high risk of bias (38.4%), while only three studies had low risk (23%). In the participant domain, two studies were considered to have a moderate risk of bias primarily due to the use of inappropriate data sources and controversial criteria (Chen et al., 2024 ; R. Li et al., 2023 ), while bias in the other studies were low. In the predictor domain, one study had an unclear risk of bias due to lack of information on how predictive factors are defined or measured (Li et al., 2024 ). The bias of outcome in one study was unclear due to no standard for diagnosing cognitive impairment described (Chen et al., 2024 ), and three was high because multiple predictive factors form part of the outcome of cognitive impairment (H. F. Chen et al., 2020 ; Li et al., 2024 ; Wang et al., 2021 ). In the analysis domain, the method of converting continuous variables into categorical variables in the two studies was not mentioned (H. F. Chen et al., 2020 ; Huang et al., 2024 ). six studies where missing data has not been properly processed (Chen et al., 2024 ; Ciulli et al., 2016 ; Li et al., 2024 ; M. Liu et al., 2022 ; Wang et al., 2021 ; L. Zhang et al., 2022 ), four studies did not consider model overfitting and underfitting (Chen et al., 2024 ; Ciulli et al., 2016 ; Qin et al., 2023 ; Wang et al., 2021 ). Therefore, these attributes should be improved in subsequent model construction. More details of the risk of bias are shown in Supplementary Table S2 . 3.5. Performance of ML Models for predicting cognitive impairment in CSVD Meta-analysis showed that the overall pooled C-index for ML models predicting cognitive impairment in CSVD was 0.85 (95% CI 0.82–0.87; P < 0.001; I 2 = 81.6%; n = 17); the pooled sensitivity was 0.82 (95% CI 0.77–0.87; P = 0.07; I 2 = 43.51%; n = 10) and the pooled specificity was 0.81 (95% CI 0.73–0.87; P < 0.001; I 2 = 72.15%; n = 10). Forest plots of the above results are presented in Figs. 2 , 3 . The results of meta-regression showed that four items including country where the data were collected, publication year, study design, and ML methods used did not significant affect the C-index (p > 0.05; Supplementary Table S3 ). 3.6. Subgroup analysis We performed the subgroup analysis of C-index based on study design, validation set generation method and model type. For study design, the pooled C-index of retrospective and prospective studies was 0.83 (95%CI 0.80–0.87, n = 5) and 0.85 (95%CI 0.82–0.88, n = 8), respectively. For validation set generation method, the pooled C-index of internal and external validation studies was 0.83 (95%CI 0.76–0.91, n = 8) and 0.86 (95%CI 0.84–0.87, n = 9), respectively. For ML algorithms, 5 prediction models were established applying the LR algorithms (Li et al., 2024 ; R. Li et al., 2023 ; Wang et al., 2021 ; L. Zhang et al., 2022 ; Zhu et al., 2024 ), and the overall pooled C-index for the LR models was 0.81 (95% CI 0.74–0.89). The overall pooled C-index for non-LR models was 0.86 (95% CI 0.84–0.89), the highest value among these subgroups. Forest plots of the above results are presented in supplemental file Figure S1 -3 . 3.7. Sensitivity analysis and publication bias The sensitivity analysis excluding studies one by one showed that the meta-analysis results of predictive value of ML for cognitive impairment in CSVD were stable and reliable ( Supplementary Figure S4 ). Egger’s test result (P = 0.439) and a symmetrically distributed funnel plot suggested no publication bias ( Supplementary Figure S5) . 4. Discussion Although the increasing use of ML in predicting cognitive impairment in CSVD, the quality of studies varies, and the results also differ. It is necessary to perform a meta-analysis for a unified evaluation. This study is the first to systematically explore the application of the latest ML prediction models for cognitive impairment in CSVD. 4.1. Principal findings In our study, the overall pooled estimation of 13 studies showed that ML models achieved high accuracy in early recognition of CSVD patients, with a C-index of 0.85 (95% CI 0.82–0.87) and sensitivity and specificity of > 80%. These results underscore the ability of ML algorithms to integrate complex data from various sources, such as neuroimaging, clinical assessments, and genetic markers, offering potential advantages. Although ML methods may not provide greater benefit than current available screening strategies, they do possess an advantage in allowing for a preferred trade-off between sensitivity and specificity (Z. Zhang et al., 2022 ). It was reported that common cognitive screening tests have similar predictive accuracy in CSVD-related cognitive impairment (Pasi et al., 2021 ). Research has shown that the MoCA has superior sensitivity of cognitive function than other screening tools, but its specificity is less than desirable (Jia et al., 2021 ; Szcześniak et al., 2021 ). However, our findings showed that ML has considerably high and balanced predictive accuracy (c-index, sensitivity, and specificity) in CSVD-related cognitive impairment and is robust for future clinical implementation. Based on the subgroup analysis, models developed utilizing non-LR methods achieved the highest C-index, implying that researchers should consider testing a broader range of candidate models to potentially identify superior predictive performance. Li et al (R. Li et al., 2023 ) developed five common ML methods (i.e., LR, SVM, Reg-Logistic, GMLVQ, GRLVQ) to predict the occurrence of CSVD-related cognitive impairment, with a data set of 889 patients. Reg-Logistic and GRLVQ were the two models with the best diagnostic accuracy in that study (C-index 0.868–0.87). Regarding the ML methods used to predict the risk of cognitive impairment in the context of CSVD, LR is the most commonly used type of model, since LR models are suitable for simple data with linear relationships between variables and outcomes (Boehm-Sturm et al., 2017 ; B. Liu et al., 2022 ; Montine et al., 2021 ). Ensemble methods like RF have achieved high precision in predicting cognitive trajectory types using longitudinal by aggregating the results of multiple decision trees, thus reducing overfitting and increasing robustness (Mohammadiarvejeh et al., 2023 ). Moreover, SVM and KNN have more advantages in image recognition and segmentation for predicting mild cognitive impairment (Al-Qazzaz et al., 2018 ; Zhang et al., 2021 ). Deep learning exhibits significant potential in the realm of image classification, attributed to its capability to automatically extract low to high-level features. Studies have demonstrated the utility of deep learning models in analyzing MRI sequence to identify subtle changes subcortical vascular cognitive impairment (Q. Chen et al., 2020 ; Wang et al., 2019 ). The most commonly used features in ML models are neuroimaging biomarkers such as white matter hyperintensities (WMHs), cerebral microbleeds (CMBs) lacunar infarction (LI), enlarged perivascular spaces (EPVS) and total CSVD burden. The preliminary research on neuroimaging biomarkers provides the imaging basis of CSVD for machine learning and can serve as primary features for input. In addition, DTI and rs-fMRI are widely applied in the study of CSVD, primarily focusing on the early identification and progression of cognitive impairment (Egle et al., 2022 ). Based on DTI and rs-fMRI parameters, structural and functional brain networks can be constructed separately. The fractional anisotropy (FA) score of DTI in CSVD patients decreases, while the mean diffusivity (MD) increases (Biesbroek et al., 2017 ), and these changes are associated with clinical outcomes such as cognitive impairment (Tuladhar et al., 2015 ), gait abnormalities (van der Holst et al., 2018 ), and emotional disorders (van Uden et al., 2015 ). Research based on rs-fMRI has found that CSVD patients exhibit abnormal functional activity in the prefrontal cortex, subcortical region, cingulate gyrus, and hippocampus (Li et al., 2012 ; Yi et al., 2012 ). Ciulli et al. utilized the mean MD value as a feature and employed a machine learning strategy combining a linear support vector machine with leave-one-out cross validation to predict the executive function of CSVD patients with mild cognitive impairment, in which sensitivity, specificity, and accuracy all reached a high level (Ciulli et al., 2016 ). Furthermore, Pantoni et al. used traditional imaging biomarkers and the fractal dimension (FD) of gray and white matter as input features to predict different neuropsychological test scores. They found that the FD value of white matter was the most commonly used feature, and FD may be an important marker for predicting cognitive decline in CSVD, which can supplement traditional imaging biomarkers (Pantoni et al., 2019 ). Our study also finds that radiomics features are frequently used as predictors in ML models for CSVD-related cognitive impairment. A recent study validated the superior accuracy and specificity of radiomics models utilizing hippocampal texture features in identifying CSVD patients with severe cognitive impairment (Huang et al., 2024 ). As for validation methods, 8 studies selected cross-validation, with different folding or iteration times. Cross-validation splits the samples into training set and testing set. This division can average the accuracy of each iteration to obtain a more robust model performance, rather than validating the model on just one test sample. 13 studies chose leave-one-out cross validation, in which case the model is trained using all data except for one data point, and then it attempts to classify the missing data points and applies the same operation to the rest of the sample in subsequent iterations. 2 studies used bootstrapping as a validation method. The model in most studies primarily relied on internal validation rather than external validation. The lack of external validation of the obtained results is considered a major limitation of current prediction models (van Kempen et al., 2021 ). The reason for this phenomenon may be due to the insufficient sample sizes for external validation. Subgroup analysis according to different validation methods showed that the pooled C-index of the external validation group were higher than those of the internal validation. This may be because the dataset used for external validation is large, allowing ML algorithms to be fully learned and trained. 4.2. Limitations and strengths This study has several potential limitations. First, non-English studies were not included in the meta-analysis, which might lead to selection bias. Second, a significant proportion of the included studies were conducted in mainland China, which may limit the generalizability of the findings to Western populations. Therefore, it is crucial for future research to develop models for cognitive impairment in CSVD patients across diverse populations to ensure their relevance and applicability in varied clinical settings. Thirdly, the heterogeneity in study designs, sample sizes, outcome measures, and a diverse array of feature selection methods across the included studies introduced potential biases and may affect the performance and applicability of ML models. Lastly, the number of validated studies and models included is relatively low, with only 3 models have external validation. Internal validation of data can result in overfitting or spectrum bias, thereby potentially overestimating the performance of machine learning models (Hu et al., 2023 ; Kuo et al., 2022 ). Despite the aforementioned limitations, this study can still offer valuable guidance for future research and clinical practice. In the first place, the major strength of our study lies in the rigorous literature search methodology and sufficiently detailed meta-analysis to be reproducible. We performed a quantitative synthesis, unprecedented in prior research, to systematically compare the models in question and described the characteristics of the validated ML models. In addition, the novel PROBAST was used to assess the risk of bias of the included predictive models. We conducted a thorough examination of more details of the model, such as data sources, processing, feature selection, and model validation (Di Tanna et al., 2020 ). Finally, we found that 5 studies had a high risk of bias in methodological quality, which may lead to overfitted prediction models. Potential biases may be prevented if studies adhere to the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) initiative in their reporting (Collins et al., 2015 ). 4.3. Future perspectives In order to make ML algorithms more robust in clinical implementation, there are some important issues that should be addressed in future research. Firstly, the multimodal imaging research of CSVD has gradually shifted from structural imaging biomarkers, to parameter analysis based on DTI and rs-fMRI, and then to constructing brain networks, leading people's understanding of CSVD from local to global. In the future, multimodal data including DTI and fMRI can be conducted to further explore the occurrence and development patterns of CSVD. Secondly, studies that achieved higher accuracy have in common the use of multidimensional and multimodal data, as well as increasingly complex classification methods. However, a balance is needed between the algorithms that achieve the higher performance, and the data and methods that more easily obtainable data in the clinical practice. In addition, although these algorithms are useful in distinguishing the brain features of CSVD, their performance is far from specific enough to hand over a complete diagnosis to automated methods. Nevertheless, when implemented in clinical practice, computer-aided diagnosis will provide a faster, easier to perform, and earlier detection method to predict the potential progression of CSVD patients to cognitive impairment. Consequently, it is recommended that clinicians enhance their capability to apply ML techniques to improve the accuracy of their diagnoses (Nagendran et al., 2020 ; Wu et al., 2023 ). Finally, the effectiveness and performance of machine learning are closely related to the size of the dataset. Therefore, Future research should prioritize the development of novel models that leverage larger sample sizes and employ rigorous study designs. Emphasis should be placed on multicenter external validation to ensure the generalizability and robustness of findings across diverse populations. 5. Conclusion This study shows the potential of ML algorithms predictive performance for cognitive impairment in patients with CSVD and can be used as a potential tool for early identification of cognitive impairment in this population. The development and validation of ML models tailored to CSVD-related cognitive impairment could facilitate earlier diagnosis and personalized treatment strategies, potentially improving patient outcomes. This study also analyzed the classification algorithms and feature source used in the selected studies, making diverse methodologies available to researchers. Our research findings also emphasize that the main limitation of the current prediction model is the lack of external validation. Additionally, this meta-analysis excluded studies that only developed models without validation and did not report model performance metrics. Overall, while the study provides important insights into the potential of machine learning for diagnosing cognitive impairment in patients with CSVD, more research is needed to fully explore and validate these approaches. Declarations Author contributions XZ and HC contributed to the study concept and design. QW and JZ retrieved and filtered the articles. QW and JZ extracted the data. QW and JZ analyzed the data. JZ, QW, and PL interpreted the data. JZ and QW drafted the manuscript. XZ and HC contributed to the critical revision of the manuscript. All authors contributed to the paper and approved the submitted version. Funding This study was financially supported by the National Natural Science Foundation of China (Grant No. 82460226), the Natural Science Foundation of Guangxi Autonomous Region (Grant No. 2023GXNSFAA026383). 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. Data availability statement Not applicable References Al-Qazzaz, N. K., Ali, S., Ahmad, S. A., Islam, M. S., & Escudero, J. (2018). Discrimination of stroke-related mild cognitive impairment and vascular dementia using EEG signal analysis. Med Biol Eng Comput , 56 (1), 137-157. https://doi.org/10.1007/s11517-017-1734-7 Biesbroek, J. M., Leemans, A., den Bakker, H., Duering, M., Gesierich, B., Koek, H. L.,…Biessels, G. J. (2017). Microstructure of Strategic White Matter Tracts and Cognition in Memory Clinic Patients with Vascular Brain Injury. Dement Geriatr Cogn Disord , 44 (5-6), 268-282. https://doi.org/10.1159/000485376 Boehm-Sturm, P., Füchtemeier, M., Foddis, M., Mueller, S., Trueman, R. C., Zille, M.,…Farr, T. D. (2017). Neuroimaging Biomarkers Predict Brain Structural Connectivity Change in a Mouse Model of Vascular Cognitive Impairment. Stroke , 48 (2), 468-475. https://doi.org/10.1161/strokeaha.116.014394 Bos, D., Wolters, F. J., Darweesh, S. K. L., Vernooij, M. W., de Wolf, F., Ikram, M. A., & Hofman, A. (2018). Cerebral small vessel disease and the risk of dementia: A systematic review and meta-analysis of population-based evidence. Alzheimers Dement , 14 (11), 1482-1492. https://doi.org/10.1016/j.jalz.2018.04.007 Chen, H. F., Huang, L. L., Li, H. Y., Qian, Y., Yang, D., Qing, Z.,…Xu, Y. (2020). Microstructural disruption of the right inferior fronto-occipital and inferior longitudinal fasciculus contributes to WMH-related cognitive impairment. CNS Neurosci Ther , 26 (5), 576-588. https://doi.org/10.1111/cns.13283 Chen, Q., Wang, Y., Qiu, Y., Wu, X., Zhou, Y., & Zhai, G. (2020). A Deep Learning-Based Model for Classification of Different Subtypes of Subcortical Vascular Cognitive Impairment With FLAIR. Front Neurosci , 14 , 557. https://doi.org/10.3389/fnins.2020.00557 Chen, X., Fang, Z., Zhao, Y., Cheng, W., Chen, H., Li, G.,…Yin, Y. (2023). Sleep Parameters and Plasma Biomarkers for Cognitive Impairment Evaluation in Patients With Cerebral Small Vessel Disease. J Gerontol B Psychol Sci Soc Sci , 78 (2), 210-219. https://doi.org/10.1093/geronb/gbac137 Chen, Y., Tozer, D., Li, R., Li, H., Tuladhar, A., De Leeuw, F. E., & Markus, H. S. (2024). Improved Dementia Prediction in Cerebral Small Vessel Disease Using Deep Learning-Derived Diffusion Scalar Maps From T1. Stroke , 55 (9), 2254-2263. https://doi.org/10.1161/strokeaha.124.047449 Ciulli, S., Citi, L., Salvadori, E., Valenti, R., Poggesi, A., Inzitari, D.,…Diciotti, S. (2016). Prediction of Impaired Performance in Trail Making Test in MCI Patients With Small Vessel Disease Using DTI Data. IEEE J Biomed Health Inform , 20 (4), 1026-1033. https://doi.org/10.1109/jbhi.2016.2537808 Collins, G. S., Reitsma, J. B., Altman, D. G., & Moons, K. G. (2015). Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. Bmj , 350 , g7594. https://doi.org/10.1136/bmj.g7594 Das, A. S., Regenhardt, R. W., Vernooij, M. W., Blacker, D., Charidimou, A., & Viswanathan, A. (2019). Asymptomatic Cerebral Small Vessel Disease: Insights from Population-Based Studies. J Stroke , 21 (2), 121-138. https://doi.org/10.5853/jos.2018.03608 Debray, T. P., Damen, J. A., Riley, R. D., Snell, K., Reitsma, J. B., Hooft, L.,…Moons, K. G. (2019). A framework for meta-analysis of prediction model studies with binary and time-to-event outcomes. Stat Methods Med Res , 28 (9), 2768-2786. https://doi.org/10.1177/0962280218785504 Di Tanna, G. L., Wirtz, H., Burrows, K. L., & Globe, G. (2020). Evaluating risk prediction models for adults with heart failure: A systematic literature review. PLoS One , 15 (1), e0224135. https://doi.org/10.1371/journal.pone.0224135 Egger, M., Davey Smith, G., Schneider, M., & Minder, C. (1997). Bias in meta-analysis detected by a simple, graphical test. Bmj , 315 (7109), 629-634. https://doi.org/10.1136/bmj.315.7109.629 Egle, M., Hilal, S., Tuladhar, A. M., Pirpamer, L., Bell, S., Hofer, E.,…Markus, H. S. (2022). Determining the OPTIMAL DTI analysis method for application in cerebral small vessel disease. Neuroimage Clin , 35 , 103114. https://doi.org/10.1016/j.nicl.2022.103114 Fu, H., Hou, D., Xu, R., You, Q., Li, H., Yang, Q.,…Bai, D. (2024). Risk prediction models for deep venous thrombosis in patients with acute stroke: A systematic review and meta-analysis. Int J Nurs Stud , 149 , 104623. https://doi.org/10.1016/j.ijnurstu.2023.104623 Grueso, S., & Viejo-Sobera, R. (2021). Machine learning methods for predicting progression from mild cognitive impairment to Alzheimer's disease dementia: a systematic review. Alzheimers Res Ther , 13 (1), 162. https://doi.org/10.1186/s13195-021-00900-w Higgins, J. P., Thompson, S. G., Deeks, J. J., & Altman, D. G. (2003). Measuring inconsistency in meta-analyses. Bmj , 327 (7414), 557-560. https://doi.org/10.1136/bmj.327.7414.557 Hu, J., Wang, Y., Guo, D., Qu, Z., Sui, C., He, G.,…Liu, X. (2023). Diagnostic performance of magnetic resonance imaging-based machine learning in Alzheimer's disease detection: a meta-analysis. Neuroradiology , 65 (3), 513-527. https://doi.org/10.1007/s00234-022-03098-2 Huang, B., Zheng, W., Mu, R., Yang, P., Li, X., Liu, F.,…Zhu, X. (2024). Disparities in the diagnostic efficacy of radiomics models in predicting various degrees of cognitive impairment in patients with cerebral small vessel disease. BMC Med Imaging , 24 (1), 257. https://doi.org/10.1186/s12880-024-01431-0 Jia, X., Wang, Z., Huang, F., Su, C., Du, W., Jiang, H.,…Zhang, B. (2021). A comparison of the Mini-Mental State Examination (MMSE) with the Montreal Cognitive Assessment (MoCA) for mild cognitive impairment screening in Chinese middle-aged and older population: a cross-sectional study. BMC Psychiatry , 21 (1), 485. https://doi.org/10.1186/s12888-021-03495-6 Kuo, R. Y. L., Harrison, C., Curran, T. A., Jones, B., Freethy, A., Cussons, D.,…Furniss, D. (2022). Artificial Intelligence in Fracture Detection: A Systematic Review and Meta-Analysis. Radiology , 304 (1), 50-62. https://doi.org/10.1148/radiol.211785 Li, C., Zheng, J., & Wang, J. (2012). An fMRI study of prefrontal cortical function in subcortical ischemic vascular cognitive impairment. Am J Alzheimers Dis Other Demen , 27 (7), 490-495. https://doi.org/10.1177/1533317512455841 Li, N., Gao, Y., Li, L. T., Hu, Y. D., Ling, L., Jia, N.,…Jiang, Y. (2024). Development and validation of a nomogram predictive model for cognitive impairment in cerebral small vessel disease: a comprehensive retrospective analysis. Front Neurol , 15 , 1373306. https://doi.org/10.3389/fneur.2024.1373306 Li, R., Harshfield, E. L., Bell, S., Burkhart, M., Tuladhar, A. M., Hilal, S.,…Markus, H. S. (2023). Predicting incident dementia in cerebral small vessel disease: comparison of machine learning and traditional statistical models. Cereb Circ Cogn Behav , 5 , 100179. https://doi.org/10.1016/j.cccb.2023.100179 Li, X., Chen, Z., Jiao, H., Wang, B., Yin, H., Chen, L.,…Qin, D. (2023). Machine learning in the prediction of post-stroke cognitive impairment: a systematic review and meta-analysis. Front Neurol , 14 , 1211733. https://doi.org/10.3389/fneur.2023.1211733 Li, Y., Kalpouzos, G., Laukka, E. J., Dekhtyar, S., Bäckman, L., Fratiglioni, L., & Qiu, C. (2022). Progression of neuroimaging markers of cerebral small vessel disease in older adults: A 6-year follow-up study. Neurobiol Aging , 112 , 204-211. https://doi.org/10.1016/j.neurobiolaging.2022.01.006 Liu, B., Meng, S., Cheng, J., Zeng, Y., Zhou, D., Deng, X.,…Li, C. (2022). Diagnosis of Subcortical Ischemic Vascular Cognitive Impairment With No Dementia Using Radiomics of Cerebral Cortex and Subcortical Nuclei in High-Resolution T1-Weighted MR Imaging. Front Oncol , 12 , 852726. https://doi.org/10.3389/fonc.2022.852726 Liu, M., Wang, Y., Zhang, H., Yang, Q., Shi, F., Zhou, Y., & Shen, D. (2022). Multiscale functional connectome abnormality predicts cognitive outcomes in subcortical ischemic vascular disease. Cereb Cortex , 32 (21), 4641-4656. https://doi.org/10.1093/cercor/bhab507 Mohammadiarvejeh, P., Klinedinst, B. S., Wang, Q., Li, T., Larsen, B., Pollpeter, A.,…Willette, A. A. (2023). Bioenergetic and vascular predictors of potential super-ager and cognitive decline trajectories-a UK Biobank Random Forest classification study. Geroscience , 45 (1), 491-505. https://doi.org/10.1007/s11357-022-00657-6 Montine, T. J., Bukhari, S. A., & White, L. R. (2021). Cognitive Impairment in Older Adults and Therapeutic Strategies. Pharmacol Rev , 73 (1), 152-162. https://doi.org/10.1124/pharmrev.120.000031 Moons, K. G. M., Wolff, R. F., Riley, R. D., Whiting, P. F., Westwood, M., Collins, G. S.,…Mallett, S. (2019). PROBAST: A Tool to Assess Risk of Bias and Applicability of Prediction Model Studies: Explanation and Elaboration. Ann Intern Med , 170 (1), W1-w33. https://doi.org/10.7326/m18-1377 Nagendran, M., Chen, Y., Lovejoy, C. A., Gordon, A. C., Komorowski, M., Harvey, H.,…Maruthappu, M. (2020). Artificial intelligence versus clinicians: systematic review of design, reporting standards, and claims of deep learning studies. Bmj , 368 , m689. https://doi.org/10.1136/bmj.m689 Nezic, D. G. (2020). Assessing the performance of risk prediction models. Eur J Cardiothorac Surg , 58 (2), 401. https://doi.org/10.1093/ejcts/ezaa071 Noroozi, M., Gholami, M., Sadeghsalehi, H., Behzadi, S., Habibzadeh, A., Erabi, G.,…Deravi, N. (2024). Machine and deep learning algorithms for classifying different types of dementia: A literature review. Appl Neuropsychol Adult , 1-15. https://doi.org/10.1080/23279095.2024.2382823 Odusami, M., Maskeliūnas, R., Damaševičius, R., & Misra, S. (2024). Machine learning with multimodal neuroimaging data to classify stages of Alzheimer's disease: a systematic review and meta-analysis. Cogn Neurodyn , 18 (3), 775-794. https://doi.org/10.1007/s11571-023-09993-5 Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D.,…Moher, D. (2021). The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. Bmj , 372 , n71. https://doi.org/10.1136/bmj.n71 Palazón-Bru, A., Martín-Pérez, F., Mares-García, E., Beneyto-Ripoll, C., Gil-Guillén, V. F., Pérez-Sempere, Á., & Carbonell-Torregrosa, M. (2020). A general presentation on how to carry out a CHARMS analysis for prognostic multivariate models. Stat Med , 39 (23), 3207-3225. https://doi.org/10.1002/sim.8660 Pantoni, L. (2010). Cerebral small vessel disease: from pathogenesis and clinical characteristics to therapeutic challenges. Lancet Neurol , 9 (7), 689-701. https://doi.org/10.1016/s1474-4422(10)70104-6 Pantoni, L., Marzi, C., Poggesi, A., Giorgio, A., De Stefano, N., Mascalchi, M.,…Diciotti, S. (2019). Fractal dimension of cerebral white matter: A consistent feature for prediction of the cognitive performance in patients with small vessel disease and mild cognitive impairment. Neuroimage Clin , 24 , 101990. https://doi.org/10.1016/j.nicl.2019.101990 Pasi, M., Sugita, L., Xiong, L., Charidimou, A., Boulouis, G., Pongpitakmetha, T.,…Biffi, A. (2021). Association of Cerebral Small Vessel Disease and Cognitive Decline After Intracerebral Hemorrhage. Neurology , 96 (2), e182-e192. https://doi.org/10.1212/wnl.0000000000011050 Qin, Q., Qu, J., Yin, Y., Liang, Y., Wang, Y., Xie, B.,…Tang, Y. (2023). Unsupervised machine learning model to predict cognitive impairment in subcortical ischemic vascular disease. Alzheimers Dement , 19 (8), 3327-3338. https://doi.org/10.1002/alz.12971 Rosenberg, G. A., Wallin, A., Wardlaw, J. M., Markus, H. S., Montaner, J., Wolfson, L.,…Hachinski, V. (2016). Consensus statement for diagnosis of subcortical small vessel disease. J Cereb Blood Flow Metab , 36 (1), 6-25. https://doi.org/10.1038/jcbfm.2015.172 Snell, K. I., Ensor, J., Debray, T. P., Moons, K. G., & Riley, R. D. (2018). Meta-analysis of prediction model performance across multiple studies: Which scale helps ensure between-study normality for the C-statistic and calibration measures? Stat Methods Med Res , 27 (11), 3505-3522. https://doi.org/10.1177/0962280217705678 Szcześniak, D., Rymaszewska, J., Zimny, A., Sąsiadek, M., Połtyn-Zaradna, K., Smith, E. E.,…Szuba, A. (2021). "Cerebral small vessel disease and other influential factors of cognitive impairment in the middle-aged: a long-term observational cohort PURE-MIND study in Poland". Geroscience , 43 (1), 279-295. https://doi.org/10.1007/s11357-020-00271-4 Tuladhar, A. M., van Norden, A. G., de Laat, K. F., Zwiers, M. P., van Dijk, E. J., Norris, D. G., & de Leeuw, F. E. (2015). White matter integrity in small vessel disease is related to cognition. Neuroimage Clin , 7 , 518-524. https://doi.org/10.1016/j.nicl.2015.02.003 van der Holst, H. M., Tuladhar, A. M., Zerbi, V., van Uden, I. W. M., de Laat, K. F., van Leijsen, E. M. C.,…de Leeuw, F. E. (2018). White matter changes and gait decline in cerebral small vessel disease. Neuroimage Clin , 17 , 731-738. https://doi.org/10.1016/j.nicl.2017.12.007 van Kempen, E. J., Post, M., Mannil, M., Kusters, B., Ter Laan, M., Meijer, F. J. A., & Henssen, D. (2021). Accuracy of Machine Learning Algorithms for the Classification of Molecular Features of Gliomas on MRI: A Systematic Literature Review and Meta-Analysis. Cancers (Basel) , 13 (11). https://doi.org/10.3390/cancers13112606 van Uden, I. W., Tuladhar, A. M., de Laat, K. F., van Norden, A. G., Norris, D. G., van Dijk, E. J.,…de Leeuw, F. E. (2015). White matter integrity and depressive symptoms in cerebral small vessel disease: The RUN DMC study. Am J Geriatr Psychiatry , 23 (5), 525-535. https://doi.org/10.1016/j.jagp.2014.07.002 Wang, Y., Lu, P., Zhan, Y., Wu, X., Qiu, Y., Wang, Z.,…Zhou, Y. (2021). The Contribution of White Matter Diffusion and Cortical Perfusion Pathology to Vascular Cognitive Impairment: A Multimode Imaging-Based Machine Learning Study. Front Aging Neurosci , 13 , 687001. https://doi.org/10.3389/fnagi.2021.687001 Wang, Y., Tu, D., Du, J., Han, X., Sun, Y., Xu, Q.,…Zhou, Y. (2019). Classification of Subcortical Vascular Cognitive Impairment Using Single MRI Sequence and Deep Learning Convolutional Neural Networks. Front Neurosci , 13 , 627. https://doi.org/10.3389/fnins.2019.00627 Wu, Y., Chao, J., Bao, M., & Zhang, N. (2023). Predictive value of machine learning on fracture risk in osteoporosis: a systematic review and meta-analysis. BMJ Open , 13 (12), e071430. https://doi.org/10.1136/bmjopen-2022-071430 Yi, L., Wang, J., Jia, L., Zhao, Z., Lu, J., Li, K.,…Han, Y. (2012). Structural and functional changes in subcortical vascular mild cognitive impairment: a combined voxel-based morphometry and resting-state fMRI study. PLoS One , 7 (9), e44758. https://doi.org/10.1371/journal.pone.0044758 Zanon Zotin, M. C., Sveikata, L., Viswanathan, A., & Yilmaz, P. (2021). Cerebral small vessel disease and vascular cognitive impairment: from diagnosis to management. Curr Opin Neurol , 34 (2), 246-257. https://doi.org/10.1097/wco.0000000000000913 Zhang, L., Gao, F., Zhang, Y., Hu, P., Yao, Y., Zhang, Q.,…Zhang, Y. (2022). Analysis of risk factors for the development of cognitive dysfunction in patients with cerebral small vessel disease and the construction of a predictive model. Front Neurol , 13 , 944205. https://doi.org/10.3389/fneur.2022.944205 Zhang, Q., Li, J., Bian, M., He, Q., Shen, Y., Lan, Y., & Huang, D. (2021). Retinal Imaging Techniques Based on Machine Learning Models in Recognition and Prediction of Mild Cognitive Impairment. Neuropsychiatr Dis Treat , 17 , 3267-3281. https://doi.org/10.2147/ndt.S333833 Zhang, Y., Hu, M., Fan, S., Cao, S., Du, B., Yin, S.,…Wei, Q. (2024). Altered Resting-State Brain Entropy in Cerebral Small Vessel Disease Patients with Cognitive Impairment. Brain Connect . https://doi.org/10.1089/brain.2024.0007 Zhang, Z., Yang, L., Han, W., Wu, Y., Zhang, L., Gao, C.,…Wu, H. (2022). Machine Learning Prediction Models for Gestational Diabetes Mellitus: Meta-analysis. J Med Internet Res , 24 (3), e26634. https://doi.org/10.2196/26634 Zhu, F., Yao, J., Feng, M., & Sun, Z. (2024). Establishment and evaluation of a clinical prediction model for cognitive impairment in patients with cerebral small vessel disease. BMC Neurosci , 25 (1), 35. https://doi.org/10.1186/s12868-024-00883-y Additional Declarations No competing interests reported. Supplementary Files Supplementary.docx Cite Share Download PDF Status: Posted Version 1 posted 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-5365831","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":376371402,"identity":"2d040a16-3ce3-4874-9b81-1bcde95e3316","order_by":0,"name":"Qi Wu#","email":"","orcid":"","institution":"Affiliated Hospital of Youjiang Medical university for Nationalities","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Wu#","suffix":""},{"id":376371403,"identity":"0c034749-09fd-4f0d-be32-70d6a02e2f24","order_by":1,"name":"Jupeng Zhang#","email":"","orcid":"","institution":"Affiliated Hospital of Youjiang Medical university for Nationalities","correspondingAuthor":false,"prefix":"","firstName":"Jupeng","middleName":"","lastName":"Zhang#","suffix":""},{"id":376371404,"identity":"f0575afb-3445-4664-b021-d41a57890995","order_by":2,"name":"Peng Lei","email":"","orcid":"","institution":"Affiliated Hospital of Youjiang Medical university for Nationalities","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Lei","suffix":""},{"id":376371405,"identity":"eba68e2f-0f7d-414e-98e0-74b53d7986ea","order_by":3,"name":"Xiqi Zhu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAu0lEQVRIiWNgGAWjYDACZgaGAyDEz8x8+AFpWiTb2dIMSLHrAIPBeR4FCaLUGhxn3niA4c+dxM2HeRgMGGpsoglqkWxmKzjAwPMscdth3gMPGI6l5TYQ0sLPzGNwgEHiMFALX4IBY8NhwlrYwFoMDidubuYxkCBKC8SWhMOJG5iJ1QL2S8KBw8YzDgMDOYEYvxicP7z5w4c/h2X7+w8ffvChxoawFpAuhgQYMwG3KjQto2AUjIJRMArwAgD8Oz/9Wx2GPAAAAABJRU5ErkJggg==","orcid":"","institution":"Affiliated Hospital of Youjiang Medical university for Nationalities","correspondingAuthor":true,"prefix":"","firstName":"Xiqi","middleName":"","lastName":"Zhu","suffix":""},{"id":376371406,"identity":"32205d48-89f8-418b-8b35-fd3cfabdd262","order_by":4,"name":"Changhui Huang","email":"","orcid":"","institution":"Affiliated Hospital of Youjiang Medical university for Nationalities","correspondingAuthor":false,"prefix":"","firstName":"Changhui","middleName":"","lastName":"Huang","suffix":""}],"badges":[],"createdAt":"2024-10-31 08:09:30","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5365831/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5365831/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":70285948,"identity":"1b3a5080-906f-49d6-947d-abe1b584d015","added_by":"auto","created_at":"2024-12-01 16:26:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":5150384,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram of the retrieval process\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-5365831/v1/d41ec4109ca5e355153b2b0d.png"},{"id":70286319,"identity":"78f13512-abf3-43fa-a353-76c06c6ba6e6","added_by":"auto","created_at":"2024-12-01 16:34:17","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":585287,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eForest plot of pooled C-index\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-5365831/v1/284912c78a95261363a575e7.png"},{"id":70285945,"identity":"926db244-893b-48dd-bdbf-9e60b88e0d97","added_by":"auto","created_at":"2024-12-01 16:26:17","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":816125,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eForest plot of pooled sensitivity and specificity\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-5365831/v1/2f08b5b2b5ac9bd73c1a6753.png"},{"id":79660450,"identity":"743f0fe0-d8aa-4b97-90dc-3f7028ece913","added_by":"auto","created_at":"2025-04-01 09:32:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6492893,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5365831/v1/ab266c1a-4ccf-4771-8c52-0f72a0b38386.pdf"},{"id":70285946,"identity":"3c5a1480-45ce-472c-8f5c-204aca5ef5db","added_by":"auto","created_at":"2024-12-01 16:26:17","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1254807,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-5365831/v1/33b26a5e35f19a484e129b14.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning Prediction Models for Cognitive Impairment in Cerebral Small Vessel Disease","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eCerebral small vessel disease (CSVD) represents a heterogeneous group of disorders affecting the small blood vessels in the brain, including arterioles, capillaries, and venules (Bos et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). These conditions are increasingly recognized as significant contributors to vascular cognitive impairment and dementia, posing a major public health challenge as populations age (Pantoni, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Zanon Zotin et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Despite its clinical importance, accurately detecting and quantifying cognitive impairment attributable to CSVD remains a complex and challenging task. There is an urgent need for effective strategies to identify individuals at risk of developing cognitive impairment (Das et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe diagnosis of CSVD-related cognitive impairment requires a combination of clinical manifestations, neuropsychological evaluation, and neuroimaging examinations (Rosenberg et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Traditional clinical assessments are subject to variability and may fail to capture the nuanced, yet critical changes in cognitive function that may occur in the early stages of disease (Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Neuropsychological assessment takes a long time and is susceptible to subjective influences from both the subjects and evaluators. Consequently, there is a growing interest in the application of machine learning (ML) approaches that can leverage large datasets to uncover patterns and predict cognitive decline more effectively than conventional methods.\u003c/p\u003e \u003cp\u003eThe application of neuroimaging machine learning in CSVD mainly includes assisting diagnosis and disease prediction. Several meta-analyses have demonstrated the efficacy of ML models in predicting cognitive decline across various conditions. For instance, Odusami et al (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) and Grueso et al (2021) provided comprehensive analysis of ML predicting progression from mild cognitive impairment to Alzheimer's disease, highlighting their potential for early detection. Moreover, Li et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) illustrated the feasibility of ML in the prediction of post-stroke cognitive impairment. However, ML has limitations and may lead to inaccurate predictions in certain clinical situations. Some researchers found that ML survival or classification models brought little improvement over traditional statistical methods, and the benefits of its assessment should be approached with caution, especially considering the limited sample size and features (R. Li et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Noroozi et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRecent work has shown that ML algorithms can classify CSVD and normal patients with high accuracy. Unfortunately, few evidence-based studies of ML models for CSVD-related cognitive impairment are currently available. As a result, this study is aim to analyze ML prediction models applied to neuroimaging data combined with other variables to predict cognitive impairment in CSVD and provide a useful reference for clinical practice and future research.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cp\u003eThis study was conducted according to the Preferred Reporting Items for a Systematic Review and Meta-analysis (PRISMA) 2020 guidelines (Page et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). It was registered on the PROSPERO website (CRD42024601473).\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Search strategy\u003c/h2\u003e \u003cp\u003eTwo researchers (QW and JZ) independently conducted a systematic search of the electronic databases MEDLINE (PubMed), Cochrane Library, Embase, and Web of Science, and the retrieval was as of 21 September 2024. The retrieval strategy was as follows: (\u0026ldquo;cerebral small vessel diseases\u0026rdquo; OR \u0026ldquo;subcortical ischemic vascular disease\u0026rdquo;) AND (\u0026ldquo;cognitive Dysfunction\u0026rdquo; OR \u0026ldquo;cognitive impairment\u0026rdquo;) AND (\u0026ldquo;artificial intelligence\u0026rdquo; OR \u0026ldquo;machine learning\u0026rdquo; OR \u0026ldquo;neural network\u0026rdquo; OR \u0026ldquo;deep learning\u0026rdquo;). The detailed strategies are available in the \u003cb\u003eSupplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e. In case of disagreement between assessors, consensus was reached through discussion and negotiation. We also identified additional relevant studies by reviewing the reference lists of the retrieved studies and review articles.\u003c/p\u003e \u003cp\u003eThis study followed the PICOTS system, recommended by the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) checklist (Palaz\u0026oacute;n-Bru et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The key items of our systematic review are described below:\u003c/p\u003e \u003cp\u003eP (Population): Patients with CSVD.\u003c/p\u003e \u003cp\u003eC (Comparator): ML methods as interventions.\u003c/p\u003e \u003cp\u003eO (Outcome): the performance for the prediction and diagnosis of ML models, including model discrimination or concordance index (C-index), specificity, sensitivity, and area under the curve (AUC).\u003c/p\u003e \u003cp\u003eT (Timing): The outcome was predicted after evaluating basic information at admission, clinical scoring scale results, and laboratory indicators.\u003c/p\u003e \u003cp\u003eS (Setting): The use of the ML prediction model is to individualize the prediction of cognitive impairment in patients with CSVD, facilitating the implementation of preventive measures to prevent adverse events.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Inclusion and exclusion criteria\u003c/h2\u003e \u003cp\u003eAll studies included had to meet the following criteria: (1) patients diagnosed with CSVD; (2) studies published in English; (3) ML was applied to predict CSVD-related cognitive impairment prediction, with a clear description of the ML models; (4) at least one measure of model performance (discrimination or calibration) was reported. Exclusion criteria were as follows: (1) only analysis of risk factors was conducted, without building complete ML models; (2) publication types such as review articles, case reports, editorials, conference abstracts and animal studies; (3) studies on the accuracy of single-factor prediction models; (4) studies that prediction models were developed, but not validated; (5) the full text could not be retrieved despite contacting the authors via email.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Literature screening\u003c/h2\u003e \u003cp\u003eEndNote 20 software (Clarivate Analytics, Philadelphia, PA, USA) were employed to manage the studies and remove duplicate items. Three investigators (QW JZ and PL) conducted literature screening and ensured that all studies met the inclusion criteria Each selected article has been screened at least twice. Any disagreement was dissolved by consulting a third reviewer (XZ).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Data extraction\u003c/h2\u003e \u003cp\u003eThe data extracted from each article was categorized into two groups: (1) Basic information: name of the first author, publication year, country, study design, sample size, age, diagnostic criteria for cognitive impairment, number of model variables and modeling variables. (2) Model information: variable selection method, handling of missing value, ML algorithms, model validation method, model performance measures. The extraction of information was performed by one reviewer, then checked by another reviewer to ensure accuracy and consistency.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Quality and Bias Assessments\u003c/h2\u003e \u003cp\u003eThe Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to assess the risk of bias and applicability of the included study. The PROBAST consists of 20 signaling questions in four distinct domains, namely participants, predictors, outcome, and analysis (Moons et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Each signaling question can be answered as \u0026ldquo;yes\u0026rdquo;, \u0026ldquo;probably yes\u0026rdquo;, \u0026ldquo;no\u0026rdquo;, \u0026ldquo;probably no\u0026rdquo; or \u0026ldquo;no information\u0026rdquo;. If at least one signaling question in a domain is answered as \u0026ldquo;no\u0026rdquo; or \u0026ldquo;probably no\u0026rdquo;, that domain should be considered at high risk of bias. Only when all domains are judged as low risk of bias, the overall bias can be considered low risk (Fu et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Two independent researchers (QW and JZ) evaluated the quality and the bias risk of the included studies. Any disagreements will be resolved by consensus by a third researcher (CH).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6. Outcome measures\u003c/h2\u003e \u003cp\u003eWe extracted the C-index as the primary outcome measure, which can be used to reflect the overall accuracy of ML models. However, this indicator alone may not fully reflect the predictive accuracy of ML models. Therefore, sensitivity and specificity were included as complementary outcome measures to evaluate the predictive accuracy of ML in CSVD-related cognitive impairment.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7. Data synthesis and statistical analysis\u003c/h2\u003e \u003cp\u003eStata version 15.1 (StataCorp LP, College Station, TX, USA) were used to conduct the meta-analyses. Given the differences in modeling variables and parameters, the C-index was preferred to be pooled using a random effects model using a random effects model while a bivariate mixed-effects model was used to calculate the pooled sensitivity and specificity. If the C-index did not report 95% CIs and standard error (SEs), we estimated the SEs using the methods by Debray (Debray et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). C-index is similar to the AUC (Nezic, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), indicates its diagnostic or prognostic discrimination ability as low (C-index\u0026thinsp;\u0026le;\u0026thinsp;0.5), modest (C-index\u0026thinsp;\u0026gt;\u0026thinsp;0.6 to 0.7), good (C-index\u0026thinsp;\u0026gt;\u0026thinsp;0.7 to 0.8), or strong (C-index\u0026thinsp;\u0026gt;\u0026thinsp;0.8) (Snell et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Cochrane Q-test and I\u003csup\u003e2\u003c/sup\u003e statistics were performed to examine heterogeneity. The I\u003csup\u003e2\u003c/sup\u003e statistics provides a measure of heterogeneity, with values of 25%, 50%, and 75% indicating low, moderate, and high heterogeneity respectively (Higgins et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2003\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSubgroup analyses, meta-regression and sensitivity analysis were also performed to gain insight into potential sources of heterogeneity. The C-index of the different ML algorithms for predicting cognitive impairment in patients with CSVD are discussed in the \u003cspan refid=\"Sec16\" class=\"InternalRef\"\u003esubgroup Analysis\u003c/span\u003e section. Sensitivity analysis was conducted by sequentially excluding each individual study and subsequently recalculating the pooled effect size for the remaining dataset. Publication bias was evaluated using Egger\u0026rsquo;s test, with p\u0026thinsp;\u0026gt;\u0026thinsp;0.05 indicating a low publication bias (Egger et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1997\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Study selection\u003c/h2\u003e \u003cp\u003eThe study search process is illustrated in the PRISMA flowchart (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). In total, 5603 records were obtained by performing electronic and manual searches. After removing duplicates and screening titles and abstracts, 89 articles remained. On the basis of full-text review, thirteen articles were included in the study.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Study characteristics\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the design and participant characteristics of the included studies. The publication years of the article ranged from 2016 to 2024; 5 out of 13 (38%) were published in 2024 (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Huang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhu et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Of the thirteen eligible studies, ten were conducted in China (H. F. Chen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Chen et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Huang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; M. Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Qin et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; L. Zhang et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhu et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), one in Italy (Ciulli et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), and two in the UK (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; R. Li et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In the studies we included, eight were prospective (including three multicenter study) (Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; M. Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; L. Zhang et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhu et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and five were retrospective (H. F. Chen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Chen et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Ciulli et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Huang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; R. Li et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Qin et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), all conducted in single centers. The source data for ML training were mostly obtained from clinical hospitals; some also included public database. Among the 13 studies, the most commonly used diagnostic criteria for cognitive impairment were Montreal Cognitive Assessment (MoCA) (n\u0026thinsp;=\u0026thinsp;6), followed by Mini-Mental State Examination (MMSE) (n\u0026thinsp;=\u0026thinsp;2) and Diagnostic and Statistical Manual of Mental Disorders (DSM) (n\u0026thinsp;=\u0026thinsp;2).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOverview of basic data of the included studies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor (year)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCountry\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eData source\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStudy design\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAge, years\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSample size, n\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDiagnostic criteria for cognitive impairment\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eIncluded variables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eModel evaluation metrics\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang, L (2022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRetrospective study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e71.06\u0026thinsp;\u0026plusmn;\u0026thinsp;9.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e159\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMoCA score\u0026thinsp;\u0026lt;\u0026thinsp;26 (an additional 1 point for education\u0026thinsp;\u0026lt;\u0026thinsp;12 years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ehypertension, homocysteine, total CSVD MRI burden, and years of schooling\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eC-index, AUC\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuang (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClinical hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e57.28\u0026thinsp;\u0026plusmn;\u0026thinsp;1.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e304\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003emild-CSVD: MoCA score 18\u0026ndash;26; sever-CSVD: MoCA score\u0026thinsp;\u0026lt;\u0026thinsp;17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eHippocampal textures\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC, Sensitivity Specificity, Accuracy, PPV, NPV\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang, Y (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRetrospective study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e65.91\u0026thinsp;\u0026plusmn;\u0026thinsp;7.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMoCA score\u0026thinsp;\u0026lt;\u0026thinsp;26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eBEN, BEN/\u003c/p\u003e \u003cp\u003eALFF, and BEN/ReHo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC, Sensitivity Specificity, Accuracy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLi, N (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRetrospective cross-sectional study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e377\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMoCA score\u0026thinsp;\u0026lt;\u0026thinsp;25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eAge, hypertension, total CSVD burden, ApoA1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC, Sensitivity Specificity, Accuracy, PPV, NPV\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhu (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRetrospective study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e66.7\u0026thinsp;\u0026plusmn;\u0026thinsp;6.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e227\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMoCA score\u0026thinsp;\u0026lt;\u0026thinsp;26 (an additional 1 point for education\u0026thinsp;\u0026lt;\u0026thinsp;12 years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eAge, homocysteine, hypertension, lacunar infarct score, total CSVD burden\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eC-index\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUK\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMulticentre (UKB, SCANS, RUN DMC)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective cohort study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e761\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eDemographic factors (n\u0026thinsp;=\u0026thinsp;3), Cognitive scores (n\u0026thinsp;=\u0026thinsp;3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eC-index\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e(continued)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor (year)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCountry\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eData source\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStudy design\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAge, years\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSample size, n\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDiagnostic criteria for cognitive impairment\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eIncluded variables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eModel evaluation metrics\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiu, M (2022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRetrospective cohort study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e50\u0026ndash;80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e197\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eneuropsychological tests scores fallen\u003c/p\u003e \u003cp\u003ebeyond \u0026plusmn;\u0026thinsp;1.5 standard deviations (SDs) from the mean;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003emultiscale features of the brain functional\u003c/p\u003e \u003cp\u003einteractions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC, Sensitivity Specificity, Accuracy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLi (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUK, Netherland, Singapore\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMulticentre (SCANS, RUN DMC, HARMONISATION)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective cohort study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDSM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eDemographic features (n\u0026thinsp;=\u0026thinsp;7), Imaging features (n\u0026thinsp;=\u0026thinsp;5 Cognitive features (n\u0026thinsp;=\u0026thinsp;3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC, Sensitivity Specificity, Accuracy, Precision\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWang (2021)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e64.71\u0026thinsp;\u0026plusmn;\u0026thinsp;6.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e113\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMoCA score, MMSE score, neuropsychological tests scores: 1.5 standard deviations below the normative mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eDTI features (n\u0026thinsp;=\u0026thinsp;8),\u003c/p\u003e \u003cp\u003eCBF features (n\u0026thinsp;=\u0026thinsp;5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC, Sensitivity Specificity, Accuracy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQin (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMulticentre (3 hospitals)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective cohort study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e64.8\u0026thinsp;\u0026plusmn;\u0026thinsp;7.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e136\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDSM-IV; CDR of \u0026ge;\u0026thinsp;0.5 on at least one domain and a global score\u0026thinsp;\u0026ge;\u0026thinsp;0.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eDTI features, rsfMRI features, clinical features Neuropsychological measures\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSensitivity, Precision Specificity, Accuracy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen, X (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e70.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eShanghai Manual of Cohort Studies on Cerebral Small Vessel Diseases\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eHBP, DM, HLP, Smoke, Drink\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC, Sensitivity, Specificity, Accuracy PPV, NPV\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen, H (2020)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeurology department of a hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective cross-sectional study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e64.8\u0026thinsp;\u0026plusmn;\u0026thinsp;1.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eBeijing version of\u003c/p\u003e \u003cp\u003ethe MoCA (MoCA-BJ), MMSE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eDTI features: AD, FA, MD, RD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSensitivity, Specificity, Accuracy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCiulli (2016)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eItaly\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClinical hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProspective, randomized, single-blind\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e75.3\u0026thinsp;\u0026plusmn;\u0026thinsp;6.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eTMT-B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMD, FA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSensitivity, Specificity, Accuracy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003eAbbreviations: NR: not reported, RF: random forests; AUC: area under the receiver operating characteristic curve; PPV: positive predictive value; NPV: negative predictive value; MMSE: Mini-Mental State Examination; MoCA: Montreal Cognitive Assessment; CDR, clinical dementia rating; DSM: Diagnostic and Statistical Manual of Mental Disorders; BEN: brain entropy, ALFF: amplitude of low frequency fluctuation, ReHo: regional homogeneity; SCANS: St George\u0026rsquo;s Cognition and Neuroimaging in Stroke; RUN DMC: Radboud University Nijmegen Diffusion Tensor and MRI Cohort; HARMONISATION: A memory clinic study in Singapore; DTI: diffusion tensor imaging; rsfMRI: resting-state fMRI CBF: cerebral blood flow; HBP: high blood pressure; DM: diabetes mellitus; HLP: hypercholesterolemia; TMT: Trail Making Test Part B\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e2\u003c/span\u003e provides the model information in the included studies. There were 21 models in the included studies, among which 17 models reported AUC or C-statistic values, ranging from 0.708 to 0.952, and 10 models reported sensitivity and specificity. In the 13 studies, the logistic regression (LR) (38.4%; n\u0026thinsp;=\u0026thinsp;5) was the most universally used algorithm, followed by the support-vector machine (SVM) (23%; n\u0026thinsp;=\u0026thinsp;3). Besides, seven studies also assessed the performance of other ML algorithms: random forest (RF), k-nearest neighbors (KNN), decision trees (DT), connectome-based prediction model (CPM), DS-GAN (diffusion scalar generative adversarial network), Reg-Logistic, generalised matrix learning vector quantisation (GMLVQ), generalised relevance learning vector quantisation (GRLVQ), and unsupervised machine learning model. For measuring deep learning performance, AUC and the Youden index were most commonly used.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMethodological characteristics of the included machine learning models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor (year)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMissing value processing\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eValidation set generation method [internal validation/external validation]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVariable screening/ feature selection method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eML algorithms\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eModel performance\u003c/p\u003e \u003cp\u003e(C-index/ Sensitivity, Specificity)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang, L (2022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBootstrapping\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eUnivariate significance level, multivariate logistic regression with stepwise regression (Forward: LR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.806 (0.735\u0026ndash;0.877)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuang (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedian interpolation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5-fold cross-validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSpearman correlation, LASSO regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eKNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.818 (0.673\u0026ndash;0.96) Sensitivity: 0.538 Specificity: 0.947\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang, Y (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eML method\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eleave-one-out cross validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM classifier\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.824(0.750\u0026ndash;0.899), Sensitivity: 0.865, Specificity: 0.618\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLi, N (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10-fold cross validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003emultivariate logistic regression analysis, LASSO regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.852 (0.781\u0026ndash;0.923) Sensitivity: 0.769, Specificity: 0.74\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhu (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedian interpolation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBootstrapping\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eUnivariate analysis, binary logistic regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.867(0.788\u0026ndash;0.947)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eInformation value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDS-GAN (consist of 2 deep learning models: a generator and a discriminator)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSCANS: C-index:0.828 (0.733, 0.923),0.903(0.828\u0026ndash;0.978)\u003c/p\u003e \u003cp\u003eRUN DMC: C-index:0.845(0.800\u0026ndash;0.890),0.858(0.815\u0026ndash;0.901)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiu, M (2022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5-fold cross-validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM classifier, GCN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCPM (with SVM as the classifier) and GCN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.821(0.747\u0026ndash;0.895), Sensitivity: 0.818, Specificity: 0.823\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e(continued)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor (year)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMissing value processing\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eValidation set generation method [internal validation/external validation]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVariable screening/ feature selection method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eML algorithms\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eModel performance\u003c/p\u003e \u003cp\u003e(C-index/ Sensitivity, Specificity)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLi (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eComplete case analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5-fold cross-validation\u003c/p\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSparse logistic regression classifier, LASSO regression, feature ranking\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLR, SVM, Reg-Logistic, GMLVQ, GRLVQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: LR 0.860 (0.826\u0026ndash;0.894),\u003c/p\u003e \u003cp\u003eReg-Logistic 0.870(0.837\u0026ndash;0.903)\u003c/p\u003e \u003cp\u003eSVM 0.858(0.824\u0026ndash;0.892)\u003c/p\u003e \u003cp\u003eGMLVQ 0.817 (0.779\u0026ndash;0.856)\u003c/p\u003e \u003cp\u003eGRLVQ 0.868 (0.835\u0026ndash;0.901)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWang (2021)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eleave-one-out cross validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSparse logistic regression classifier, LASSO regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.708 (0.667\u0026ndash;0.740), Sensitivity: 0.77, Specificity: 0.641\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQin (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRemove highly missing variables, fill missings with model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInternal validation\u003c/p\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLASSO regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eUnsupervised machine learning model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSensitivity: 0.795, Specificity: 0.962\u003c/p\u003e \u003cp\u003eSensitivity: 0.711, Specificity: 0.938\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen, X (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eComplete case analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,000 times\u003c/p\u003e \u003cp\u003ecross-validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStatistical analysis, k-means clustering\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.952 (0.911\u0026ndash;0.993) Sensitivity: 0.\u0026nbsp;981, Specificity:\u0026nbsp;0.896\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen, H (2020)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eimputation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBootstrapping\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSpearman correlation, RF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSensitivity: 0.818, Specificity:\u0026nbsp;0.79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCiulli (2016)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eleave-one-out cross validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAveraging Mean Diffusivity (MD) and Fractional Anisotropy (FA) maps within 50 Regions of Interest (ROIs)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSensitivity: 0.895, Specificity:\u0026nbsp;0.714\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eAbbreviations: NR: not reported, ML: machine learning; LASSO: Least Absolute Shrinkage and Selection Operator; RF: random forest; KNN: k-nearest neighbors; CPM: connectome-based prediction model; GCN: graph convolutional network; LR: logistic regression; SVM: support-vector machine; Reg-Logistic: regularised logistic regression with elastic net penalty; GRLVQ: generalised relevance learning vector quantisation; GMLVQ: generalised matrix learning vector quantisation; DT: decision tree; SLR: sparse logistic regression DS-GAN: diffusion scalar generative adversarial network\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eResults of subgroup analyses of C-index by study design, validation set generation method and machine learning type\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSubgroups\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eModels\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePooled C-index (95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP Value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eI\u003csup\u003e2\u003c/sup\u003e (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOverall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.85 (0.82, 0.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e81.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy design\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRetrospective\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.83 (0.80, 0.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u0026thinsp;=\u0026thinsp;0.792\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProspective\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.85 (0.82, 0.88)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e86.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValidation method\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInternal validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.83 (0.76, 0.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e91\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.86 (0.84, 0.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u0026thinsp;=\u0026thinsp;0.515\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.81 (0.74, 0.89)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e89.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon\u0026ndash;logistic regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.86 (0.84, 0.89)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e63.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFeature selection is an important step for ML training. Some researchers favored features that are statistically correlated and easily obtainable with CSVD (i.e., structural MRI, demographic, and cognitive results), whereas others incorporated factors based on established knowledge from previous models. In the included studies, 4 studies used DTI (diffusion tensor imaging) data as input features (H. F. Chen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Ciulli et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Qin et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), 3 studies used rs-fMRI (resting-state functional MRI) data as input features (M. Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Qin et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and one used a radiomics model with hippocampal texture features (Huang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Model validation\u003c/h2\u003e \u003cp\u003eIn the included studies, all ML models were internally or externally validated. Among them, internal validation was conducted in ten studies using random split, k-fold cross-validation or bootstrapping, while the remaining three conducted external validation. Models that only developed without validation were not included.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Quality assessment\u003c/h2\u003e \u003cp\u003eFive studies had a high risk of bias (38.4%), while only three studies had low risk (23%). In the participant domain, two studies were considered to have a moderate risk of bias primarily due to the use of inappropriate data sources and controversial criteria (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; R. Li et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), while bias in the other studies were low. In the predictor domain, one study had an unclear risk of bias due to lack of information on how predictive factors are defined or measured (Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The bias of outcome in one study was unclear due to no standard for diagnosing cognitive impairment described (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and three was high because multiple predictive factors form part of the outcome of cognitive impairment (H. F. Chen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In the analysis domain, the method of converting continuous variables into categorical variables in the two studies was not mentioned (H. F. Chen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Huang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). six studies where missing data has not been properly processed (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Ciulli et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; M. Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; L. Zhang et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), four studies did not consider model overfitting and underfitting (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Ciulli et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Qin et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Therefore, these attributes should be improved in subsequent model construction. More details of the risk of bias are shown in \u003cb\u003eSupplementary Table S2\u003c/b\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.5. Performance of ML Models for predicting cognitive impairment in CSVD\u003c/h2\u003e \u003cp\u003eMeta-analysis showed that the overall pooled C-index for ML models predicting cognitive impairment in CSVD was 0.85 (95% CI 0.82\u0026ndash;0.87; P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; I\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;81.6%; n\u0026thinsp;=\u0026thinsp;17); the pooled sensitivity was 0.82 (95% CI 0.77\u0026ndash;0.87; P\u0026thinsp;=\u0026thinsp;0.07; I\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;43.51%; n\u0026thinsp;=\u0026thinsp;10) and the pooled specificity was 0.81 (95% CI 0.73\u0026ndash;0.87; P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; I\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;72.15%; n\u0026thinsp;=\u0026thinsp;10). Forest plots of the above results are presented in Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe results of meta-regression showed that four items including country where the data were collected, publication year, study design, and ML methods used did not significant affect the C-index (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05; \u003cb\u003eSupplementary Table S3\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.6. Subgroup analysis\u003c/h2\u003e \u003cp\u003eWe performed the subgroup analysis of C-index based on study design, validation set generation method and model type. For study design, the pooled C-index of retrospective and prospective studies was 0.83 (95%CI 0.80\u0026ndash;0.87, n\u0026thinsp;=\u0026thinsp;5) and 0.85 (95%CI 0.82\u0026ndash;0.88, n\u0026thinsp;=\u0026thinsp;8), respectively. For validation set generation method, the pooled C-index of internal and external validation studies was 0.83 (95%CI 0.76\u0026ndash;0.91, n\u0026thinsp;=\u0026thinsp;8) and 0.86 (95%CI 0.84\u0026ndash;0.87, n\u0026thinsp;=\u0026thinsp;9), respectively. For ML algorithms, 5 prediction models were established applying the LR algorithms (Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; R. Li et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; L. Zhang et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Zhu et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and the overall pooled C-index for the LR models was 0.81 (95% CI 0.74\u0026ndash;0.89). The overall pooled C-index for non-LR models was 0.86 (95% CI 0.84\u0026ndash;0.89), the highest value among these subgroups. Forest plots of the above results are presented in supplemental file \u003cb\u003eFigure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e-3\u003c/b\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.7. Sensitivity analysis and publication bias\u003c/h2\u003e \u003cp\u003eThe sensitivity analysis excluding studies one by one showed that the meta-analysis results of predictive value of ML for cognitive impairment in CSVD were stable and reliable (\u003cb\u003eSupplementary Figure S4\u003c/b\u003e). Egger\u0026rsquo;s test result (P\u0026thinsp;=\u0026thinsp;0.439) and a symmetrically distributed funnel plot suggested no publication bias (\u003cb\u003eSupplementary Figure S5)\u003c/b\u003e.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eAlthough the increasing use of ML in predicting cognitive impairment in CSVD, the quality of studies varies, and the results also differ. It is necessary to perform a meta-analysis for a unified evaluation. This study is the first to systematically explore the application of the latest ML prediction models for cognitive impairment in CSVD.\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Principal findings\u003c/h2\u003e \u003cp\u003eIn our study, the overall pooled estimation of 13 studies showed that ML models achieved high accuracy in early recognition of CSVD patients, with a C-index of 0.85 (95% CI 0.82\u0026ndash;0.87) and sensitivity and specificity of \u0026gt;\u0026thinsp;80%. These results underscore the ability of ML algorithms to integrate complex data from various sources, such as neuroimaging, clinical assessments, and genetic markers, offering potential advantages. Although ML methods may not provide greater benefit than current available screening strategies, they do possess an advantage in allowing for a preferred trade-off between sensitivity and specificity (Z. Zhang et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). It was reported that common cognitive screening tests have similar predictive accuracy in CSVD-related cognitive impairment (Pasi et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Research has shown that the MoCA has superior sensitivity of cognitive function than other screening tools, but its specificity is less than desirable (Jia et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Szcześniak et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). However, our findings showed that ML has considerably high and balanced predictive accuracy (c-index, sensitivity, and specificity) in CSVD-related cognitive impairment and is robust for future clinical implementation.\u003c/p\u003e \u003cp\u003eBased on the subgroup analysis, models developed utilizing non-LR methods achieved the highest C-index, implying that researchers should consider testing a broader range of candidate models to potentially identify superior predictive performance. Li et al (R. Li et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) developed five common ML methods (i.e., LR, SVM, Reg-Logistic, GMLVQ, GRLVQ) to predict the occurrence of CSVD-related cognitive impairment, with a data set of 889 patients. Reg-Logistic and GRLVQ were the two models with the best diagnostic accuracy in that study (C-index 0.868\u0026ndash;0.87). Regarding the ML methods used to predict the risk of cognitive impairment in the context of CSVD, LR is the most commonly used type of model, since LR models are suitable for simple data with linear relationships between variables and outcomes (Boehm-Sturm et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; B. Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Montine et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Ensemble methods like RF have achieved high precision in predicting cognitive trajectory types using longitudinal by aggregating the results of multiple decision trees, thus reducing overfitting and increasing robustness (Mohammadiarvejeh et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Moreover, SVM and KNN have more advantages in image recognition and segmentation for predicting mild cognitive impairment (Al-Qazzaz et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Deep learning exhibits significant potential in the realm of image classification, attributed to its capability to automatically extract low to high-level features. Studies have demonstrated the utility of deep learning models in analyzing MRI sequence to identify subtle changes subcortical vascular cognitive impairment (Q. Chen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe most commonly used features in ML models are neuroimaging biomarkers such as white matter hyperintensities (WMHs), cerebral microbleeds (CMBs) lacunar infarction (LI), enlarged perivascular spaces (EPVS) and total CSVD burden. The preliminary research on neuroimaging biomarkers provides the imaging basis of CSVD for machine learning and can serve as primary features for input. In addition, DTI and rs-fMRI are widely applied in the study of CSVD, primarily focusing on the early identification and progression of cognitive impairment (Egle et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Based on DTI and rs-fMRI parameters, structural and functional brain networks can be constructed separately. The fractional anisotropy (FA) score of DTI in CSVD patients decreases, while the mean diffusivity (MD) increases (Biesbroek et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), and these changes are associated with clinical outcomes such as cognitive impairment (Tuladhar et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), gait abnormalities (van der Holst et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), and emotional disorders (van Uden et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Research based on rs-fMRI has found that CSVD patients exhibit abnormal functional activity in the prefrontal cortex, subcortical region, cingulate gyrus, and hippocampus (Li et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Yi et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Ciulli et al. utilized the mean MD value as a feature and employed a machine learning strategy combining a linear support vector machine with leave-one-out cross validation to predict the executive function of CSVD patients with mild cognitive impairment, in which sensitivity, specificity, and accuracy all reached a high level (Ciulli et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Furthermore, Pantoni et al. used traditional imaging biomarkers and the fractal dimension (FD) of gray and white matter as input features to predict different neuropsychological test scores. They found that the FD value of white matter was the most commonly used feature, and FD may be an important marker for predicting cognitive decline in CSVD, which can supplement traditional imaging biomarkers (Pantoni et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Our study also finds that radiomics features are frequently used as predictors in ML models for CSVD-related cognitive impairment. A recent study validated the superior accuracy and specificity of radiomics models utilizing hippocampal texture features in identifying CSVD patients with severe cognitive impairment (Huang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAs for validation methods, 8 studies selected cross-validation, with different folding or iteration times. Cross-validation splits the samples into training set and testing set. This division can average the accuracy of each iteration to obtain a more robust model performance, rather than validating the model on just one test sample. 13 studies chose leave-one-out cross validation, in which case the model is trained using all data except for one data point, and then it attempts to classify the missing data points and applies the same operation to the rest of the sample in subsequent iterations. 2 studies used bootstrapping as a validation method. The model in most studies primarily relied on internal validation rather than external validation. The lack of external validation of the obtained results is considered a major limitation of current prediction models (van Kempen et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The reason for this phenomenon may be due to the insufficient sample sizes for external validation. Subgroup analysis according to different validation methods showed that the pooled C-index of the external validation group were higher than those of the internal validation. This may be because the dataset used for external validation is large, allowing ML algorithms to be fully learned and trained.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Limitations and strengths\u003c/h2\u003e \u003cp\u003eThis study has several potential limitations. First, non-English studies were not included in the meta-analysis, which might lead to selection bias. Second, a significant proportion of the included studies were conducted in mainland China, which may limit the generalizability of the findings to Western populations. Therefore, it is crucial for future research to develop models for cognitive impairment in CSVD patients across diverse populations to ensure their relevance and applicability in varied clinical settings. Thirdly, the heterogeneity in study designs, sample sizes, outcome measures, and a diverse array of feature selection methods across the included studies introduced potential biases and may affect the performance and applicability of ML models. Lastly, the number of validated studies and models included is relatively low, with only 3 models have external validation. Internal validation of data can result in overfitting or spectrum bias, thereby potentially overestimating the performance of machine learning models (Hu et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Kuo et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite the aforementioned limitations, this study can still offer valuable guidance for future research and clinical practice. In the first place, the major strength of our study lies in the rigorous literature search methodology and sufficiently detailed meta-analysis to be reproducible. We performed a quantitative synthesis, unprecedented in prior research, to systematically compare the models in question and described the characteristics of the validated ML models. In addition, the novel PROBAST was used to assess the risk of bias of the included predictive models. We conducted a thorough examination of more details of the model, such as data sources, processing, feature selection, and model validation (Di Tanna et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Finally, we found that 5 studies had a high risk of bias in methodological quality, which may lead to overfitted prediction models. Potential biases may be prevented if studies adhere to the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) initiative in their reporting (Collins et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e4.3. Future perspectives\u003c/h2\u003e \u003cp\u003eIn order to make ML algorithms more robust in clinical implementation, there are some important issues that should be addressed in future research. Firstly, the multimodal imaging research of CSVD has gradually shifted from structural imaging biomarkers, to parameter analysis based on DTI and rs-fMRI, and then to constructing brain networks, leading people's understanding of CSVD from local to global. In the future, multimodal data including DTI and fMRI can be conducted to further explore the occurrence and development patterns of CSVD. Secondly, studies that achieved higher accuracy have in common the use of multidimensional and multimodal data, as well as increasingly complex classification methods. However, a balance is needed between the algorithms that achieve the higher performance, and the data and methods that more easily obtainable data in the clinical practice. In addition, although these algorithms are useful in distinguishing the brain features of CSVD, their performance is far from specific enough to hand over a complete diagnosis to automated methods. Nevertheless, when implemented in clinical practice, computer-aided diagnosis will provide a faster, easier to perform, and earlier detection method to predict the potential progression of CSVD patients to cognitive impairment. Consequently, it is recommended that clinicians enhance their capability to apply ML techniques to improve the accuracy of their diagnoses (Nagendran et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Wu et al., \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Finally, the effectiveness and performance of machine learning are closely related to the size of the dataset. Therefore, Future research should prioritize the development of novel models that leverage larger sample sizes and employ rigorous study designs. Emphasis should be placed on multicenter external validation to ensure the generalizability and robustness of findings across diverse populations.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study shows the potential of ML algorithms predictive performance for cognitive impairment in patients with CSVD and can be used as a potential tool for early identification of cognitive impairment in this population. The development and validation of ML models tailored to CSVD-related cognitive impairment could facilitate earlier diagnosis and personalized treatment strategies, potentially improving patient outcomes.\u003c/p\u003e \u003cp\u003eThis study also analyzed the classification algorithms and feature source used in the selected studies, making diverse methodologies available to researchers. Our research findings also emphasize that the main limitation of the current prediction model is the lack of external validation. Additionally, this meta-analysis excluded studies that only developed models without validation and did not report model performance metrics. Overall, while the study provides important insights into the potential of machine learning for diagnosing cognitive impairment in patients with CSVD, more research is needed to fully explore and validate these approaches.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXZ and HC contributed to the study concept and design. QW and JZ retrieved and filtered the articles. QW and JZ extracted the data. QW and JZ analyzed the data. JZ, QW, and PL interpreted the data. JZ and QW drafted the manuscript. XZ and HC contributed to the critical revision of the manuscript. All authors contributed to the paper and approved the submitted version.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was financially supported by the National Natural Science Foundation of China (Grant No. 82460226), the Natural Science Foundation of Guangxi Autonomous Region (Grant No. 2023GXNSFAA026383).\u003c/p\u003e\n\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\u003eData availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAl-Qazzaz, N. K., Ali, S., Ahmad, S. A., Islam, M. S., \u0026amp; Escudero, J. (2018). Discrimination of stroke-related mild cognitive impairment and vascular dementia using EEG signal analysis. \u003cem\u003eMed Biol Eng Comput\u003c/em\u003e,\u003cem\u003e 56\u003c/em\u003e(1), 137-157. https://doi.org/10.1007/s11517-017-1734-7\u003c/li\u003e\n\u003cli\u003eBiesbroek, J. M., Leemans, A., den Bakker, H., Duering, M., Gesierich, B., Koek, H. L.,\u0026hellip;Biessels, G. J. (2017). Microstructure of Strategic White Matter Tracts and Cognition in Memory Clinic Patients with Vascular Brain Injury. \u003cem\u003eDement Geriatr Cogn Disord\u003c/em\u003e,\u003cem\u003e 44\u003c/em\u003e(5-6), 268-282. https://doi.org/10.1159/000485376\u003c/li\u003e\n\u003cli\u003eBoehm-Sturm, P., F\u0026uuml;chtemeier, M., Foddis, M., Mueller, S., Trueman, R. C., Zille, M.,\u0026hellip;Farr, T. D. (2017). Neuroimaging Biomarkers Predict Brain Structural Connectivity Change in a Mouse Model of Vascular Cognitive Impairment. \u003cem\u003eStroke\u003c/em\u003e,\u003cem\u003e 48\u003c/em\u003e(2), 468-475. https://doi.org/10.1161/strokeaha.116.014394\u003c/li\u003e\n\u003cli\u003eBos, D., Wolters, F. J., Darweesh, S. K. L., Vernooij, M. W., de Wolf, F., Ikram, M. A., \u0026amp; Hofman, A. (2018). Cerebral small vessel disease and the risk of dementia: A systematic review and meta-analysis of population-based evidence. \u003cem\u003eAlzheimers Dement\u003c/em\u003e,\u003cem\u003e 14\u003c/em\u003e(11), 1482-1492. https://doi.org/10.1016/j.jalz.2018.04.007\u003c/li\u003e\n\u003cli\u003eChen, H. F., Huang, L. L., Li, H. Y., Qian, Y., Yang, D., Qing, Z.,\u0026hellip;Xu, Y. (2020). Microstructural disruption of the right inferior fronto-occipital and inferior longitudinal fasciculus contributes to WMH-related cognitive impairment. \u003cem\u003eCNS Neurosci Ther\u003c/em\u003e,\u003cem\u003e 26\u003c/em\u003e(5), 576-588. https://doi.org/10.1111/cns.13283\u003c/li\u003e\n\u003cli\u003eChen, Q., Wang, Y., Qiu, Y., Wu, X., Zhou, Y., \u0026amp; Zhai, G. (2020). A Deep Learning-Based Model for Classification of Different Subtypes of Subcortical Vascular Cognitive Impairment With FLAIR. \u003cem\u003eFront Neurosci\u003c/em\u003e,\u003cem\u003e 14\u003c/em\u003e, 557. https://doi.org/10.3389/fnins.2020.00557\u003c/li\u003e\n\u003cli\u003eChen, X., Fang, Z., Zhao, Y., Cheng, W., Chen, H., Li, G.,\u0026hellip;Yin, Y. (2023). Sleep Parameters and Plasma Biomarkers for Cognitive Impairment Evaluation in Patients With Cerebral Small Vessel Disease. \u003cem\u003eJ Gerontol B Psychol Sci Soc Sci\u003c/em\u003e,\u003cem\u003e 78\u003c/em\u003e(2), 210-219. https://doi.org/10.1093/geronb/gbac137\u003c/li\u003e\n\u003cli\u003eChen, Y., Tozer, D., Li, R., Li, H., Tuladhar, A., De Leeuw, F. E., \u0026amp; Markus, H. S. (2024). Improved Dementia Prediction in Cerebral Small Vessel Disease Using Deep Learning-Derived Diffusion Scalar Maps From T1. \u003cem\u003eStroke\u003c/em\u003e,\u003cem\u003e 55\u003c/em\u003e(9), 2254-2263. https://doi.org/10.1161/strokeaha.124.047449\u003c/li\u003e\n\u003cli\u003eCiulli, S., Citi, L., Salvadori, E., Valenti, R., Poggesi, A., Inzitari, D.,\u0026hellip;Diciotti, S. (2016). Prediction of Impaired Performance in Trail Making Test in MCI Patients With Small Vessel Disease Using DTI Data. \u003cem\u003eIEEE J Biomed Health Inform\u003c/em\u003e,\u003cem\u003e 20\u003c/em\u003e(4), 1026-1033. https://doi.org/10.1109/jbhi.2016.2537808\u003c/li\u003e\n\u003cli\u003eCollins, G. S., Reitsma, J. B., Altman, D. G., \u0026amp; Moons, K. G. (2015). Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. \u003cem\u003eBmj\u003c/em\u003e,\u003cem\u003e 350\u003c/em\u003e, g7594. https://doi.org/10.1136/bmj.g7594\u003c/li\u003e\n\u003cli\u003eDas, A. S., Regenhardt, R. W., Vernooij, M. W., Blacker, D., Charidimou, A., \u0026amp; Viswanathan, A. (2019). Asymptomatic Cerebral Small Vessel Disease: Insights from Population-Based Studies. \u003cem\u003eJ Stroke\u003c/em\u003e,\u003cem\u003e 21\u003c/em\u003e(2), 121-138. https://doi.org/10.5853/jos.2018.03608\u003c/li\u003e\n\u003cli\u003eDebray, T. P., Damen, J. A., Riley, R. D., Snell, K., Reitsma, J. B., Hooft, L.,\u0026hellip;Moons, K. G. (2019). A framework for meta-analysis of prediction model studies with binary and time-to-event outcomes. \u003cem\u003eStat Methods Med Res\u003c/em\u003e,\u003cem\u003e 28\u003c/em\u003e(9), 2768-2786. https://doi.org/10.1177/0962280218785504\u003c/li\u003e\n\u003cli\u003eDi Tanna, G. L., Wirtz, H., Burrows, K. L., \u0026amp; Globe, G. (2020). Evaluating risk prediction models for adults with heart failure: A systematic literature review. \u003cem\u003ePLoS One\u003c/em\u003e,\u003cem\u003e 15\u003c/em\u003e(1), e0224135. https://doi.org/10.1371/journal.pone.0224135\u003c/li\u003e\n\u003cli\u003eEgger, M., Davey Smith, G., Schneider, M., \u0026amp; Minder, C. (1997). Bias in meta-analysis detected by a simple, graphical test. \u003cem\u003eBmj\u003c/em\u003e,\u003cem\u003e 315\u003c/em\u003e(7109), 629-634. https://doi.org/10.1136/bmj.315.7109.629\u003c/li\u003e\n\u003cli\u003eEgle, M., Hilal, S., Tuladhar, A. M., Pirpamer, L., Bell, S., Hofer, E.,\u0026hellip;Markus, H. S. (2022). Determining the OPTIMAL DTI analysis method for application in cerebral small vessel disease. \u003cem\u003eNeuroimage Clin\u003c/em\u003e,\u003cem\u003e 35\u003c/em\u003e, 103114. https://doi.org/10.1016/j.nicl.2022.103114\u003c/li\u003e\n\u003cli\u003eFu, H., Hou, D., Xu, R., You, Q., Li, H., Yang, Q.,\u0026hellip;Bai, D. (2024). Risk prediction models for deep venous thrombosis in patients with acute stroke: A systematic review and meta-analysis. \u003cem\u003eInt J Nurs Stud\u003c/em\u003e,\u003cem\u003e 149\u003c/em\u003e, 104623. https://doi.org/10.1016/j.ijnurstu.2023.104623\u003c/li\u003e\n\u003cli\u003eGrueso, S., \u0026amp; Viejo-Sobera, R. (2021). Machine learning methods for predicting progression from mild cognitive impairment to Alzheimer\u0026apos;s disease dementia: a systematic review. \u003cem\u003eAlzheimers Res Ther\u003c/em\u003e,\u003cem\u003e 13\u003c/em\u003e(1), 162. https://doi.org/10.1186/s13195-021-00900-w\u003c/li\u003e\n\u003cli\u003eHiggins, J. P., Thompson, S. G., Deeks, J. J., \u0026amp; Altman, D. G. (2003). Measuring inconsistency in meta-analyses. \u003cem\u003eBmj\u003c/em\u003e,\u003cem\u003e 327\u003c/em\u003e(7414), 557-560. https://doi.org/10.1136/bmj.327.7414.557\u003c/li\u003e\n\u003cli\u003eHu, J., Wang, Y., Guo, D., Qu, Z., Sui, C., He, G.,\u0026hellip;Liu, X. (2023). Diagnostic performance of magnetic resonance imaging-based machine learning in Alzheimer\u0026apos;s disease detection: a meta-analysis. \u003cem\u003eNeuroradiology\u003c/em\u003e,\u003cem\u003e 65\u003c/em\u003e(3), 513-527. https://doi.org/10.1007/s00234-022-03098-2\u003c/li\u003e\n\u003cli\u003eHuang, B., Zheng, W., Mu, R., Yang, P., Li, X., Liu, F.,\u0026hellip;Zhu, X. (2024). Disparities in the diagnostic efficacy of radiomics models in predicting various degrees of cognitive impairment in patients with cerebral small vessel disease. \u003cem\u003eBMC Med Imaging\u003c/em\u003e,\u003cem\u003e 24\u003c/em\u003e(1), 257. https://doi.org/10.1186/s12880-024-01431-0\u003c/li\u003e\n\u003cli\u003eJia, X., Wang, Z., Huang, F., Su, C., Du, W., Jiang, H.,\u0026hellip;Zhang, B. (2021). A comparison of the Mini-Mental State Examination (MMSE) with the Montreal Cognitive Assessment (MoCA) for mild cognitive impairment screening in Chinese middle-aged and older population: a cross-sectional study. \u003cem\u003eBMC Psychiatry\u003c/em\u003e,\u003cem\u003e 21\u003c/em\u003e(1), 485. https://doi.org/10.1186/s12888-021-03495-6\u003c/li\u003e\n\u003cli\u003eKuo, R. Y. L., Harrison, C., Curran, T. A., Jones, B., Freethy, A., Cussons, D.,\u0026hellip;Furniss, D. (2022). Artificial Intelligence in Fracture Detection: A Systematic Review and Meta-Analysis. \u003cem\u003eRadiology\u003c/em\u003e,\u003cem\u003e 304\u003c/em\u003e(1), 50-62. https://doi.org/10.1148/radiol.211785\u003c/li\u003e\n\u003cli\u003eLi, C., Zheng, J., \u0026amp; Wang, J. (2012). An fMRI study of prefrontal cortical function in subcortical ischemic vascular cognitive impairment. \u003cem\u003eAm J Alzheimers Dis Other Demen\u003c/em\u003e,\u003cem\u003e 27\u003c/em\u003e(7), 490-495. https://doi.org/10.1177/1533317512455841\u003c/li\u003e\n\u003cli\u003eLi, N., Gao, Y., Li, L. T., Hu, Y. D., Ling, L., Jia, N.,\u0026hellip;Jiang, Y. (2024). Development and validation of a nomogram predictive model for cognitive impairment in cerebral small vessel disease: a comprehensive retrospective analysis. \u003cem\u003eFront Neurol\u003c/em\u003e,\u003cem\u003e 15\u003c/em\u003e, 1373306. https://doi.org/10.3389/fneur.2024.1373306\u003c/li\u003e\n\u003cli\u003eLi, R., Harshfield, E. L., Bell, S., Burkhart, M., Tuladhar, A. M., Hilal, S.,\u0026hellip;Markus, H. S. (2023). Predicting incident dementia in cerebral small vessel disease: comparison of machine learning and traditional statistical models. \u003cem\u003eCereb Circ Cogn Behav\u003c/em\u003e,\u003cem\u003e 5\u003c/em\u003e, 100179. https://doi.org/10.1016/j.cccb.2023.100179\u003c/li\u003e\n\u003cli\u003eLi, X., Chen, Z., Jiao, H., Wang, B., Yin, H., Chen, L.,\u0026hellip;Qin, D. (2023). Machine learning in the prediction of post-stroke cognitive impairment: a systematic review and meta-analysis. \u003cem\u003eFront Neurol\u003c/em\u003e,\u003cem\u003e 14\u003c/em\u003e, 1211733. https://doi.org/10.3389/fneur.2023.1211733\u003c/li\u003e\n\u003cli\u003eLi, Y., Kalpouzos, G., Laukka, E. J., Dekhtyar, S., B\u0026auml;ckman, L., Fratiglioni, L., \u0026amp; Qiu, C. (2022). Progression of neuroimaging markers of cerebral small vessel disease in older adults: A 6-year follow-up study. \u003cem\u003eNeurobiol Aging\u003c/em\u003e,\u003cem\u003e 112\u003c/em\u003e, 204-211. https://doi.org/10.1016/j.neurobiolaging.2022.01.006\u003c/li\u003e\n\u003cli\u003eLiu, B., Meng, S., Cheng, J., Zeng, Y., Zhou, D., Deng, X.,\u0026hellip;Li, C. (2022). Diagnosis of Subcortical Ischemic Vascular Cognitive Impairment With No Dementia Using Radiomics of Cerebral Cortex and Subcortical Nuclei in High-Resolution T1-Weighted MR Imaging. \u003cem\u003eFront Oncol\u003c/em\u003e,\u003cem\u003e 12\u003c/em\u003e, 852726. https://doi.org/10.3389/fonc.2022.852726\u003c/li\u003e\n\u003cli\u003eLiu, M., Wang, Y., Zhang, H., Yang, Q., Shi, F., Zhou, Y., \u0026amp; Shen, D. (2022). Multiscale functional connectome abnormality predicts cognitive outcomes in subcortical ischemic vascular disease. \u003cem\u003eCereb Cortex\u003c/em\u003e,\u003cem\u003e 32\u003c/em\u003e(21), 4641-4656. https://doi.org/10.1093/cercor/bhab507\u003c/li\u003e\n\u003cli\u003eMohammadiarvejeh, P., Klinedinst, B. S., Wang, Q., Li, T., Larsen, B., Pollpeter, A.,\u0026hellip;Willette, A. A. (2023). Bioenergetic and vascular predictors of potential super-ager and cognitive decline trajectories-a UK Biobank Random Forest classification study. \u003cem\u003eGeroscience\u003c/em\u003e,\u003cem\u003e 45\u003c/em\u003e(1), 491-505. https://doi.org/10.1007/s11357-022-00657-6\u003c/li\u003e\n\u003cli\u003eMontine, T. J., Bukhari, S. A., \u0026amp; White, L. R. (2021). Cognitive Impairment in Older Adults and Therapeutic Strategies. \u003cem\u003ePharmacol Rev\u003c/em\u003e,\u003cem\u003e 73\u003c/em\u003e(1), 152-162. https://doi.org/10.1124/pharmrev.120.000031\u003c/li\u003e\n\u003cli\u003eMoons, K. G. M., Wolff, R. F., Riley, R. D., Whiting, P. F., Westwood, M., Collins, G. S.,\u0026hellip;Mallett, S. (2019). PROBAST: A Tool to Assess Risk of Bias and Applicability of Prediction Model Studies: Explanation and Elaboration. \u003cem\u003eAnn Intern Med\u003c/em\u003e,\u003cem\u003e 170\u003c/em\u003e(1), W1-w33. https://doi.org/10.7326/m18-1377\u003c/li\u003e\n\u003cli\u003eNagendran, M., Chen, Y., Lovejoy, C. A., Gordon, A. C., Komorowski, M., Harvey, H.,\u0026hellip;Maruthappu, M. (2020). Artificial intelligence versus clinicians: systematic review of design, reporting standards, and claims of deep learning studies. \u003cem\u003eBmj\u003c/em\u003e,\u003cem\u003e 368\u003c/em\u003e, m689. https://doi.org/10.1136/bmj.m689\u003c/li\u003e\n\u003cli\u003eNezic, D. G. (2020). Assessing the performance of risk prediction models. \u003cem\u003eEur J Cardiothorac Surg\u003c/em\u003e,\u003cem\u003e 58\u003c/em\u003e(2), 401. https://doi.org/10.1093/ejcts/ezaa071\u003c/li\u003e\n\u003cli\u003eNoroozi, M., Gholami, M., Sadeghsalehi, H., Behzadi, S., Habibzadeh, A., Erabi, G.,\u0026hellip;Deravi, N. (2024). Machine and deep learning algorithms for classifying different types of dementia: A literature review. \u003cem\u003eAppl Neuropsychol Adult\u003c/em\u003e, 1-15. https://doi.org/10.1080/23279095.2024.2382823\u003c/li\u003e\n\u003cli\u003eOdusami, M., Maskeliūnas, R., Dama\u0026scaron;evičius, R., \u0026amp; Misra, S. (2024). Machine learning with multimodal neuroimaging data to classify stages of Alzheimer\u0026apos;s disease: a systematic review and meta-analysis. \u003cem\u003eCogn Neurodyn\u003c/em\u003e,\u003cem\u003e 18\u003c/em\u003e(3), 775-794. https://doi.org/10.1007/s11571-023-09993-5\u003c/li\u003e\n\u003cli\u003ePage, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D.,\u0026hellip;Moher, D. (2021). The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. \u003cem\u003eBmj\u003c/em\u003e,\u003cem\u003e 372\u003c/em\u003e, n71. https://doi.org/10.1136/bmj.n71\u003c/li\u003e\n\u003cli\u003ePalaz\u0026oacute;n-Bru, A., Mart\u0026iacute;n-P\u0026eacute;rez, F., Mares-Garc\u0026iacute;a, E., Beneyto-Ripoll, C., Gil-Guill\u0026eacute;n, V. F., P\u0026eacute;rez-Sempere, \u0026Aacute;., \u0026amp; Carbonell-Torregrosa, M. (2020). A general presentation on how to carry out a CHARMS analysis for prognostic multivariate models. \u003cem\u003eStat Med\u003c/em\u003e,\u003cem\u003e 39\u003c/em\u003e(23), 3207-3225. https://doi.org/10.1002/sim.8660\u003c/li\u003e\n\u003cli\u003ePantoni, L. (2010). Cerebral small vessel disease: from pathogenesis and clinical characteristics to therapeutic challenges. \u003cem\u003eLancet Neurol\u003c/em\u003e,\u003cem\u003e 9\u003c/em\u003e(7), 689-701. https://doi.org/10.1016/s1474-4422(10)70104-6\u003c/li\u003e\n\u003cli\u003ePantoni, L., Marzi, C., Poggesi, A., Giorgio, A., De Stefano, N., Mascalchi, M.,\u0026hellip;Diciotti, S. (2019). Fractal dimension of cerebral white matter: A consistent feature for prediction of the cognitive performance in patients with small vessel disease and mild cognitive impairment. \u003cem\u003eNeuroimage Clin\u003c/em\u003e,\u003cem\u003e 24\u003c/em\u003e, 101990. https://doi.org/10.1016/j.nicl.2019.101990\u003c/li\u003e\n\u003cli\u003ePasi, M., Sugita, L., Xiong, L., Charidimou, A., Boulouis, G., Pongpitakmetha, T.,\u0026hellip;Biffi, A. (2021). Association of Cerebral Small Vessel Disease and Cognitive Decline After Intracerebral Hemorrhage. \u003cem\u003eNeurology\u003c/em\u003e,\u003cem\u003e 96\u003c/em\u003e(2), e182-e192. https://doi.org/10.1212/wnl.0000000000011050\u003c/li\u003e\n\u003cli\u003eQin, Q., Qu, J., Yin, Y., Liang, Y., Wang, Y., Xie, B.,\u0026hellip;Tang, Y. (2023). Unsupervised machine learning model to predict cognitive impairment in subcortical ischemic vascular disease. \u003cem\u003eAlzheimers Dement\u003c/em\u003e,\u003cem\u003e 19\u003c/em\u003e(8), 3327-3338. https://doi.org/10.1002/alz.12971\u003c/li\u003e\n\u003cli\u003eRosenberg, G. A., Wallin, A., Wardlaw, J. M., Markus, H. S., Montaner, J., Wolfson, L.,\u0026hellip;Hachinski, V. (2016). Consensus statement for diagnosis of subcortical small vessel disease. \u003cem\u003eJ Cereb Blood Flow Metab\u003c/em\u003e,\u003cem\u003e 36\u003c/em\u003e(1), 6-25. https://doi.org/10.1038/jcbfm.2015.172\u003c/li\u003e\n\u003cli\u003eSnell, K. I., Ensor, J., Debray, T. P., Moons, K. G., \u0026amp; Riley, R. D. (2018). Meta-analysis of prediction model performance across multiple studies: Which scale helps ensure between-study normality for the C-statistic and calibration measures? \u003cem\u003eStat Methods Med Res\u003c/em\u003e,\u003cem\u003e 27\u003c/em\u003e(11), 3505-3522. https://doi.org/10.1177/0962280217705678\u003c/li\u003e\n\u003cli\u003eSzcześniak, D., Rymaszewska, J., Zimny, A., Sąsiadek, M., Połtyn-Zaradna, K., Smith, E. E.,\u0026hellip;Szuba, A. (2021). \u0026quot;Cerebral small vessel disease and other influential factors of cognitive impairment in the middle-aged: a long-term observational cohort PURE-MIND study in Poland\u0026quot;. \u003cem\u003eGeroscience\u003c/em\u003e,\u003cem\u003e 43\u003c/em\u003e(1), 279-295. https://doi.org/10.1007/s11357-020-00271-4\u003c/li\u003e\n\u003cli\u003eTuladhar, A. M., van Norden, A. G., de Laat, K. F., Zwiers, M. P., van Dijk, E. J., Norris, D. G., \u0026amp; de Leeuw, F. E. (2015). White matter integrity in small vessel disease is related to cognition. \u003cem\u003eNeuroimage Clin\u003c/em\u003e,\u003cem\u003e 7\u003c/em\u003e, 518-524. https://doi.org/10.1016/j.nicl.2015.02.003\u003c/li\u003e\n\u003cli\u003evan der Holst, H. M., Tuladhar, A. M., Zerbi, V., van Uden, I. W. M., de Laat, K. F., van Leijsen, E. M. C.,\u0026hellip;de Leeuw, F. E. (2018). White matter changes and gait decline in cerebral small vessel disease. \u003cem\u003eNeuroimage Clin\u003c/em\u003e,\u003cem\u003e 17\u003c/em\u003e, 731-738. https://doi.org/10.1016/j.nicl.2017.12.007\u003c/li\u003e\n\u003cli\u003evan Kempen, E. J., Post, M., Mannil, M., Kusters, B., Ter Laan, M., Meijer, F. J. A., \u0026amp; Henssen, D. (2021). Accuracy of Machine Learning Algorithms for the Classification of Molecular Features of Gliomas on MRI: A Systematic Literature Review and Meta-Analysis. \u003cem\u003eCancers (Basel)\u003c/em\u003e,\u003cem\u003e 13\u003c/em\u003e(11). https://doi.org/10.3390/cancers13112606\u003c/li\u003e\n\u003cli\u003evan Uden, I. W., Tuladhar, A. M., de Laat, K. F., van Norden, A. G., Norris, D. G., van Dijk, E. J.,\u0026hellip;de Leeuw, F. E. (2015). White matter integrity and depressive symptoms in cerebral small vessel disease: The RUN DMC study. \u003cem\u003eAm J Geriatr Psychiatry\u003c/em\u003e,\u003cem\u003e 23\u003c/em\u003e(5), 525-535. https://doi.org/10.1016/j.jagp.2014.07.002\u003c/li\u003e\n\u003cli\u003eWang, Y., Lu, P., Zhan, Y., Wu, X., Qiu, Y., Wang, Z.,\u0026hellip;Zhou, Y. (2021). The Contribution of White Matter Diffusion and Cortical Perfusion Pathology to Vascular Cognitive Impairment: A Multimode Imaging-Based Machine Learning Study. \u003cem\u003eFront Aging Neurosci\u003c/em\u003e,\u003cem\u003e 13\u003c/em\u003e, 687001. https://doi.org/10.3389/fnagi.2021.687001\u003c/li\u003e\n\u003cli\u003eWang, Y., Tu, D., Du, J., Han, X., Sun, Y., Xu, Q.,\u0026hellip;Zhou, Y. (2019). Classification of Subcortical Vascular Cognitive Impairment Using Single MRI Sequence and Deep Learning Convolutional Neural Networks. \u003cem\u003eFront Neurosci\u003c/em\u003e,\u003cem\u003e 13\u003c/em\u003e, 627. https://doi.org/10.3389/fnins.2019.00627\u003c/li\u003e\n\u003cli\u003eWu, Y., Chao, J., Bao, M., \u0026amp; Zhang, N. (2023). Predictive value of machine learning on fracture risk in osteoporosis: a systematic review and meta-analysis. \u003cem\u003eBMJ Open\u003c/em\u003e,\u003cem\u003e 13\u003c/em\u003e(12), e071430. https://doi.org/10.1136/bmjopen-2022-071430\u003c/li\u003e\n\u003cli\u003eYi, L., Wang, J., Jia, L., Zhao, Z., Lu, J., Li, K.,\u0026hellip;Han, Y. (2012). Structural and functional changes in subcortical vascular mild cognitive impairment: a combined voxel-based morphometry and resting-state fMRI study. \u003cem\u003ePLoS One\u003c/em\u003e,\u003cem\u003e 7\u003c/em\u003e(9), e44758. https://doi.org/10.1371/journal.pone.0044758\u003c/li\u003e\n\u003cli\u003eZanon Zotin, M. C., Sveikata, L., Viswanathan, A., \u0026amp; Yilmaz, P. (2021). Cerebral small vessel disease and vascular cognitive impairment: from diagnosis to management. \u003cem\u003eCurr Opin Neurol\u003c/em\u003e,\u003cem\u003e 34\u003c/em\u003e(2), 246-257. https://doi.org/10.1097/wco.0000000000000913\u003c/li\u003e\n\u003cli\u003eZhang, L., Gao, F., Zhang, Y., Hu, P., Yao, Y., Zhang, Q.,\u0026hellip;Zhang, Y. (2022). Analysis of risk factors for the development of cognitive dysfunction in patients with cerebral small vessel disease and the construction of a predictive model. \u003cem\u003eFront Neurol\u003c/em\u003e,\u003cem\u003e 13\u003c/em\u003e, 944205. https://doi.org/10.3389/fneur.2022.944205\u003c/li\u003e\n\u003cli\u003eZhang, Q., Li, J., Bian, M., He, Q., Shen, Y., Lan, Y., \u0026amp; Huang, D. (2021). Retinal Imaging Techniques Based on Machine Learning Models in Recognition and Prediction of Mild Cognitive Impairment. \u003cem\u003eNeuropsychiatr Dis Treat\u003c/em\u003e,\u003cem\u003e 17\u003c/em\u003e, 3267-3281. https://doi.org/10.2147/ndt.S333833\u003c/li\u003e\n\u003cli\u003eZhang, Y., Hu, M., Fan, S., Cao, S., Du, B., Yin, S.,\u0026hellip;Wei, Q. (2024). Altered Resting-State Brain Entropy in Cerebral Small Vessel Disease Patients with Cognitive Impairment. \u003cem\u003eBrain Connect\u003c/em\u003e. https://doi.org/10.1089/brain.2024.0007\u003c/li\u003e\n\u003cli\u003eZhang, Z., Yang, L., Han, W., Wu, Y., Zhang, L., Gao, C.,\u0026hellip;Wu, H. (2022). Machine Learning Prediction Models for Gestational Diabetes Mellitus: Meta-analysis. \u003cem\u003eJ Med Internet Res\u003c/em\u003e,\u003cem\u003e 24\u003c/em\u003e(3), e26634. https://doi.org/10.2196/26634\u003c/li\u003e\n\u003cli\u003eZhu, F., Yao, J., Feng, M., \u0026amp; Sun, Z. (2024). Establishment and evaluation of a clinical prediction model for cognitive impairment in patients with cerebral small vessel disease. \u003cem\u003eBMC Neurosci\u003c/em\u003e,\u003cem\u003e 25\u003c/em\u003e(1), 35. https://doi.org/10.1186/s12868-024-00883-y\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"cognitive impairment, machine learning, prediction model, cerebral small vessel disease, meta-analysis","lastPublishedDoi":"10.21203/rs.3.rs-5365831/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5365831/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eEarly identification of cerebral small vessel disease (CSVD) patients with a higher risk of developing cognitive impairment is essential for timely intervention and improvement of patient prognosis. The advancement of medical imaging and computing capabilities provides new methods for early detection of cognitive disorders. Machine learning (ML) has emerged as a promising technique for cognitive impairment in CSVD. This study aims to conduct a thorough meta-analysis and comparison of published ML prediction models for cognitive impairment in patients with CSVD.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eIn September 2024, relevant studies were retrieved from four databases: PubMed, Embase, Web of Science, and the Cochrane Library. The Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to assess the risk of bias of the ML models. The random effects model was used for meta-analysis of C-index, while a bivariate mixed-effects model was used to calculate the pooled sensitivity and specificity with their 95% confidence intervals (CIs). In addition, to limit the influence of heterogeneity, we also performed sensitivity analyses, a meta-regression, and subgroup analysis.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eTwenty-one prediction models from thirteen studies, involving 3444 patients met criteria for inclusion. The reported C-index ranged from 0.708 to 0.952. The pooled C-index, sensitivity, and specificity were 0.85 (95% CI 0.82\u0026ndash;0.87), 0.82 (95% CI 0.77\u0026ndash;0.87), and 0.81 (95% CI 0.73\u0026ndash;0.87). As one of the most commonly used ML methods, logistic regression achieved a total merged C-index of 0.81, while non logistic regression models performed better with a total merged C-index of 0.86.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eML models holds significant promise in forecasting the risk of cognitive impairment in patients with CSVD. However, future high-quality research that externally validates the algorithm through prospective studies with larger, more diverse cohorts is needed before it can be introduced into clinical practice.\u003c/p\u003e","manuscriptTitle":"Machine Learning Prediction Models for Cognitive Impairment in Cerebral Small Vessel Disease","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-01 16:26:12","doi":"10.21203/rs.3.rs-5365831/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a7a1843c-644e-4036-a58a-48394265bdf4","owner":[],"postedDate":"December 1st, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-04-01T09:24:10+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-01 16:26:12","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5365831","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5365831","identity":"rs-5365831","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}