Early identification of vascular cognitive impairment from a multimodal perspective: a combined diagnosis from targeted cognitive assessments, imaging biomarkers, and molecular fluid biomarkers to ecological behavioral characteristics

Abstract

Vascular cognitive impairment (VCI), the second leading cause of dementia, is characterized by heterogeneous pathophysiology and a potentially reversible early phase, underscoring the need for timely identification. This review synthesizes advances across four complementary domains – targeted cognitive assessments, imaging biomarkers, molecular fluid biomarkers, and ecological behavioral characteristics – conceptualized as the TIME framework. Emerging markers, including the peak width of skeletonized mean diffusivity (PSMD), oxygen extraction fraction, brain‐derived extracellular vesicles, and digital gait metrics, enable the detection of microvascular injury before overt cognitive decline. Given the limitations of single modalities, we advocate for multimodal integration via machine learning to capture the disease continuum from vascular insult to clinical impairment. Establishing a standardized, pathophysiologically anchored classification system, analogous to the AT(N) framework in Alzheimer's disease, is essential to advance precision risk stratification and early intervention in VCI.

Highlights

• We propose a novel TIME framework that integrates targeted cognitive assessments, imaging biomarkers, molecular fluid biomarkers, and ecological behavioral characteristics for the early identification of VCI.

• Emerging biomarkers including PSMD, OEF, BDEVs, and digital gait metrics enable the detection of early microvascular injury prior to overt cognitive decline.

• Multimodal data integration combined with machine learning can capture the full continuum of VCI from vascular insult to clinical cognitive impairment.

• A standardized, pathophysiology‐based classification system analogous to the AT(N) framework in AD is urgently needed for VCI.

1. INTRODUCTION

Vascular cognitive impairment (VCI) represents a spectrum of cognitive disorders driven by cerebrovascular disease or vascular risk factors, encompassing a clinical continuum from mild cognitive impairment (MCI) to overt dementia. ¹ , ² , ³ It is defined as a syndrome characterized by cognitive deficits in at least one domain, corroborated by clinical stroke or subclinical vascular brain injury. ⁴ The complex pathogenesis and profound mechanistic heterogeneity underlying VCI pose significant challenges in establishing universally accepted clinical and research criteria. ⁵ , ⁶ Clinically, this heterogeneity manifests in its ability to present as an isolated entity or to co‐occur with neurodegenerative diseases; notably, such mixed pathologies account for a substantial proportion of dementia cases among the aging population. ⁷

Epidemiological data indicate that approximately 30% of individuals aged 65 and older suffer from cognitive impairment or dementia prior to death. Projections suggest that the global prevalence of dementia will escalate to 150 million by 2050. Within this landscape, VCI accounts for over 20% of all cases, establishing itself as the second leading cause of cognitive decline and dementia worldwide, surpassed only by Alzheimer's disease (AD). ⁸ , ⁹ As global aging rises, the prevalence of cerebrovascular diseases rises commensurately with age, further elevating the lifetime risk of dementia in the elderly. Consequently, VCI has emerged as a major global public health challenge. ¹⁰

As a formidable challenge to public health in aging societies, the development of effective early identification methods for VCI is paramount for ensuring global health security. ¹¹ Unlike the irreversible neurodegeneration typical of AD, VCI offers a critical window for intervention where timely treatment may mitigate disease progression. Despite this potential, current clinical practice is hindered by a lack of globally standardized and specific cognitive scales, neuroimaging features, and hematological biomarkers. Recognizing the profound individual variability in physiological and psychological profiles – such as age and educational attainment – we propose a novel, integrated “TIME” framework. By synthesizing existing cognitive assessments, imaging modalities, and fluid biomarkers with emerging behavioral characteristics, this approach aims to provide a clinically adaptive methodology for the early and precise identification of VCI.

1.1. Multimodal characterization of neurovascular unit (NVU) injury and theoretical basis of TIME framework

Although the clinical presentation of VCI is highly heterogeneous, its core pathophysiological basis arises primarily from structural damage and functional dysfunction of the NVU. ¹² , ¹³ The NVU is a complex multicellular composite composed of neurons, glial cells, pericytes, and the microvascular network, ¹⁴ , ¹⁵ which maintains stable cerebral blood flow and matches the metabolic demands of neuronal activity through precise neurovascular coupling (NVC). ¹⁶ When pathological insults such as ischemia or chronic hypoperfusion occur, the integrity of the NVU is disrupted, triggering a cascade of injuries spanning from molecular micro changes to macroscopic behavioral alterations. ¹⁷

Against this background, the clinically applicable “TIME” framework proposed in this review is not a simple parallel listing of unimodal tools but a comprehensive assessment system constructed based on the longitudinal evolutionary logic of “mechanism‐structure‐function‐behavior” following NVU injury. Specifically:

1. T: Targeted cognitive assessments, representing the ultimate functional outcome, quantify the decline in specific neuropsychological domains (e.g., executive function, memory) caused by cumulative NVU injury. ¹⁸

2. I: Imaging biomarkers, representing structural and perfusion levels, directly visualize microvascular lesions, blood‐brain barrier (BBB) disruption, and subsequent impairment of brain structural networks and perfusion mediated by NVU dysfunction. ¹⁹

3. M: Molecular fluid biomarkers, representing molecular and cellular mechanisms, precisely capture early endothelial injury, neuroinflammation, and microenvironmental imbalance signals of the NVU at the microscopic level. ²⁰

4. E: Ecological behavioral characteristics, representing the dynamic integration and behavioral response of the entire neural network in real‐world environments, reflect the systemic outcome of the transition from microscopic pathology to macroscopic phenotype. ²¹

This review synthesizes advances across four complementary domains – targeted cognitive assessments, imaging biomarkers, molecular fluid biomarkers, and ecological behavioral characteristics – conceptualized as the TIME framework.

1.2. Literature search strategy

This review systematically searched the PubMed, Web of Science, and Embase databases from inception to March 2026.The search focused on the early identification of VCI, covering studies related to cognitive assessment scales, neuroimaging features, fluid biomarkers, behavioral characteristics, and multimodal integrated diagnosis. Peer‐reviewed original articles, systematic reviews, meta‐analyses, and international consensuses were included. Irrelevant literature was excluded by screening titles and abstracts, and the final included studies were confirmed after full‐text review.

2. STAGING FRAMEWORK FOR EARLY IDENTIFICATION OF VCI: DEFINING THE EARLY WINDOW

To achieve effective early intervention for VCI, it is necessary to move beyond static disease definitions and adopt a staging framework that reflects the progressive pathophysiological characteristics of VCI. To clearly define the “early” stage, this review defines this critical intervention window as encompassing two consecutive periods: the preclinical stage and the MCI stage. This definition intentionally excludes late stages such as stable vascular cognitive impairment no dementia (VCIND) and vascular dementia (VaD), aiming to focus on the earliest node at which NVU function has just begun to deviate, at which intervention is most likely to reverse or delay disease progression. Based on the following staging strategy.

In the preclinical stage, individuals have no objective cognitive impairment, or only subjective cognitive decline (SCD). ²² The primary goal of clinical management is the early detection and stratification of at‐risk populations. Therefore, ideal biomarkers must possess extremely high detection sensitivity to capture the most subtle dysfunction signals of the NVU, even at the cost of sacrificing partial specificity. Assessment at this stage focuses on “wide‐range screening” to identify high‐risk individuals.

In the MCI stage, objective domain‐specific cognitive decline begins to manifest. ²³ The diagnostic goal then shifts to etiological confirmation and intervention strategy development. At this point, ideal biomarkers must not only maintain high sensitivity but also high specificity to effectively distinguish vascular pathology from other neurodegenerative diseases.

From this staging perspective, the selection criteria for TIME multimodal biomarkers evolve dynamically, emphasizing high sensitivity to capture upstream mechanistic signals in the preclinical stage and requiring both high sensitivity and specificity for etiological differentiation in the MCI stage. Table 1 summarizes the evolutionary characteristics of each dimension of biomarkers across the VCI continuum in a timeline model, revealing the cross‐stage transition from “functional compensation” to “structural decompensation” and from “subclinical fluctuations” to “explicit functional impairment” (Table 1).

TABLE 1.

Comparison of scales for early diagnosis of VCI.

Tool	Domains assessed	Clinical application and advantages	Diagnostic performance (illustrative values)	Limitations	References
MoCA	Executive function, attention, language, visuospatial ability, memory, orientation	Preferred screening tool for VCI; covers comprehensive domains; sensitive to frontal‐subcortical dysfunction	Approximately 93% sensitivity for MCI; superior to MMSE	Substantial influence from education and cultural background; prone to false positives	16 , 17
MMSE	Orientation, memory, attention, calculation, language	Suitable for large‐scale epidemiological surveys and assessment of moderate to severe dementia; high level of standardization	Low sensitivity for detecting MCI conversion (∼36%); when combined with Clock Drawing Test, sensitivity can increase to 93.7%	Pronounced ceiling effects; lacks in‐depth evaluation of executive function and processing speed; high miss rate for early‐stage identification	20 , 21 , 22
CDR/QDRS	Memory, judgment, community affairs, home/hobbies, personal care (functional assessment)	Standard for staging dementia; emphasizes social function and daily living abilities; commonly used as endpoint in clinical trials	High concordance with clinical staging; consistency between QDRS and CDR reaches 70.66%	Traditional CDR is time‐consuming and requires professional training; QDRS exhibits scoring discrepancies in memory items	18 , 26
DR‐CERAD+IQCODE	Comprehensive cognitive domains; retrospective informant questionnaire	Enables identification across entire disease spectrum; combining tools compensates for individual limitations at different disease stages	AUC for identifying MCI reaches 0.863 (DR‐CERAD); IQCODE achieves an AUC of 0.884 for differentiating dementia	Relatively complex administration procedure; requires informant reports	30 , 31 , 32
Machine Learning Models	Multimodal data (imaging + scales + biomarkers)	Direction of precision medicine; applied in stroke subtype diagnosis, risk stratification, and prognostic prediction	Performance continuously improves with algorithm optimization; capable of processing high‐dimensional features	Lack of standardized validation; clinical interpretability ("black box" issue) remains a challenge	33

3. DEFINITION AND CLASSIFICATION OF VCI

VCI is characterized as a progressive, lifelong process encompassing a spectrum that ranges from MCI to overt dementia. The primary etiology of VCI involves structural brain alterations driven by small vessel disease, large vessel disease, cardioembolism, and intracranial hemorrhage, or the synergistic effects of these vascular pathologies. ⁸ , ²⁴ Conceptually, VCI comprises two major stages: VCIND and VaD. Furthermore, based on distinct vascular pathological features, VCI can be categorized into five clinical subtypes: mixed dementia, post‐stroke cognitive impairment (PSCI), multi‐infarct dementia, and subcortical ischemic VaD. ¹⁸ , ²⁵

4. TOOLS FOR EARLY IDENTIFICATION

4.1. Targeted cognitive assessments (T)

In a meta‐analysis encompassing 86 studies, Salvadori et al. evaluated the performance of eight commonly utilized cognitive scales in identifying cognitive deficits among patients with cerebral small vessel disease (cSVD). Their findings demonstrated that all assessed scales could effectively discriminate between cSVD patients and healthy controls. ²⁶ Building upon established clinical evidence, we systematically summarize and compare the primary and supplementary screening scales currently in mainstream use based on their diagnostic performance and clinical applicability.

4.1.1. Clinical screening: Montreal Cognitive Assessment (MoCA)

MoCA is an assessment tool for screening MCI, covering six core domains: executive function (EFC), attention (AC), language (LANG), visuospatial ability (VIS), memory (MEM), and orientation (ORIEN).

Evidence indicates that MoCA yields high sensitivity for detecting MCI and early‐stage dementia. Specifically, the scale achieves 93% sensitivity (95% CI: 0.88 to 0.96) for MCI, which is significantly higher than the 89% sensitivity of the Mini‐Mental State Examination (MMSE). MoCA also shows clear advantages in identifying non‐AD cognitive impairments, particularly VCI. ²⁷ Owing to its broader coverage of cognitive domains, MoCA is superior for detecting non‐AD cognitive disorders such as VCI. ²⁷ Educational attainment is a major confounder of cognitive test performance. To optimize screening performance, studies have examined educational effects on MoCA subdomains, showing that EFC is most strongly influenced by education, with mild effects on AC, while the other four domains are not significantly associated with education level. Some subdomains, including MEM and ORIEN, are correlated with Marker for Vascular Cognitive Impairment and Dementia (MarkVCID) biomarkers – namely, white matter hyperintensities (WMHs) and free water. ²⁸ However, the MoCA score is also affected by factors beyond education, including ethnicity, age, and regional cultural background. Therefore, reducing false‐positive (FP) results remains an important direction for future investigation.

4.1.2. Clinical screening: MMSE

MMSE is a standardized, non‐self‐rated, multi‐domain cognitive tool widely used in clinical, community, and research settings to screen and evaluate global cognitive function. ¹ MMSE is frequently applied in large‐scale epidemiologic studies, such as those exploring environmental influences on cognitive impairment. ¹ Notably, like MoCA, MMSE is sensitive to educational and cultural background and often exhibits a ceiling effect among highly educated individuals. ¹

On the one hand, MMSE retains value for identifying certain VCI subtypes. A study on VCIND reported that MMSE (cut‐off score: 28) distinguished VCIND from healthy controls with 80% sensitivity and 70% specificity; when combined with the Clock Drawing Test, sensitivity improved to 93.7%, ¹ supporting combined screening as a strategy to overcome its limited early detection capacity.

On the other hand, a consistent evidence‐based consensus documents the limitations of MMSE in early VCI. A systematic review of 11 heterogeneous studies comprising 1569 MCI patients identified only one study focusing on MCI‐to‐VaD conversion, which found a MMSE sensitivity of 36% and specificity of 80%. Based on a 6.2% VaD incidence rate, this single tool would miss five true converters and overdiagnose 19 cases per 100 MCI patients, indicating very low early diagnostic utility. ²⁹ Moreover, about 33% of patients with early dementia are missed despite normal MMSE scores (25 to 30), and 41% of those with mild dementia have normal MMSE values, reflecting insufficient sensitivity for mild VCI. ³⁰ Its sensitivity for detecting specific cognitive changes, such as verbal decline after electroconvulsive therapy, is only 3.6% to 11.1%. ³¹ These deficits stem from inadequate coverage of core VCI‐impaired domains – executive function, processing speed, and attention – which is especially prominent in frontal‐subcortical injury subtypes such as cSVD. In addition, administration and scoring errors (exceeding 10% for place orientation and attention tasks) further undermine data reliability, highlighting the need for standardized protocols in future research. ¹

In summary, MMSE has clinical value for identifying moderate to severe VaD, and combined strategies or tool updates can partially improve its sensitivity. Nevertheless, it performs poorly as a stand‐alone early screening tool for VCI; its application should be strictly bounded and complemented by more domain‐specific instruments.

4.1.3. Clinical staging: Clinical Dementia Rating (CDR)

The CDR assesses six key functional domains. After independent scoring of each domain, a clinician derives a global score (CDR Global Score [CDR‐G]). ³² As an internationally accepted gold standard for staging functional impairment in AD and related dementias (including VCI), the CDR has been validated with high reliability. ³³ It is widely used as a primary endpoint in clinical trials, directly reflecting patients’ symptomatic, functional, and survival status, and enabling evaluation of treatment effects on cognitive function (ADAS‐Cog) and activities of daily living (CDR‐Sum of Boxes [CDR‐SB]). ³⁴ However, traditional CDR assessment is time‐consuming and dependent on specialized personnel, restricting its current use mainly to research settings. To address this efficiency bottleneck, investigators have developed an electronic, remotely self‐administered, auto‐scored CDR, whose concordance with the conventional CDR is currently under validation. ³⁵ , ³⁶

4.1.4. Clinical staging: Quick Dementia Rating System (QDRS)

To meet the demand for ultra‐brief tools in clinical screening and large‐scale epidemiologic surveys, the QDRS was developed. Without specialized clinicians, it produces results in 3 to 5 min that show 70.66% agreement with the total CDR score, greatly improving the feasibility of dementia screening. ³³ Studies report a FP rate of 39.2% and false‐negative (FN) rate of 11.3%, with scoring discrepancies in the memory item as the leading cause of inconsistent global ratings. ³⁷ , ³⁸ In primary care settings, QDRS combined with passive digital markers (PDMs) significantly increases early dementia detection and diagnostic evaluation referral rates (adjusted odds ratio [OR] = 1.31). ³⁹

4.1.5. Combined diagnostic approaches

Each of the aforementioned cognitive assessment tools possesses certain strengths and limitations. Therefore, selecting the appropriate scale based on the clinical context is crucial for optimizing the diagnostic pathway. Although scales such as the MoCA and MMSE are applicable for rapid cognitive screening, comprehensive assessment relies more on tools including the CDR and Dementia Rating Scale (DRS). Given that a single scale possesses inherent emphases and limitations, the combined application of multiple assessment tools serves as a pivotal strategy to facilitate accurate clinical diagnosis. ⁴⁰

A cross‐sectional study from the 2023 Cog‐Aging cohort demonstrated that the combined use of the DRS‐CERAD (DR‐CERAD) and the short form of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) effectively distinguished between different clinical stages of VCI. ² , ⁴¹ Specifically, the DR‐CERAD performed best at discriminating normal cognition from MCI (area under the curve [AUC] = 0.863) and normal cognition from dementia (AUC = 0.987). ⁴² In the diagnostic task of differentiating MCI from dementia, the IQCODE showed the most outstanding performance among all evaluated tools, achieving an AUC of 0.884, demonstrating its superior discriminatory power and potential clinical value. ²⁹ These findings confirm the complementary value of different scales at various nodes along the VCI disease trajectory. It is important to emphasize that regardless of the scale chosen, the interpretation of results must always be integrated with the clinical context. Moreover, it remains challenging for any single tool to completely avoid the influence of factors such as educational level, cultural background, and age on test scores.

Fortunately, the advent of machine learning is substantially improving this situation. Daidone et al. systematically elaborated on the diverse applications of machine learning in stroke medicine, demonstrating that machine learning models exhibit significant efficacy in imaging analysis, stroke subtype diagnosis, risk stratification, treatment guidance, and prognostic prediction, thereby offering a novel technological pathway for precision in stroke care ⁴³ (Table 2).

TABLE 2.

Summary of biomarkers for early diagnosis of VCI.

Biomarker category and key indicators	BEST classification	Validation stage	Early value	Current challenges and specificity	References
Brain‐derived extracellular vesicles (BDEVs) (e.g., cystatin C, specific protein cargos)	Susceptibility / monitoring	Exploratory Research (Peripheral Blood)	Acts as a "liquid biopsy" crossing the BBB; reflects real‐time vascular brain injury repair and pathological progression	Extraction protocols lack standardization; lack of direct, large‐scale prospective validation in pure VCI cohorts.	12 , 93 , 94 , 95 , 96 , 97 , 98 , 161 , 162 , 163
Blood‐brain barrier (BBB) markers (sPDGFRβ, LCN2, MMPs)	Diagnostic / susceptibility	Clinical validation (CSF) Exploratory (peripheral blood)	sPDGFRβ is highly sensitive for capturing early pericyte injury and hippocampal BBB microvascular leakage	CSF versus blood discrepancy: excellent performance in CSF, but peripheral blood validation remains insufficient due to dilution effects	99 , 100 , 101 , 102 , 103 , 104 , 105
Iron metabolism markers (QSM‐detected iron deposition, serum ferritin, hepcidin)	Diagnostic / prognostic	Clinical validation	Regional basal ganglia iron overload is independently associated with executive decline and MCI conversion	Peripheral iron metrics need further correlation with brain deposition; requires high‐field MRI (QSM); interactions with Aβ co‐pathology require cautious interpretation	106 , 107 , 108 , 109 , 110 , 116
Neuroinflammation markers (CRP, IL‐6, GFAP regulators)	Monitoring / susceptibility	Clinical validation (traditional) Exploratory (novel)	Captures the "inflammatory dysfunction" core mechanism (elevated basal levels with blunted stimulated responses)	Traditional broad‐spectrum markers (CRP/IL‐6) lack CNS specificity and yield conflicting results; static levels fail to reflect dynamic in vivo activity	117 , 118 , 119 , 120 , 121 , 122
Oxidative stress (OS) markers (glutathione, oxidized lipids, Nrf2 pathway)	Predictive / prognostic	Exploratory / early clinical	Reduced baseline glutathione independently predicts subsequent executive function decline, serving as a potential upstream driver	Requires longitudinal multi‐omics data to elucidate the temporal causality between OS and neuroinflammation in VCI progression	123 , 124
Classic AD‐related biomarkers (p‐tau181/217, NfL)	Diagnostic (in AD) / prognostic	Clinical application (in AD) Exploratory (in pure VCI)	P‐tau identifies comorbid AD pathology; NfL sensitively reflects generalized neuroaxonal injury and predicts cognitive decline rate	Caution required: highly validated in AD, but lack specificity for VCI. Mostly indicate mixed pathology and cannot independently diagnose pure vascular etiology.	105 , 116
Neuronal injury and angiogenesis markers (NSE, VEGF)	Monitoring	Early clinical validation	Indicate post‐ischemic neuronal damage and compensatory angiogenesis following cerebral hypoperfusion.	Easily confounded by systemic inflammation; low diagnostic value when used as a single marker	104

Note: BEST framework: Biomarkers, EndpointS, and other tools (FDA criteria) [127]. Validation progression: To bridge the gap from “exploratory research” to “clinical application,” international prospective consortia, such as MarkVCID2 [128], are currently conducting large‐scale, cross‐platform validation. By establishing standardized acquisition protocols and harmonized cut‐off values for markers like endothelial function and microstructural injury, these multicenter efforts are essential to overcome the current translational challenges in VCI.

4.1.6. Strategy selection in clinical application scenarios

Despite the wide variety of available cognitive assessment scales, no single scale represents a universally optimal choice. Selecting and appropriately combining tools based on specific clinical scenarios and evaluation objectives is essential for the efficient identification of VCI. Focusing on this rationale, this section outlines a clinical context‐based decision framework to guide standardized selection of cognitive scales.

For large‐scale community‐based or primary care screening settings, the MMSE or QDRS is recommended. In high‐throughput evaluation settings, the MMSE achieves a practical balance between assessment scope and administration time, with a mean paper‐based testing duration of 6.21 min. ⁴⁴ Although its sensitivity for MCI is lower than that of the MoCA, ⁴⁰ its simplicity renders it suitable for population‐level screening. As a rapid informant‐reported questionnaire, the QDRS further improves screening efficiency in primary care settings with limited clinician time. ³³

For precise evaluation in memory or neurology specialist clinics, the MoCA is recommended. The core objective in specialist settings is sensitive detection of early or subtle cognitive impairment, and the MoCA compensates for the limitations of the MMSE in identifying MCI. The MoCA comprehensively assesses cognitive domains commonly affected in VCI, including executive function and attention. At a cut‐off score of <24, the sensitivity is 79.5% (95% CI: 67.1 to 88.0) and specificity is 83.7% (95% CI: 75.4 to 89.6). ⁴⁰ Notably, the CDR is not a cognitive screening tool but the gold standard for functional staging. For patients with a confirmed diagnosis, the CDR can be used to stage functional impairment ³³ and has shown value in monitoring individualized progression. ⁴⁵

Supplementing performance‐based cognitive assessments with informant‐rated tools like the IQCODE is crucial for mitigating educational confounding in individuals with lower schooling, as the latter evaluates decline from a historical baseline rather than absolute performance. ⁴⁶

The diagnostic paradigm for VCI is shifting from reliance on cut‐off values of single scales toward multidimensional data fusion. Integrating performance‐based tests such as the MoCA with functional rating scales such as the CDR, and analyzing multi‐source data using machine learning algorithms, helps construct more accurate and individualized cognitive profiles. Such multimodal integration strategies enhance early identification efficacy and align closely with the TIME framework proposed in this review.

4.2. Imaging biomarkers (I)

Neuroimaging assessment serves as the cornerstone of VCI diagnosis. Its application has evolved from descriptive concepts such as “extensive WMHs” to refined evaluations based on complex pathophysiological mechanisms. The framework of this section is highly consistent with the Standards for Reporting Vascular Changes on Neuroimaging‐2 (STRIVE‐2), ⁴⁷ which systematically define a core panel of markers including recent subcortical small infarcts, presumed vascular lacunes, WMHs, perivascular spaces (PVSs), cerebral microbleeds (CMBs), and the newly added cortical microinfarcts and cortical superficial siderosis (cSS).To establish clearer pathophysiological links, this section reorganizes imaging markers into four key domains in accordance with the latest expert consensus: microstructural integrity, metabolism and hemodynamics, vascular morphology and structure, and peripheral vascular windows.

4.2.1. Markers of microstructural integrity

These markers provide a window into early microstructural damage to brain white matter, which often precedes the appearance of macroscopic lesions on conventional scans.

4.2.1.1. Peak width of skeletonized mean diffusivity (PSMD)

PSMD is a neuroimaging biomarker derived from diffusion tensor imaging (DTI) or diffusion‐weighted magnetic resonance imaging (DWI). Its core computational method involves analyzing the distribution of mean diffusivity (MD) values across a skeletonized representation of whole‐brain white matter, quantifying the peak width as a summary metric. ³⁸ , ⁴⁸ , ⁴⁹ PSMD serves as a quantitative indicator of microstructural white matter integrity in cSVD. ³⁷ This metric has been robustly associated with global cognitive decline (β = −0.8, P < 0.001), an association that remains significant after adjusting for demographic factors and apolipoprotein E (APOE) ε4/ε4 status. ³⁷ A key advantage of PSMD is its ability to capture cognitive impairment beyond that explained by traditional WMH volume, detecting microstructural injury variations not apparent on conventional imaging. Its measurement reproducibility and cross‐platform consistency have been validated through multicenter studies, including the MarkVCID‐1 consortium and three independent cohorts (total sample size N = 7622). Nevertheless, clinical application remains constrained by the complexity of processing pipelines and heterogeneity across different analytical tools. ² , ⁷ Notably, recent evidence from the Framingham Heart Study demonstrates that a high burden of PVSs is significantly associated with elevated PSMD (p < 0.001), suggesting that PSMD may serve as a sensitive quantitative marker of early white matter damage resulting from PVS dysfunction. ⁵⁰

4.2.1.2. Brain free water (FW)

FW refers to the water component within brain tissue (particularly the brain parenchyma) that is not bound by cellular membranes or macromolecular structures and can diffuse freely within the extracellular space. ⁵¹ , ⁵² Within the framework of diffusion magnetic resonance imaging (dMRI), it is quantified as the free water fraction (FWF) or free water volume fraction (FISO), serving as an imaging marker reflecting extracellular water content. ⁵³ , ⁵⁴ As a dMRI‐derived metric, FW is frequently employed to quantify the diffusion component of water molecules in the interstitial space. ⁵⁵ It specifically reflects the proportion of freely diffusing water molecules in the extracellular space of brain tissue, effectively eliminates partial volume effects from cerebrospinal fluid (CSF), and provides a sensitive indicator for neuroinflammation, vascular injury, and microstructural alterations. ⁵⁴ , ⁵⁵ FW levels in typical white matter and key gray matter networks fully mediate the link between cardiovascular biomarkers and 5‐year cognitive decline. ⁵⁵ This metric exhibits a domain‐specific pattern of impairment: FW levels within the default mode network are significantly associated with memory decline, whereas FW levels in the executive control network correlate with EF deterioration. As an emerging quantitative biomarker, threshold definitions and clinical interpretation standards for FW have yet to be harmonized across different centers. ²

Recent investigations have further elucidated the mechanistic significance of FW. In patients with cSVD, the association between high burden of basal ganglia PVSs and WMH volume was fully mediated by FW (β = 0.082, 95% CI: 0.012 to 0.207, p < 0.05), suggesting that impaired interstitial fluid drainage – as reflected by elevated FW – constitutes a critical intermediate step linking enlarged PVSs to white matter injury. ⁵⁶ Moreover, the Framingham Heart Study has confirmed that high PVS burden is independently associated with elevated whole‐brain FW (p < 0.001), with substantial spatial overlap between these two markers along white matter tracts. ⁵⁰

4.2.2. Metabolic and hemodynamic markers

This category of markers evaluates the functional response of the brain to vascular stress and can reveal physiological impairments in oxygen supply and blood flow regulation prior to the occurrence of tissue death.

4.2.2.1. Oxygen extraction fraction (OEF)

OEF is a physiological parameter that quantifies the brain tissue's capacity to extract oxygen from the blood, reflecting the efficiency of tissue oxygen utilization and the dynamic equilibrium between oxygen supply and consumption. ⁵⁷ , ⁵⁸ As a key indicator of cerebral oxygen extraction efficiency, OEF can now be assessed non‐invasively and without contrast agents using advanced MRI techniques such as accelerated T2 relaxation under phase contrast (aTRUPC). ⁵⁹ Studies have demonstrated that OEF is significantly correlated with declines in EF. Specifically, each unit increase in cortical and subcortical OEF is associated with a decrease in EF scores of 0.044 (p = 0.0043) and 0.038 (p = 0.030), respectively, suggesting that this metric exhibits good domain specificity for executive impairment. ⁵⁹ Furthermore, OEF shows a positive correlation with vascular risk scores, reflecting compensatory increases in oxygen extraction by brain tissue under chronic hypoperfusion. ⁵⁹ This elevation appears independently of AD pathological markers, underscoring its potential value in identifying vascular etiology. ⁵⁹ However, the sensitivity and specificity of this biomarker require further validation through large‐scale longitudinal studies. ⁵⁹ Additionally, due to its limited technical availability, standardized protocols for acquisition and post‐processing have yet to be fully established. ²

4.2.2.2. Cerebrovascular reactivity (CVR)

CVR is a hemodynamic metric derived from MRI that reflects the capacity of cerebral blood vessels to modulate their caliber in response to vasoactive stimuli. ⁶⁰ Studies have demonstrated a significant positive association between gray matter CVR and MoCA scores in patients with cognitive impairment (β = 14.97, 95% CI: 3.79 to 26.14, p = 0.009), an association that was particularly pronounced in the EF domain. ⁶¹ Critically, CVR exhibits high specificity for cerebrovascular etiology, as its alterations appear independently of AD pathological markers. However, CVR measurement necessitates dedicated gas stimulation devices and real‐time end‐tidal carbon dioxide monitoring, resulting in substantial technical barriers and operational costs. ² Encouragingly, the MarkVCID2 consortium has initiated multicenter CVR standardization efforts aimed at establishing acquisition protocols and quality control frameworks that are reproducible across different scanner platforms and field strengths, thereby laying the groundwork for clinical application. ⁶²

4.2.3. Vascular morphological and structural markers

These markers represent the classic, visible signatures of established cerebrovascular disease and are used to quantify the cumulative burden of macroscopic parenchymal damage.

4.2.3.1. White matter hyperintensities

WMHs are the most characteristic, widely applied, and best‑validated imaging feature of cSVD in clinical practice. Their burden is strongly associated with cognitive decline and the risk of incident dementia in patients with VCI. Studies have shown that individuals with extensive WMHs carry approximately twice the risk of developing dementia compared with those with mild or no WMHs. ⁶³ , ⁶⁴ In the setting of acute stroke, the severity of WMHs on baseline MRI is a strong predictive risk marker for post‑stroke cognitive impairment. ⁶⁵ Currently, two mainstream approaches are used for WMH assessment in clinical and research settings: visual rating scales and quantitative volumetry, which serve distinct scenarios and provide complementary value in the early detection of VCI.

WMH visual rating scales: Visual rating scales are the most widely used tools in routine clinical practice. Their key advantages include ease of use, no requirement for specialized post‑processing software, and high consistency across different imaging platforms. ⁶⁶ The Fazekas scale is internationally recognized and recommended by the STRIVEv2 consensus; it grades periventricular WMHs and deep WMHs separately from 0 to 3 (periventricular WMH: 0 = no lesions; 1 = cap‑like or linear hyperintensities; 2 = smooth halo‑like hyperintensities; 3 = irregular periventricular hyperintensities extending into deep white matter; deep WMH: 0 = no lesions; 1 = punctate lesions; 2 = lesions beginning to confluence; 3 = large confluent lesions). ⁶⁷ Other commonly used tools include the Scheltens scale, which enables more detailed regional assessment of WMH burden. ⁶⁸

Visual rating scales are particularly suitable for large‑scale community screening, clinical evaluation in primary care settings, and emergency scenarios, allowing rapid risk stratification of VCI. ⁶⁹ However, their limitations are substantial: They have low sensitivity to subtle, progressive changes in WMH burden and cannot detect the small, subthreshold volumetric alterations in early VCI that precede overt cognitive decline. ⁶⁷

WMH quantitative volumetry: WMH quantitative volumetry is derived from T2 fluid‐attenuated inversion recovery MRI sequences, using semi‑automatic or fully automatic segmentation algorithms to generate continuous, objective values of total WMH volume, usually expressed as a percentage of intracranial volume to adjust for individual differences in brain size. ⁶⁷ Compared with visual rating scales, quantitative volumetry has markedly higher sensitivity to subtle changes in lesion burden and can detect subthreshold microstructural damage that cannot be captured by ordinal visual scoring. ⁷⁰ Longitudinal studies have confirmed that the annual rate of change in WMH volume is independently associated with the degree of decline in core cognitive domains affected by VCI, ⁷¹ making it an ideal biomarker for early disease monitoring and treatment evaluation in research settings.

However, the clinical translation of WMH quantitative volumetry remains limited. A multicenter study pooled WMH data from 15,653 individuals and found that data across independent cohorts could not be directly integrated for cross‑study analysis without complex spatial standardization and statistical harmonization, due to the lack of standardized segmentation protocols and universal diagnostic thresholds across different imaging platforms and sequences. ⁷² Cross‐platform standardization deficiency is a key methodological limitation restricting routine WMH quantitative research in primary care.

In summary, visual rating scales are the preferred tool for initial clinical screening and rapid risk stratification, whereas WMH quantitative volumetry is more suitable for precise early detection, longitudinal follow‑up, and trial endpoint evaluation in research settings. The combined use of both methods enables comprehensive assessment of WMH burden throughout the full disease course of VCI.

4.2.3.2. Infarct lesions

The impact of infarcts on cognition is highly dependent on their location and type. Studies have demonstrated that infarcts located in the left prefrontal cortex, left thalamus, and right parietal lobe are directly associated with the occurrence of post‐stroke cognitive impairment (false discovery rate‐corrected q < 0.01; OR > 20% at the voxel level). ⁶⁶ Furthermore, the number and total volume of lacunes (old small deep infarcts) correlate positively with the severity of impairment in cognitive domains typically affected by vascular pathology, such as EF and processing speed. ⁷³ , ⁷⁴ Although computed tomography (CT) can detect some old infarcts, the evidence supporting its utility as an independent predictor of post‐stroke cognitive impairment remains inconsistent, with conflicting findings across large‐scale studies and meta‐analyses, necessitating further validation. ⁷⁵ , ⁷⁶ , ⁷⁷

4.2.3.3. Cerebral microbleeds and microvascular injury

CMBs represent another important marker of cSVD. Research suggests that CMBs are associated with an increased risk of dementia in both general and high‐risk elderly populations (hazard ratio [HR] = 1.98, 95% CI: 1.55 to 2.53, p < 0.001). ⁶³ However, two memory clinic studies demonstrated no association between CMB and incident dementia, with no statistical heterogeneity observed across studies (relative risk [RR] = 1.02, 95% CI: 0.68 to 1.51, p = 0.94). ⁷⁸ This discrepancy suggests that the relationship between CMBs and dementia may be population‐dependent. In cognitively normal individuals from the general population and those with high vascular risk, CMBs may serve as markers of early vascular brain injury, predicting long‐term dementia risk. In contrast, among memory clinic patients with established cognitive impairment, CMBs may represent merely one of several coexisting pathologies, exerting no independent additive effect on already manifest cognitive decline. Furthermore, as CMBs cannot be reliably detected by CT, susceptibility‐weighted imaging or T2*‐weighted gradient‐recalled echo sequences on MRI remain the gold standard for their detection. Additionally, cerebral microinfarcts, representing more subtle ischemic injury, have demonstrated visibility on 3T MRI, along with associated changes in surrounding cortical thickness, and have been shown to correlate with cognitive function, exhibiting potential as predictive markers. ⁷⁹

4.2.4. Emerging markers in the STRIVE‐2 criteria

Notably, the STRIVE‐2 criteria specifically emphasize the importance of cortical microinfarcts (CMIs) and cSS in the evaluation framework of cSVD. ⁴⁷ cSS is mainly detected by conventional susceptibility‐weighted imaging (SWI) and the novel quantitative susceptibility mapping (QSM), which reflect iron levels in the cortical and subcortical regions. ⁸⁰ , ⁸¹ Studies have shown that QSM quantifies the burden of cortical iron deposition and is positively correlated with WMH burden. ⁸¹ In the COMPASS‐ND cohort, global cSVD scores and hypertensive arteriopathy‐specific cSVD scores that included cSS were independently associated with more severe cognitive diagnoses (adjusted OR = 1.14 to 1.18). ⁴⁷

CMIs are novel MRI markers of cerebrovascular disease. They are easily missed on conventional imaging but can be identified as small focal signal abnormalities within the cortex using high‐resolution 3T MRI and predict accelerated cognitive decline. ⁷⁹ A recent longitudinal study found that baseline CMIs significantly predicted new incident CMIs within 2 years, and incidental CMIs were independently associated with cognitive decline and incident dementia, further confirming the clinical value of CMIs as core imaging biomarkers of VCI. ⁸²

Nevertheless, the widespread application of these markers in routine clinical practice remains challenging. They represent a more refined dimension of vascular damage affecting cognitive function and represent an important direction for future research on imaging biomarkers.

4.2.4.1. Brain atrophy and cerebrovascular function

Brain atrophy in VCI frequently manifests as brain tissue loss secondary to vascular injury. ¹ Notably, voxel‐based morphometry analyses suggest that patterns of gray matter atrophy in patients with subcortical VCI may differ from those observed in AD. ⁸³ Furthermore, a meta‐analysis demonstrated that CT‐visible brain atrophy serves as a significant predictor of PSCI (four studies, N = 558, OR = 2.80, 95% CI:  1.21 to 6.51). ⁷⁵ As a key indicator for assessing cerebrovascular regulatory capacity, CVR offers a dynamic functional dimension that extends beyond conventional static structural assessments. Research has confirmed that impaired CVR is closely associated with global cognitive decline in older adults, suggesting its potential as a predictive marker for VCI. ⁶¹

4.2.4.2. Thalamic gray matter volume and cerebellar perfusion

Thalamic gray matter volume, derived from structural MRI, and cerebellar perfusion, assessed using arterial spin labeling, together constitute a novel combined assessment metric. ⁸⁴ In patients with subcortical vascular MCI, reduced thalamic volume and altered cerebellar perfusion are significantly associated with gait parameters, including gait speed and Timed Up and Go test results (p < 0.025). ⁸⁴ Compared to AD, these alterations exhibit potential specificity for vascular etiology. However, the precise mechanisms linking these imaging changes to gait impairment remain to be elucidated, and translation to clinical practice will require integration within a multimodal assessment framework. ⁸⁴

4.2.5. Peripheral vascular window

4.2.5.1. Radial peripapillary capillary (RPC) perfusion area

RPC perfusion area, derived from optical coherence tomography angiography (OCTA), enables non‐invasive quantitative assessment of retinal microcirculation. In patients with asymptomatic internal carotid artery stenosis, the RPC perfusion area demonstrated superior diagnostic performance in distinguishing vascular MCI from VaD compared to total cSVD burden scores and WMH volume, with an optimal cut‐off value of 0.068 mm². ⁸⁵ This technique offers rapid acquisition, requires no contrast agent, exhibits high patient tolerability, and holds promise for large‐scale community‐based screening. However, existing evidence is entirely confined to populations with internal carotid artery stenosis, and its generalizability to cohorts with other vascular etiologies remains to be validated. ⁸⁵

Furthermore, a meta‐analysis encompassing 36 studies revealed significant reductions in retinal vessel density among patients with AD and MCI. Compared to cognitively normal controls, pooled standardized mean differences (SMDs) for vascular density ranged from −0.69 to −0.38 in the AD group and from −0.36 to −0.17 in the MCI group. Nevertheless, the available evidence is heavily concentrated within the AD spectrum, with systematic validation in VCI populations remaining largely absent. ⁸⁶ Additionally, methodological heterogeneity, inadequate control of confounding factors, and the lack of standardized protocols may constitute major barriers to the clinical translation of retinal biomarkers. ⁸⁷

4.2.5.2. Multimodal imaging combination strategy

Each imaging biomarker provides unique pathophysiological insights, yet its true diagnostic value lies in the integration and quantitative analysis of multiple markers. In contrast, the multimodal imaging combination strategy can construct a more comprehensive individualized pathophysiological profile by analyzing the total aggregated burden of multiple biomarkers.

1. Complementarity between structure and function: Combining structural biomarkers with hemodynamic and metabolic biomarkers enables effective differentiation between individuals with similar static lesion burdens but vastly disparate vascular elasticity and metabolic compensatory capacities. This rationale helps explain clinical heterogeneity whereby some patients present severe imaging findings but mild clinical symptoms.

2. Early warning from microstructural to macrostructural changes: Integrating microstructural biomarkers indicative of early potential damage with macrostructural biomarkers allows for the identification of high‐risk individuals with disrupted white matter microstructural integrity and on the verge of physiological stress decompensation, prior to the emergence of extensive confluent white matter lesions.

3. Machine learning‐driven decision‐level integration: Future core advances will depend on machine learning algorithms capable of integrating multi‐source heterogeneous data from structural, diffusion, perfusion, and functional imaging. Using feature‐level or decision‐level integration strategies, multidimensional imaging data can be transformed into robust, individualized risk scores or subtype classifications for cognitive impairment associated with cSVD. This is the precise goal of large‐scale research initiatives such as the MarkVCID project, which aims to validate the value of multimodal biomarker combinations in clinical risk stratification and disease monitoring.

In summary, the choice of imaging technique should be guided by specific clinical questions and available resources. MRI is the first choice due to its ultra‐high sensitivity to microstructures and microbleeds, while CT is suitable for acute evaluation owing to its speed and accessibility. Ultimately, the goal is to move beyond simple lesion counting and establish a dynamic evaluation framework based on aggregated analysis and holistic interpretation of multimodal imaging.

4.2.6. Imaging technology selection and future directions: shared strengths and translational challenges

While these imaging markers transcend conventional WMH assessment to multidimensionally reflect vascular and neurovascular pathophysiology and enable preclinical cognitive risk stratification, their clinical translation is hindered by inadequate standardized protocols, limited multicenter longitudinal evidence, and insufficient validation for widespread application in VCI. ⁵⁵ , ⁸⁸

Regarding integrative trends, the MarkVCID2 consortium has completed baseline enrollment of 1883 participants from diverse racial backgrounds and will prospectively validate the aforementioned candidate markers over a 3‐year period, aiming to establish clinically meaningful thresholds for two application scenarios: participant selection and disease progression monitoring. ⁶² Notably, these research objectives align closely with the Framework for Clinical Trials in Cerebral Small Vessel Disease (FINESSE) framework, which provides consensus‐based recommendations for optimizing clinical trial design in VCI and cSVD. ⁸⁹ The FINESSE framework defines the overarching strategic direction, emphasizing the use of validated biomarkers for trial population enrichment and as surrogate endpoints. ⁸⁹ In turn, MarkVCID2 offers actionable operational tools to achieve these goals, including standardized imaging acquisition protocols, cross‐platform data harmonization, and clinically practical cut‐off values for candidate biomarkers. ⁹⁰ Furthermore, multimodal strategies combining plasma biomarkers such as S100B, neurofilament light chain (NfL), and quantitative BBB function assessments have emerged as important directions for improving early diagnostic accuracy in VCI. ⁹¹ Nevertheless, the development of algorithms specifically tailored to capture vascular etiology remains in its early exploratory stages. ⁹²

In clinical practice, MRI offers irreplaceable advantages for detecting WMHs, acute infarcts, CMBs, and microstructural alterations due to its high soft tissue resolution and multi‐sequence capabilities, establishing it as the preferred tool for VCI assessment. CT, by contrast, plays a pivotal role in acute stroke screening, emergency evaluation, and resource‐limited settings owing to its rapid acquisition, widespread availability, lower cost, and absence of absolute contraindications. Future research must focus on quantifying core imaging markers and standardizing diagnostic thresholds, advancing the integration of multimodal imaging with artificial intelligence to construct more robust predictive models, and optimizing imaging assessment strategies across diverse clinical scenarios and population cohorts, thereby addressing the diagnostic challenges posed by the marked heterogeneity of VCI (Figure 1).

FIGURE 1.

Imaging features. This schematic illustrates advanced MRI‐based markers capturing distinct pathophysiological dimensions of VCI. OEF reflects cerebral metabolic reserve, with elevations indicating compensatory responses to chronic hypoperfusion. PSMD quantifies microstructural white matter integrity loss beyond conventional white matter hyperintensities. FW detects extracellular fluid accumulation secondary to neuroinflammation and impaired interstitial fluid drainage. The RPC perfusion area enables non‐invasive retinal microvascular assessment. CVR measures dynamic vasodilatory capacity. Despite their potential for preclinical risk stratification, clinical translation remains constrained by lack of standardized acquisition protocols, cross‐platform variability, and limited multicenter validation. Ongoing initiatives such as MarkVCID2 aim to establish clinically meaningful thresholds and harmonize methodologies across diverse populations.

4.3. Molecular fluid biomarkers (M)

4.3.1. Brain‐derived extracellular vesicles (BDEVs)

As key nanoscale carriers released from the central nervous system (CNS) into the periphery, BDEVs enable the precise transport of molecular information from the brain to bodily fluids by virtue of their capacity to cross the BBB. ⁹³ , ⁹⁴ , ⁹⁵ Owing to their ability to dynamically reflect cerebral pathophysiological states in real time, BDEVs have emerged as highly promising biomarkers in the field of non‐invasive liquid biopsy for CNS disorders, demonstrating substantial potential for the early identification of VCI.

First, BDEVs are intimately associated with core pathological processes in VCI and VCID. The 2025 VCID conference proceedings indicated that the quantity or protein composition of specific BDEV subtypes in plasma correlates significantly with cerebral inflammatory responses and cognitive decline, suggesting that BDEVs may directly reflect disease activity in VCID. ¹² Second, the cargo carried by BDEVs carries pathophysiological significance. For instance, in ischemic stroke models, BDEVs have been shown to transport neuroprotective molecules such as cystatin C, indirectly suggesting their involvement in and reflection of repair mechanisms and pathological progression following vascular brain injury. ⁹⁶ Given that ischemic stroke constitutes a major etiology of VCID, this finding indirectly suggests that BDEVs may reflect reparative or pathological processes following vascular brain injury, thereby holding potential value as VCID‐related biomarkers. Furthermore, evidence supporting the role of BDEVs in other cerebrovascular pathological conditions lends credence to their applicability in VCID. For instance, BDEVs have been shown to exhibit highly expressed phosphatidylserine following traumatic brain injury, which can activate the extrinsic coagulation pathway and contribute to coagulation dysfunction. ⁹⁷

Although these studies did not directly validate the association between BDEVs and VCID, the close relationship they reveal between BDEVs and vascular pathological processes – such as coagulation abnormalities and endothelial injury – suggests that BDEVs may also play an important role in VCI. The implicated biological mechanisms may have cross‐disease relevance. Multi‐omics analyses have revealed that BDEVs carrying specific risk genes harbor cargo that can promote neuroinflammation and vascular endothelial dysfunction, implicating BDEVs as potentially significant contributors to key pathogenic mechanisms in VCID. ⁹⁸ Consequently, BDEVs possess potential as dynamic indicators reflecting disease progression in this condition.

However, this field remains in its early stages of exploration. Despite the aforementioned indirect evidence, direct validation studies in VCID populations remain scarce. Future prospective studies and harmonized detection standards are urgently needed to establish the clinical utility of BDEVs.

4.3.2. BBB biomarkers

As an essential regulator of CNS homeostasis via restricting peripheral harmful substances and stabilizing the neural microenvironment, BBB disruption, especially early hippocampal microvascular leakage, drives aging and cognitive decline prior to typical pathological changes; thus, biomarkers reflecting BBB dysfunction and neurovascular unit injury are critical for early VCI identification. ⁹⁹ , ¹⁰⁰

Among molecules reflecting BBB damage, soluble platelet‐derived growth factor receptor beta (sPDGFRβ) demonstrates particularly high sensitivity. As a specific product of pericyte injury, CSF levels of sPDGFRβ increase significantly with aging and pericyte loss, and this elevation occurs independently of amyloid beta (Aβ) or tau pathology, directly correlating with hippocampal BBB disruption. ¹⁰¹ , ¹⁰² Additionally, peripheral blood lipocalin‐2 (LCN2) has emerged as a potential screening tool. Studies have found that serum LCN2 levels are significantly upregulated in the early stages of cognitive impairment and correlate negatively with MoCA scores (p < 0.001), suggesting its potential as a novel target for monitoring cerebrovascular inflammation and enabling early diagnosis of VCI. ¹⁰³

For clinically prevalent mixed dementia (Mx), single‐domain biomarkers often prove insufficient for accurate classification. A biomarker panel incorporating matrix remodeling factors (matrix metalloproteinase [MMP]‐1/3/10), angiogenic factors (vascular endothelial growth factor‐C [VEGF‐C], placental growth factor [PlGF]), and cytokines (interleukin (IL)‐2/6/13) has demonstrated significant potential for enhancing diagnostic accuracy by capturing the underlying BBB‐related pathological mechanisms. ¹⁰⁴ Investigating CSF protein profiles distinguishing patients with subcortical vascular disease (SVD) and Mx from healthy controls, one research team identified a biomarker combination of significant value for differential diagnosis between SVD and AD. This panel included total tau (t‐tau), tau phosphorylated at threonine 181 (p‐tau181), and Aβ1‐42 reflecting AD pathology, NfL and myelin basic protein (MBP) reflecting white matter injury, and MMP‐9 and tissue inhibitor of metalloproteinases‐1 (TIMP‐1) reflecting matrix remodeling. The data indicated that this combination achieved 89% sensitivity and 90% specificity for distinguishing subcortical vascular disease from AD. Further analysis revealed that MBP and NfL were specific markers for identifying SVD, whereas t‐tau and Aβ1‐42 primarily contributed to AD identification. These findings suggest that combining CSF markers spanning distinct pathological dimensions enhances diagnostic accuracy in differentiating SVD from AD. ¹⁰⁵

In summary, although multiple high‐value biomarkers have been identified, the inherent heterogeneity of the disease precludes any single marker from serving as a diagnostic gold standard. Therefore, future research should prioritize the development of multi‐marker diagnostic models grounded in pathophysiological mechanisms. By integrating peripheral markers such as LCN2 for initial screening, followed by precise assessment using sPDGFRβ and imaging indicators, early intervention and individualized management of VCI and its mixed pathologies may become achievable.

4.3.3. Iron metabolism, inflammation, and oxidative stress (OS) markers

4.3.3.1. Iron and its metabolism

Dysregulation of iron metabolism represents a key biological process for the early identification of VCI. Regarding neuroimaging, studies utilizing QSM have confirmed that patients with cSVD exhibit significantly increased iron deposition in bilateral basal ganglia structures, including the caudate nucleus, putamen, and globus pallidus. This increased iron load is independently associated with declines in EF and global cognition: QSM values in the putamen correlate negatively with MMSE scores (β = −0.35, p = 0.003), while caudate nucleus QSM values correlate negatively with EF scores (β = −0.28, p = 0.01) and are also linked to compromised CNS integrity. ¹⁰⁶ , ¹⁰⁷ Furthermore, elevated magnetic susceptibility in the entorhinal cortex and putamen predicts conversion from cognitively normal aging to MCI, with substantially higher risk observed in individuals harboring concomitant Aβ deposition (entorhinal cortex: HR = 2.00, 95% CI: 1.23 to 3.23, p = 0.005 for the overall cohort; HR = 3.59, 95% CI: 1.70 to 7.57, p < 0.001 for the PET‐positive subgroup). ¹⁰⁸ Additionally, cortical iron deposition is independently associated with declining MMSE and MoCA scores and exhibits spatial colocalization with cortical atrophy patterns. ¹⁰⁹ Ultra‐high‐field 7T MRI has further revealed that iron deposition specifically within the hippocampal CA1 subfield and that the subiculum mediates memory decline. Regarding peripheral iron metabolism, elevated serum ferritin levels (female > 200 ng/mL, male > 300 ng/mL) significantly increase the risk of cognitive impairment (OR = 1.39, 95% CI: 1.11 to 1.74), with particularly pronounced susceptibility observed in individuals aged 60 to 69 years. Serum iron exhibits a U‐shaped non‐linear association with cognition, whereas the trend for low iron levels did not reach statistical significance ¹¹⁰ Mechanistically, hepcidin deficiency leads to hippocampal iron accumulation, neural stem cell depletion, and elevated tumor necrosis factor α (TNF‐α), elucidating the driving role of the iron metabolism–neuroinflammation axis. ¹¹¹ apoE4 disrupts brain iron homeostasis by interfering with endosomal maturation and transferrin receptor expression. ¹¹² Iron overload activates the TFR1/THBS1 pathway, triggering ferroptosis, while iron chelation and antioxidant interventions ameliorate neurological function. ¹¹³ , ¹¹⁴ Additionally, in cerebral amyloid angiopathy accompanied by inferior temporal cortical iron deposition, perceptual speed decline is particularly pronounced (p = 0.008). ¹¹⁵ Synthesizing available evidence, Delaby et al. demonstrated that among blood‐based biomarkers for diagnosing VaD, NfL, and hepcidin exhibited particularly high diagnostic value, with performance superior to many other circulating biomarkers under investigation and potentially even surpassing certain CSF markers. ¹¹⁶ Collectively, this evidence establishes brain iron homeostasis as a core component of the multimodal biomarker panel for VCI and provides translational targets for early intervention.

4.3.3.2. Neuroinflammation

Neuroinflammation constitutes a core mechanism in VCI; however, identifying reliable and specific circulating inflammatory markers remains challenging. The application of traditional broad‐spectrum inflammatory markers has significant limitations. For instance, different isoforms of C‐reactive protein (CRP) – the native pentameric (nCRP) and monomeric (mCRP) forms – possess distinct and even opposing biological functions, and early studies that failed to differentiate between them yielded conflicting conclusions. ¹¹⁷ Findings regarding IL‐6 have also been inconsistent, although a meta‐analysis suggested that blood levels of IL‐6 may be specifically elevated in VaD compared to AD, offering some differential diagnostic value. ¹⁰⁴ , ¹¹⁸ Other studies found no differences in static IL‐6 levels between VaD and AD. ¹¹⁹ More critically, research has revealed a phenomenon of “inflammatory dysfunction” in VCI patients, characterized by a state of elevated basal inflammation alongside significantly impaired capacity of peripheral immune cells to release pro‐inflammatory cytokines such as TNF‐α and IL‐6 upon ex vivo stimulation. ¹²⁰ This indicates that merely detecting static circulating factor levels is insufficient for assessing complex in vivo inflammatory activity; dynamic interpretation incorporating immune cell functional responses is required.

Consequently, research frontiers have shifted toward novel markers with greater CNS specificity and functional relevance. On the one hand, regulators targeting glial cells have emerged as focal points; for instance, inhibiting the microglial key receptor CSF1R has been shown to ameliorate vascular white matter lesions and associated cognitive deficits by attenuating neuroinflammation. ¹²¹ On the other hand, liquid biopsy techniques enabling the capture of brain‐derived exosomes from peripheral blood, coupled with detection of their cargo containing specific neuroinflammation‐related proteins such as neural cell adhesion molecule, offer a novel avenue for non‐invasive assessment of central inflammatory status. ¹²²

4.3.3.3. Oxidative stress

OS is considered a key driver of VCI pathogenesis, potentially operating upstream of neuroinflammation. Compared to dynamically complex inflammatory markers, OS‐related biomarkers demonstrate potential as early, stable predictors. A 4‐year longitudinal study revealed that reduced baseline levels of circulating glutathione, a key antioxidant (adjusted RR = 1.70, 95% CI: 1.02 to 2.85, p = 0.04), independently predicted subsequent EF decline, whereas concurrently measured classical inflammatory marker CRP showed no such association. ¹²³ This finding challenges the traditional view of neuroinflammation as the sole early driver and suggests that redox imbalance may play a more central role in initiating cognitive decline.

At the molecular level, dysfunction of Nrf2, the core transcription factor governing antioxidant responses, is closely linked to VCI pathology. The Nrf2 signaling pathway declines with age, and loss of its function exacerbates cellular senescence while promoting the production of senescence‐associated secretory phenotype (SASP) factors, thereby worsening inflammatory status in critical brain regions such as the hippocampus. ¹²⁴ This elucidates the fundamental link between oxidative stress and neuroinflammation. Consequently, monitoring circulating markers reflecting OS status – such as glutathione, oxidized lipids, and indicators of Nrf2 pathway activity – not only aids in early VCI identification but also provides crucial insights into the temporal sequence of disease pathogenesis. Future research should leverage longitudinal multi‐omics data to further delineate the interactive network between oxidative stress and neuroinflammation in VCI progression (Table 3).

TABLE 3.

Timeline model of evolution of T/I/M/E multimodal biomarkers in VCI.

Disease stage	Preclinical stage	Mild cognitive impairment (MCI) stage	Dementia stage (VaD)	Reference
Core Characteristics	Asymptomatic or only subjective cognitive decline (SCD) is present. The pathophysiological cascade has initiated, but objective cognitive tests remain within the normal range.	Objective cognitive decline; impairment in one or more cognitive domains, but not yet severely impacting instrumental activities of daily living (IADL).	Significant cognitive decline sufficient to interfere with social function and independent living.	1 , 4 , 22 , 23
(T) Targeted cognitive assessments	Normal range: MoCA/MMSE scores are normal. Subjective complaints may be present but are not supported by objective neuropsychological evidence.	Mild abnormality: Significant decline in MoCA scores (more sensitive than MMSE) Clinical Dementia Rating (CDR) Global Score of 0.5	Marked abnormality: Severely reduced MoCA/MMSE scores CDR Global Score ≥ 1.0	22 , 27 , 30 , 32 , 40
(I) Imaging biomarkers	Upstream mechanistic markers: Blood‐brain barrier (BBB) injury (sPDGFRβ↑) Impaired white matter microstructural integrity (PSMD↑, FW↑) Compensatory changes in cerebral oxygen metabolism and blood flow regulation (OEF↑, CVR↓) Dysregulation of iron homeostasis (increased iron deposition in brain regions↑)	Markers of accumulated pathological burden: Increased white matter hyperintensity (WMH) volume Reduced radial peripapillary capillary (RPC) perfusion area Thalamic gray matter atrophy	Widespread structural damage: Severe WMHs, multiple lacunar infarcts, significant global or regional atrophy	37 , 55 , 59 , 61 , 63 , 66 , 71 , 73 , 75 , 85 , 101 , 102 , 106 , 108
(M) Molecular fluid biomarkers	Early sensitive markers: Alterations in molecular cargo of brain‐derived extracellular vesicles (BDEVs)	Disease‐specific and differential markers: Elevated plasma neurofilament light chain (NfL) levels Upregulation of specific inflammatory factors (e.g., LCN2) AD‐related pathological markers (for differential diagnosis)	Markers of disease severity and prognosis: Biomarker levels are markedly abnormal, primarily used for disease staging and progression monitoring	12 , 98 , 103 , 105 , 116
(E) Ecological behavioral characteristics	Subclinical changes: Subtle alterations in gait variability or keystroke dynamics, detectable only through high‐precision wearable sensors	Observable motor‐behavioral changes: Reduced gait speed Increased gait variability Possible emergence of AGED phenotype (apathy, gait disturbance, executive dysfunction)	Overt functional impairment: Prominent gait apraxia, balance disorders, and mobility difficulties emerge	126 , 127 , 128 , 130 , 132 , 134

4.4. Validation stages of biomarkers and the BEST classification framework

Although the aforementioned biomarkers show great potential in elucidating the pathophysiological mechanisms of VCI, their maturity in clinical translation varies substantially. To provide clinicians with a clear translational roadmap, we systematically reclassified the above biomarkers based on the BEST (Biomarkers, EndpointS, and other tools) framework proposed by the US Food and Drug Administration (FDA) ¹²⁵ and clearly defined their respective validation stages, namely, exploratory research, clinical validation, and commercial application (Table 3).

First, caution is required regarding the application bias of established AD biomarkers in VCI. Plasma NfL and p‐tau181/217 show excellent performance in AD diagnosis (AUC > 0.90). However, elderly individuals often present with mixed pathology. NfL only reflects generalized neuroaxonal injury, while p‐tau mainly indicates tau protein burden; neither can specifically identify pure vascular lesions. Therefore, within the BEST framework for VCI, they are more suitable for comorbidity identification and prognostic assessment rather than specific diagnostic indicators.

Second, asymmetric validation between central and peripheral compartments represents a major barrier to current translation. Taking sPDGFRβ as an example, its value as a CSF marker for BBB damage is well recognized. However, its validation in peripheral blood is still preliminary due to blood dilution effects and limited detection sensitivity. In addition, emerging targets such as BDEVs remain in the exploratory validation stage, and their monitoring value in large clinical cohorts needs further clarification.

To address these gaps in validation, international prospective consortia such as MarkVCID2 are filling the evidence base through multicenter collaboration. Unlike traditional exploratory studies, MarkVCID2 focuses on standardized clinical validation of endothelial function, white matter microstructural injury, and inflammatory markers in trans‐ethnic, cross‐device, large‐sample cohorts. This process aims to transform “research hotspots” into clinically practical tools with unified cut‐off values, thereby enabling precise subtyping of VCI. ⁶²

4.5. Ecological behavioral characteristics (E)

4.5.1. Gait characteristics

Beyond traditional diagnostic approaches, numerous investigators have observed that gait abnormalities – such as reduced gait speed and increased step length variability – frequently precede overt cognitive decline, suggesting their potential as early surrogate markers for preclinical VCI. ¹²⁶ Studies have demonstrated significant associations between multiple gait parameters, including stride velocity and swing velocity, and cognitive impairment attributable to cSVD (stride velocity: OR = 0.95, 95% CI: 0.92 to 0.98;swing velocity: OR = 0.73, 95% CI: 0.57 to 0.94). ¹²⁷ Further research has revealed that gait variability, defined as fluctuations in step timing or length, not only correlates closely with neurodegeneration and cognitive impairment but also exhibits differential patterns across dementia subtypes. ¹²⁸ Moreover, combining gait assessment with oculomotor features, such as anti‐saccade accuracy, may enhance screening efficiency for early cSVD‐related cognitive impairment, providing a more comprehensive reference for identifying at‐risk populations. ¹²⁹

Building upon these theoretical foundations, one research team developed a predictive model utilizing wearable inertial sensors to capture gait speed and variability. This model demonstrated accuracy in discriminating between cognitively impaired individuals and healthy controls that approached that of the MMSE (development dataset: AUC difference = 0.026, p = 0.542; validation dataset: AUC difference = 0.070, p = 0.330). ¹³⁰

4.5.2. Other behavioral characteristics and combined gait diagnostics

Beyond gait analysis, researchers have also investigated facial expression characteristics as potential behavioral markers. By extracting four categories of facial expressions from video recordings of patients and employing decision tree models for prediction and assessment, studies have revealed a relatively significant association between dementia and emotional states. ¹³¹ This finding suggests that facial behavioral features – including action units, emotion categories, valence‐arousal dimensions, and facial embeddings – may enable earlier and more non‐invasive diagnosis of cognitive impairment.

Gait and keystroke assessments, as non‐invasive behavioral indicators, are effective for early MCI and dementia screening. ¹³² Combined detection significantly elevates diagnostic accuracy (e.g., AUC > 0.9) but remains ineffective for subjective cognitive decline; meanwhile, keystroke features assist in differentiating dementia subtypes. ¹³³ Furthermore, AI‐based analysis and the apathy, gait, and executive dysfunction (AGED) phenotype provide novel pathways for early VCI identification, compensating for conventional evaluation deficiencies. ¹²⁹ , ¹³⁴

Gait analysis, as a classic behavioral paradigm with a long research history, can be optimally utilized for early behavioral assessment when combined with machine learning approaches. By employing long short‐term memory (LSTM) network classifiers trained on a series of gait behavioral indicators, ¹³⁵ researchers have developed recognition methods that function effectively outside specialized clinical environments, contributing positively to early identification and diagnosis. Furthermore, some scholars have proposed that Multi‐view Gait Generative Adversarial Networks (MvGGAN) can expand existing gait datasets, providing sufficient samples for deep learning‐based cross‐view gait recognition methods. ¹³⁶ Collectively, these advances suggest that integrating artificial intelligence and deep learning into diagnostic techniques holds potential to create novel value for clinical assessment.

4.5.3. Mild behavioral impairment and neuropsychiatric characteristics

Beyond motor manifestations such as gait disturbance, mild behavioral impairment (MBI) is an important early behavioral biomarker of VCI. ¹³⁷ Cognitive decline associated with VCI is often accompanied by subtle neuropsychiatric and behavioral changes, including apathy, irritability, anxiety, emotional dysregulation, and reduced social engagement. ¹³⁸ , ¹³⁹ Such behavioral alterations typically precede significant cognitive decline and may reflect vascular injury involving the subcortical prefrontal circuits. ¹³⁹ , ¹⁴⁰ In clinical assessment, several standardized tools are available for evaluating MBI, such as the Mild Behavioral Impairment Checklist (MBI‐C), which quantifies the severity of neuropsychiatric symptoms in at‐risk individuals and facilitates the early identification of prodromal behavioral changes related to vascular pathology. ¹³⁹ Accordingly, incorporating MBI assessment into the TIME framework enables the capture of non‐cognitive, neuropsychiatric features that are highly specific to vascular brain injury, thereby improving the sensitivity of early VCI identification.

4.5.4. Assessment of MBI

In addition to traditional gait analysis parameters, simple and clinically feasible metrics are widely adopted for gait assessment in VCI, such as the Timed 25 Foot Walk (T25FW). This test is straightforward to administer and time‐efficient, and by measuring short‐distance walking speed, it effectively reflects motor dysfunction associated with subcortical vascular injury. ¹⁴¹ Its simplicity and wide applicability render it an important tool for evaluating gait disturbance in both clinical evaluation and population‐based screening.

5. MULTIMODAL TIME PERSPECTIVE

Current research demonstrates a notable paradigm shift from single‐dimensional assessments toward multimodal integrated diagnostic approaches. Evidence from the COMPASS‐ND cohort has revealed that total cSVD burden assessed by MRI, along with its etiological subtypes – such as cerebral amyloid angiopathy‐related cSVD (CAA‐cSVD) and hypertensive arteriopathy‐related cSVD (HTNA‐cSVD) – constitute key neuroimaging indicators characterizing cognitive impairment. The study demonstrated that increased cSVD burden was significantly associated with lower MoCA scores (β = −0.56, p < 0.01), higher CDR‐SB scores (β = 0.20, p < 0.01), and reduced neuropsychological composite scores (β = −0.06, p < 0.01). ^{142 143} These findings further substantiate the idea that deep integration of neuroimaging phenotypes with multidimensional cognitive assessments constitutes a core strategy for developing highly sensitive identification systems and achieving precise clinical stratification in VCI.

The Singapore MCI cohort study employed unsupervised K‐means clustering to classify patients into two endophenotypes: hypoperfusion‐dominant (P+) and neuroinflammation‐dominant (I+). This investigation deeply integrated arterial spin labeling (ASL)‐MRI, peripheral blood biomarkers (glial fibrillary acidic protein, NfL, Aβ42/40, oligomerized amyloid beta), and the Visual Cognitive Assessment Test (VCAT), revealing distinct biological characteristics and cognitive profiles across subtypes. Mediation analysis further elucidated a pathological cascade wherein hypoperfusion promotes inflammatory development, leading to non‐neurodegenerative changes and ultimately resulting in cognitive impairment. ¹⁴⁴ This mechanism‐oriented integration strategy of “biomarker‐cognitive scale” provides critical evidence supporting precise subtyping and individualized intervention for VCI. Furthermore, as a large‐scale, multicenter cohort specifically targeting VCI, the MarkVCID consortium studies have established a highly reliable paradigm for early VCI identification through deep integration of refined cognitive assessments with cutting‐edge biomarkers. This research confirmed that the DTI‐based imaging marker PSMD was significantly negatively correlated with global cognitive function (β = −0.8, p < 0.001), with diagnostic value independent of WMHs. ³⁷ Simultaneously, it revealed the pathological mechanism whereby elevated peripheral blood PlGF levels drive cognitive impairment by mediating white matter interstitial fluid accumulation and increased free water. ¹⁴⁵ Integrating multimodal MRI (e.g., OEF), biospecimens, and cognitive assessments provides a robust and precise framework for the early screening, identification, and progression monitoring of VCI. ⁶² , ¹⁴⁶ , ¹⁴⁷

In summary, establishing a more comprehensive perspective for early identification may represent a future direction; however, numerous challenges currently persist. First, the complexity of pathological mechanisms precludes any single biomarker from fully capturing the trajectory of VCI, while substantial overlap between the mechanisms of various diseases and VCI pathogenesis further compounds the difficulty of etiological differentiation. Second, compared to the National Institute on Aging‐Alzheimer's Association biomarker framework for AD, VCI still lacks standardized biomarker criteria capable of systematically mapping onto its pathological processes. ¹⁴⁸

Based on the evidence provided by the aforementioned multicenter cohort studies, the application of machine learning in multimodal data fusion has demonstrated substantial potential; however, the hierarchical structure of its ensemble methods still requires clear conceptual definition. ¹⁴⁹ At present, the mainstream ensemble strategies can be summarized into three levels:

1. Data‐level integration, ¹⁵⁰ which directly fuses raw neuroimaging, behavioral time‐series signals, and humoral biomarkers;

2. Feature‐level integration, ¹⁵¹ which extracts high‐order features (e.g., PSMD, cerebral OEF, gait parameters) for joint representation;

3. Decision‐level integration, ¹⁵² which consolidates the outputs of prediction models independently constructed for each modality to achieve final classification or diagnostic decision‐making.

The TIME framework established in this review achieves cross‐modal information coordination and complementarity across the three aforementioned integration levels by introducing temporal dynamic analysis.

To address the significant pathological and clinical heterogeneity of VCI, ¹⁵³ , ¹⁵⁴ this study proposes a dual integrated diagnostic strategy with disease stage as the vertical axis and pathophysiological subtype as the horizontal axis, based on the longitudinal evolution of NVU injury. Given the insidious onset, prolonged disease course, and poor reversibility of VCI, ¹⁵⁵ in the preclinical stage without obvious cognitive complaints, diffusion tensor imaging metrics (which sensitively reflect the loss of brain microstructural integrity) and digital gait behavioral metrics (which capture subcortical circuit dysfunction, e.g., bradyphrenia and shuffling gait caused by chronic cSVD) ¹²⁷ together constitute the dominant modalities for early risk identification. Feature‐level integration is optimal at this stage to maximize the capture of subtle functional fluctuations that have not yet met the criteria for dementia, thereby enabling highly sensitive early screening. Integrating multimodal MRI, biospecimens, and cognitive assessments provides a robust and precise framework for the early screening, identification, and progression monitoring of VCI.

This strategy is highly consistent with cutting‐edge international research on VCI subtyping and the intervention guideline of “preventing disease before onset, and preventing progression after onset.” ¹⁵⁶ For instance, multimodal data clustering has identified two dominant endophenotypes: cerebral hypoperfusion and neuroinflammation activation. ¹⁵⁷ For cerebral hypoperfusion, the TIME framework can prioritize feature‐level fusion of imaging and behavioral metrics; for neuroinflammation activation, it can focus on decision‐level integration of biomarkers reflecting blood flow and neuronal injury, such as VEGF and neuron‐specific enolase.

Recent studies have further validated the superior performance of multimodal machine learning frameworks in this field. Using multimodal MRI features, one team constructed a deep learning model based on the AutoGluon platform, which achieved an AUC value of 0.926 to 0.878 in differentiating MCI due to cSVD from normal cognition, ¹⁵⁸ significantly outperforming traditional algorithms. Another team developed a transformer‐based deep learning model for cognitive impairment attributable to SVD, with model development and external validation conducted in 783 subjects; the model yielded an AUC of 0.841 in the training set and 0.859 and 0.749 in the validation sets, respectively, ¹⁵⁹ surpassing conventional machine learning models. These findings provide a methodological foundation for the clinical translation of the TIME framework (Table 4).

TABLE 4.

Multimodal integration strategy for VCI based on TIME framework.

Classification	Dominant assessment modalities	Recommended integration level
Preclinical stage	Neuroimaging, digital gait analysis	Feature‐level integration
Clinical stage (by subtype)	Neuropsychological scales, fluid biomarkers	Decision‐level integration
Hypoperfusion‐dominant subtype	Perfusion imaging, gait kinetic parameters	Feature‐level integration
Neuroinflammation‐dominant subtype	Inflammation‐related biomarkers, cognitive scales	Decision‐level integration

Furthermore, although clinical assessment scales such as the MoCA effectively discriminate severity levels in VCI and dementia, they demonstrate insufficient specificity and sensitivity for early diagnosis and remain limited in their ability to localize specific regional brain damage, thereby constraining their value in guiding subsequent clinical treatment. In summary, single diagnostic modalities are insufficient for achieving comprehensive and systematic early identification.

To facilitate clinical diagnosis, we propose a novel diagnostic framework that integrates distinct diagnostic perspectives to achieve enhanced early precision identification. TIME integrates the diagnostic advantages of Targeted cognitive assessments, Imaging biomarkers, Molecular fluid biomarkers, and Ecological behavioral characteristics. Unlike traditional fragmented and non‐standardized identification approaches, this framework enables more rapid and accurate detection and diagnosis of cognitive impairment at early stages. We also advocate for systematically incorporating TIME into machine learning frameworks to establish large‐scale data models capable of more comprehensive and systematic detection and identification of early cognitive impairment with enhanced specificity and sensitivity. As artificial intelligence technologies continue to mature within the medical domain, integrating multimodal diagnostic approaches into intelligent healthcare systems holds promise to become an important developmental direction for early identification and precise diagnosis of VCI (Figure 2).

FIGURE 2.

Pathophysiological cascade of vascular cognitive impairment and the multimodal TIME diagnostic framework. This conceptual framework synthesizes four complementary diagnostic dimensions: targeted cognitive assessments (e.g., MoCA, CDR), imaging biomarkers (microstructural, metabolic, and hemodynamic markers), molecular fluid biomarkers (brain‐derived extracellular vesicles, BBB integrity markers, iron metabolism indicators), and ecological behavioral characteristics (gait variability, keystroke dynamics, facial expressions). The TIME approach addresses VCI's inherent heterogeneity by capturing the disease continuum from preclinical vascular injury to overt cognitive decline. Rather than relying on any single modality with limited specificity, this framework leverages deep integration of multi‐source data via machine learning algorithms to construct a continuous, biologically anchored identification network. Such multimodal fusion enables mechanistic subtyping (e.g., hypoperfusion‐dominant vs neuroinflammation‐dominant endophenotypes), dynamic risk stratification, and, ultimately, personalized intervention strategies analogous to the AT(N) framework in AD.

6. CLINICAL TRANSLATION

6.1. Barriers to clinical translation of TIME framework and corresponding solutions

Although the TIME framework provides a promising multimodal approach for the early and precise identification of VCI, a substantial standardization gap remains in its translation from bench research to widespread clinical application. This lack of standardization directly undermines the comparability of findings across different research cohorts. To advance the clinical translation of the TIME framework, we must acknowledge and address technical barriers across the following three dimensions in future work.

First, variability across imaging vendors and scanning sequences compromises the quantitative comparability of imaging biomarkers (I). Although emerging microstructural and metabolic biomarkers such as PSMD and OEF have been preliminarily validated in multicenter cohorts including MarkVCID, their quantitative values are highly sensitive to scanning hardware. Differences in field strength (1.5T, 3T, and even 7T), gradient coil performance, and b‐value settings in DTI acquisition sequences across vendors can lead to significant deviations in water diffusion signals. ¹⁶⁰ Without rigorous cross‐platform data harmonization, imaging thresholds derived from single‐center studies cannot be directly extrapolated to other centers, severely limiting their generalizability.

Second, there is a lack of consensus on protocols for sample collection, processing, and storage for biomarkers (M). Taking the highly promising BDEVs as an example, methods for isolating BDEVs from brain tissue and blood have not been standardized, and heterogeneity in extraction and purification techniques severely hinders clinical translation. For instance, common procedures such as collagenase digestion may damage the structure and function of extracellular vesicle (EV) proteins. ¹⁶¹ Although polymer precipitation and immunoprecipitation (IP) methods have gained popularity for plasma BDEV enrichment, poor protocol consistency results in low comparability across studies. ¹⁶² In recent years, innovations including low‐speed centrifugation based on polyethylene glycol (PEG) precipitation combined with flow cytometry and nano‐flow cytometry requiring only minimal plasma volumes have markedly improved EV recovery and detection sensitivity. ¹⁶³ Despite these technological advances, protocols tailored to the specific pathological context of VCI remain fragmented, and no unified VCI‐specific operational guidelines have been established internationally. This technical heterogeneity renders direct cross‐laboratory comparisons of reported biomarker concentrations largely infeasible.

Third, sensor model and placement in behavioral measurements (E) severely interfere with data consistency. In behavioral assessments such as digital gait analysis, data accuracy is highly dependent on hardware parameters and acquisition environments. Inertial measurement units (IMUs) are used in 67% of gait studies, yet no uniform standards govern sensor placement, ¹⁶⁴ impairing the comparability and generalizability of results. Meanwhile, sensor placement directly yields distinct kinetic profiles for metrics including stride time variability and stride length, ¹⁶⁵ highlighting an urgent need for standardized testing guidelines. 

TABLE 5.

Feasibility‐efficacy balance analysis of assessment tools across TIME dimensions.

Dimension	Assessment tool / biomarker	Feasibility (time, cost, accessibility)	Diagnostic efficacy (sensitivity and specificity)	Recommended clinical scenarios	References
T (Targeted cognitive assessments)	MMSE	Short duration (∼6 min); simple to administer; highly accessible	Effective for moderate‐to‐severe dementia screening; good specificity	Large‐scale community screening; primary care settings	1 , 29 , 44
	MoCA	Moderate duration (∼10 to 15 min); requires professional guidance	High sensitivity; significantly superior to MMSE in detecting mild VCI	Memory clinics; precise specialist evaluations	27 , 40
I (Imaging biomarkers)	CT	Fast acquisition; widespread availability; low cost	Detects large infarcts, hemorrhages, and moderate to severe atrophy	Emergency screening; resource‐limited settings	75
	MRI (e.g., PSMD, OEF)	Time‐consuming; high cost; requires specialized equipment	Captures microstructural damage and metabolic compensation long before structural lesions appear	Etiological differentiation; precise early diagnosis; research cohorts	37 , 59
	OCTA (RPC)	Non‐invasive; rapid; no contrast agent required	Retinal microcirculation serves as a window to brain microvessels; shows potential for early detection	Longitudinal monitoring of high‐risk populations; aiding AD versus VCI differentiation	85
M (Molecular fluid biomarkers)	Plasma NfL	Convenient blood sampling; testing standardization is improving	Highly sensitive reflection of neuroaxonal injury; high prognostic value	Monitoring of disease activity and prognosis	116
	CSF sPDGFRβ	Invasive lumbar puncture; low patient acceptance	Very early specific indicator of pericyte injury and BBB disruption	Research settings; etiological investigation of complex cases	101 , 102
E (Ecological behavioral characteristics)	Digital gait analysis	Non‐invasive; allows continuous monitoring; easy to integrate via wearables	Decline in gait speed/variability often precedes objective cognitive decline	Routine physicals for community‐dwelling older adults; preclinical stage screening	126 , 127 , 130

To fundamentally resolve the aforementioned heterogeneity, this review recommends that the international research community jointly establish a VCI‐specific open database modeled after the Alzheimer's Disease Neuroimaging Initiative (ADNI). A core principle of this platform is that multicenter studies must submit raw unprocessed data rather than only summary statistics. Only through open sharing of raw data can researchers worldwide employ unified postprocessing pipelines and machine learning‐based feature extraction algorithms to achieve true normative harmonization across scanners and centers, thereby establishing generalizable diagnostic thresholds for the TIME framework across diverse clinical settings.

6.2. Clinical feasibility and stratified application strategy of TIME framework

The TIME framework, although developed from a multimodal perspective to help with early identification of VCI, is not intended to become a mandatory checklist that requires every patient to undergo all cognitive, imaging, molecular, and behavioral assessments. It is more like a flexible, tiered strategy that can be used depending on available resources, patient risk profile, and the specific diagnostic goal.

In clinical work, it can be applied in a step‐by‐step manner. For example, in community or primary care settings, first‑level screening should prioritize tools that are practical, low‑cost, and suitable for broad implementation. If a patient has vascular risk factors, a history of stroke, subjective cognitive complaints, slowed gait, or suspected cognitive decline, simple cognitive screening scales (such as MMSE or QDRS) can be used, along with measuring gait speed and routine vascular or metabolic laboratory tests. If neuroimaging is clinically indicated and accessible, non‑contrast CT can help identify large infarcts or structural lesions, although it is indeed not sensitive enough for early microvascular injury. The purpose of this level is not to make a definitive etiological diagnosis but to identify individuals who may need more detailed evaluation.

When patients reach memory clinics, neurology clinics, or regional specialty centers, the second‑level assessment can include more specific and mechanistically informative methods. For example, MoCA can be used to more sensitively detect early cognitive impairment, CDR can be used for functional staging, long‑term follow‑up, multi‑sequence MRI can be used to assess the burden of vascular lesions and microstructural damage, and, if feasible, certain specific blood or CSF biomarkers can be added. At this level, TIME mainly assists with etiological differentiation, risk stratification, and individualized follow‑up, rather than relying on any single test or a single cut‐off value.

In research settings, TIME can also serve as an organizing framework to integrate multimodal data. For example, cognitive test results, imaging indicators, fluid biomarkers, and daily behavioral signals can be analyzed through feature‑level or decision‑level fusion models to identify VCI‑related endophenotypes, predict disease progression, or select more suitable participants for clinical trials. This type of research application could help push the field toward a biologically based classification system – similar to what has been done in AD, while acknowledging that vascular pathology is much more heterogeneous.

Therefore, the practical value of TIME lies in its adaptability. During screening, it focuses on sensitivity, operability, and scalability; during specialist evaluation, it focuses on specificity, mechanistic interpretation, and long‑term monitoring; and in research, it provides a structure for integrating multimodal data and testing hypotheses. This tiered approach also shows that TIME complements existing clinical workflows rather than replacing them (Table 5).

7. SUMMARY AND OUTLOOK

In summary, VCI as a dynamically evolving process driven by complex vascular pathologies has shifted its early identification paradigm from unidimensional symptom description toward multimodal, cross‐scale mechanistic synthesis. The TIME (cogniTive scales, Imaging features, bioMarkers, bEhavioral characteristics) integrative framework proposed in this review systematically deconstructs highly sensitive cognitive scales, with advanced imaging techniques reflecting microstructural and metabolic disturbances (e.g., PSMD, OEF), liquid biopsy markers revealing BBB disruption and iron dyshomeostasis, and behavioral dynamic features capturing preclinical functional fluctuations. This framework illuminates the substantial potential of multi‐source heterogeneous data integration in overcoming the heterogeneity challenge inherent in VCI.

Although technical barriers persist – including the lack of unified pathological criteria, high inter‐center study heterogeneity, and difficulties in identifying mixed pathologies – the maturation of data from international prospective cohorts such as MarkVCID has catalyzed a growing consensus regarding the establishment of VCI‐specific biological definitions analogous to the AT(N) framework in the AD field. Importantly, the complementarity of these initiatives – with FINESSE providing a strategic blueprint for optimizing VCI clinical trials and MarkVCID2 delivering a standardized toolkit for biomarker validation – underscores that the field is advancing toward a unified mechanism‐anchored system for diagnosis and prognostication. Looking ahead, machine learning models driven by big data will enable deep decoupling and non‐linear fusion of multidimensional TIME data. This advancement will not only reshape our temporal understanding of the cascade of NVU injury but also foster deep integration between digital behavioral monitoring and precision medicine systems. Ultimately, this trajectory promises to provide revolutionary scientific paradigms and technological support for early intervention, precise risk stratification, and individualized dynamic diagnosis and treatment in VCI.

AUTHOR CONTRIBUTIONS

Conception and design: Zhiying Chen, Wenqi Song. Administrative support: Zhiying Chen, Yuanyuan Xiang. Provision of study materials: Moxin Wu, Xiaoping Yin. Collection and assembly of data: Moxin Wu, Xiaoping Yin. Data analysis and interpretation: Zhiying Chen. Manuscript writing: All authors. Final approval of manuscript: All authors. The final version was revised by Zhiying Chen.

CONFLICT OF INTEREST STATEMENT

The authors report no conflict of interest. Author disclosures are available in the Supporting Information.

CONSENT FOR PUBLICATION

Not applicable

AI DISCLOSURE STATEMENT

For the submitted manuscript, artificial intelligence (AI) tools were only applied for limited auxiliary work, and all core academic work was independently completed by the research team. Detailed usage breakdown is listed below:

The overall topic selection, logical framework, novel viewpoints, and content arrangement, as well as the original draft of this review, were all independently completed by the author team. AI tools only assist with three auxiliary tasks:

1. English translation of manuscript;

2. Grammar revision and academic polishing of sentences;

3. Preliminary screening of literature titles and abstract keywords.

All literature recommended by AI was manually retrieved and verified on PubMed by the authors. Relevant content was summarized and composed after careful reading of original publications; no AI‐generated text was directly copied into the manuscript.

Reference formatting was finished via EndNote during PubMed citation.

All figures were hand‐drawn independently by the authors with BioRender, and no AI was involved in any figure creation.

All authors bear full academic responsibility for the authenticity and integrity of all contents, data, and citations of this manuscript.

Supporting information

Supporting Information

DATA AVAILABILITY STATEMENT

The datasets used and/or analyzed during this study are available from the corresponding author on reasonable request.