Neuroimaging

TL;DR — MRI is the central in-vivo tool for vascular cognitive impairment because no single sequence captures infarction, small-vessel disease, haemorrhagic markers, and atrophy (Duering 2023, PMID 37236211). A minimum useful protocol combines structural T1, T2/FLAIR, diffusion, and susceptibility-sensitive imaging; vascular and perfusion sequences answer targeted questions (Razek 2021, PMID 33434866). CT remains valuable for acute haemorrhage and gross infarction but underdetects small and haemorrhagic lesions. Imaging demonstrates injury and patterns; it does not by itself prove that lesions caused the cognitive syndrome.

Sequence map

Sequence	Primary information	Important limitation
T1 3D	anatomy, atrophy, segmentation	atrophy nonspecific
T2/FLAIR	WMH, oedema, lacunes	WMH etiology uncertain
DWI/ADC	recent infarction	transient/chronic lesions differ
T2*/SWI	microbleeds, siderosis, blood products	field/sequence affects counts
TOF/contrast MRA	stenosis/occlusion	flow artifacts
ASL/perfusion	regional flow	low flow may be cause or consequence
DTI	microstructural/network injury	harmonization and model issues
DCE	BBB permeability	contrast/model dependence

Structural markers

STRIVE and STRIVE-2 standardize recent small subcortical infarcts, lacunes, WMH, perivascular spaces, microbleeds, superficial siderosis, and atrophy (Wardlaw 2013, PMID 23867200; Duering 2023, PMID 37236211).

Marker	Report
Infarct	age, volume, territory, strategic network relation
WMH	distribution, severity, progression
Lacunes	number and location
Microbleeds	count and lobar/deep/mixed distribution
Siderosis	focal/disseminated
PVS	location and severity
Atrophy	global and regional pattern

How reproducible is a visual rating?

The measurement most often reported in vascular-dementia literature — a WMH grade from a visual scale — has a reproducibility floor that is rarely quoted alongside it. When 395 post-stroke MRIs were re-scored under 13 different published rating scales, agreement between scales ranged from more than 80% of patients receiving an equivalent grade at best to 0.4% for periventricular and 18% for deep WMH at worst; some scales showed floor effects and others ceiling effects with age, and only 1 of 7 periventricular, 5 of 9 deep and 1 of 3 combined scales correlated significantly with hypertension — a risk factor all of them should have detected (Mäntylä 1997, PMID 9259759). Inter-observer reliability is not a fixed property of a scale either: across 494 subjects in five cohorts rated with seven scales, weighted κ for Fazekas ran 0.89 and 0.72 in the cohort with the most WMH but 0.20 and 0.24 in the cohort with the least, and the authors argue the high values partly reflect a ceiling effect rather than genuinely better agreement (Wardlaw 2004, PMID 15164192). Two consequences follow for this condition: apparent between-study disagreement about WMH–cognition associations is partly instrument variance, and reliability estimates borrowed from a heavily affected cohort do not transfer to an early-disease or prevention trial.

The pessimistic reading above needs one qualification, which sharpens rather than softens it. Three raters of differing experience applying the Manolio, Fazekas–Schmidt and Scheltens scales to 74 baseline and follow-up scans from five European centres achieved fair-to-good interrater agreement at baseline (κ 0.59–0.78) — with Fazekas–Schmidt significantly better than Manolio (P=0.003) — and all three scales correlated highly with each other (Spearman 0.712–0.806) and with volumetric measurement (Kendall W 0.37–0.57, all P<0.001). But on follow-up scans, direct interrater agreement for detecting progression collapsed to κ 0.19–0.39, even though all three raters independently found significant progression (Kapeller 2003, PMID 12574557). Cross-sectional visual rating is therefore adequate and progression rating is not — which is exactly the wrong way round for a field that has concluded (see small-vessel disease) that WMH change carries roughly three times the cognitive effect of WMH presence. Any progression endpoint needs volumetry or automated segmentation, not a visual scale.

CT is less hopeless for this purpose than usually assumed. In 70 IST-3 patients scanned with both modalities, intrarater CT–MRI agreement was weighted κ 0.55–0.75 for periventricular WMH components and 0.44–0.70 for atrophy scales, falling to κ 0.18–0.44 for basal-ganglia ratings, largely because prominent perivascular spaces were misclassified on CT; MRI remained more sensitive for small and cavitated infarcts (Ferguson 2018, PMID 29576397).

CT

CT is fast and widely available, making it indispensable in acute stroke/haemorrhage. In chronic cognitive evaluation it detects large infarcts, established lacunes, confluent WMH, and atrophy but is less sensitive than MRI for small infarcts, microbleeds, siderosis, and subtle regional patterns (Razek 2021, PMID 33434866). A normal CT does not exclude VCI.

Medial temporal atrophy: a rating that beats its own volumetry

The atrophy marker used to argue against a vascular attribution has a comparative validation worth quoting, because it inverts the usual assumption that quantification beats eyeballing. In 143 unselected memory-clinic patients (41 Alzheimer disease, 36 other dementias including vascular dementia, 66 non-demented), visual 0–4 MTA rating and stereological volumetry both separated Alzheimer disease from non-demented subjects and other dementias from non-demented subjects. Combining MMSE with visual MTA gave sensitivity 95% for Alzheimer disease and 85% for other dementias at specificity 96%, and volumetry added nothing over MMSE alone (Wahlund 2000, PMID 11032615). Two cautions for use in this condition: the "other dementias" group pooled vascular, frontotemporal and unspecified cases, so the 85% figure is not a vascular-dementia sensitivity; and MTA at these thresholds separates dementia from non-dementia far better than it separates one dementia from another.

Pattern attribution

Pattern	Supports	Does not prove
multiple territorial infarcts	cumulative embolic/large-vessel injury	dementia causation
deep WMH/lacunes/microbleeds	hypertensive arteriopathy	absence of Alzheimer disease
strictly lobar microbleeds/siderosis	CAA	parenchymal Alzheimer stage
temporal-pole/external-capsule WMH	CADASIL suspicion	pathogenic NOTCH3 variant
medial temporal atrophy	Alzheimer-pattern degeneration	pure Alzheimer etiology
watershed infarcts	hypoperfusion episode	chronic low-flow mechanism

Cortical cerebral microinfarcts

Cortical cerebral microinfarcts are the STRIVE-2 marker least likely to appear in a clinical report, because detecting them requires deliberate looking. In 135 patients imaged within 3 months of spontaneous intracerebral haemorrhage, 3T MRI found 100 cortical microinfarcts in 57 patients (42%), while 7T MRI in the 40-patient subset found 59 lesions in 28 patients (70%) — the field-strength gap is the detection problem stated numerically. Frequency did not differ between lobar and non-lobar ICH (41% vs 43% at 3T), but depth within the cortex did: superficial-layer microinfarcts occurred in 30% of lobar versus 5% of non-lobar cases (RR 2.7, 95% CI 1.5–5.0), suggesting laminar position may index the underlying arteriopathy where lesion count does not. Presence was associated with prior TIA/ischaemic stroke (RR 2.7, 95% CI 1.1–6.4) and with none of the conventional SVD markers (Jolink 2025, PMID 40537072). A marker uncorrelated with the rest of the SVD panel is either measuring something the panel misses or measuring noise; the longitudinal data favour the former.

In 475 memory-clinic patients followed at least twice over 5 years, cortical microinfarcts predicted accelerated volume loss at 2 years (total brain β = −1.94, 95% CI −3.07 to −0.82; grey matter β = −1.00, −1.69 to −0.30; white matter β = −0.95, −1.54 to −0.35, all P-interaction ≤0.002) and interacted with atrophy on cognition: at year 5, patients with high volume loss plus a single microinfarct scored β = −1.83 (−2.68 to −0.97) lower on global cognition and those with multiple microinfarcts β = −3.13 (−4.21 to −2.05), with the steepest trajectory in the multiple-microinfarct group (Huang 2025, PMID 40796328). The effect was strongest in executive function, memory, language and visuospatial domains — i.e. not confined to the executive profile usually attributed to vascular injury.

CAA imaging

Boston v2.0 probable CAA combines haemorrhagic and selected white-matter features. In autopsy-standard cases, sensitivity was 74.5% (65.4–82.4) and specificity 95.0% (83.1–99.4); derivation, temporal, and geographic cohorts ranged 74.8–92.5% sensitivity and 81.5–89.5% specificity (Charidimou 2022, PMID 35841910). In 54 non-haemorrhagic MRI–pathology cases, probable CAA sensitivity fell to 28.6% (13.2–48.7) with specificity 65.3% (44.3–82.8) (Switzer 2024, PMID 38710005). Performance depends on clinical selection and adequate susceptibility imaging.

Where MRI is unavailable, the Edinburgh CT-only and CT-APOE criteria carry prognostic information rather than diagnostic certainty: across 1,620 patients with lobar ICH from eight cohorts, cumulative 5-year recurrent-ICH incidence ran 12% (low risk), 16% (intermediate; adjusted subdistribution HR 1.68, 95% CI 1.21–2.32), and 26% (high risk; aSHR 2.97, 95% CI 1.50–5.89) (Rodrigues 2025, PMID 40975099). CT can therefore stratify risk even when it cannot characterize the arteriopathy — see CAA.

Quantitative imaging

Method	Candidate value	Validation need
WMH volume	continuous burden/progression	cross-scanner robustness
Lacune/microbleed detection	scalable lesion counts	false-positive review
Brain volume	global/regional loss	head-size and pipeline harmonization
Peak-width skeletonized diffusivity	diffuse white-matter injury	external thresholds
Connectomics	network disconnection	clinical interpretability
ASL	perfusion phenotype	repeatability
retinal OCT-A	accessible microvascular correlate	brain specificity

PSMD: the marker closest to being usable

Peak-width of skeletonized mean diffusivity is the DTI summary furthest along toward trial use because it is fully automated, takes minutes, needs no manual segmentation, and reduces the whole white-matter skeleton to a single number — the 95th minus 5th percentile of mean diffusivity — which sidesteps the boundary-definition problem that limits WMH segmentation (Zanon Zotin 2023, PMID 36692402). The same review is explicit about what is not yet established: PSMD is not specific to small-vessel disease, is sensitive to any diffuse white-matter process including neurodegeneration and demyelination, and has no external cut-point.

Its mechanistic position is as a mediator rather than a competitor to the visible markers. In 273 community-dwelling older adults stratified by SVD burden, PSMD fully mediated the association between enlarged perivascular spaces and both MMSE and sorting-test performance, and partially mediated the associations for total SVD burden, WMH volume, lacunes and microbleeds; a chain model placed hypertension and coronary disease upstream, acting on cognition through SVD and then through PSMD (Hu 2025, PMID 40585818). If that ordering is right, PSMD measures the tissue consequence that the lesion counts only proxy — which is the argument for using it as an endpoint, and equally the reason it cannot be used to attribute a cause.

Systematic review of computer-aided MRI-marker extraction found 70 classical image-processing, machine-learning, and deep-learning studies for microbleeds, perivascular spaces, and lacunes, but no generalisable pipeline validated across research and clinical cohorts (Jiang 2022, PMID 35914668). Machine-learning discrimination in curated datasets is not equivalent to prospective diagnostic utility.

How good is automated WMH segmentation, actually?

The largest test to date trained 2D UNet and squeeze-and-excitation UNet models on 2,408 FLAIR scans from three Korean hospitals and validated them on 6,013 scans from six others — 8,421 patients with acute ischaemic stroke, the setting where stroke lesions most obscure WMH borders.

Model	Internal Dice (95% CI)	External Dice (95% CI)	Volume correlation
UNet	0.659 (0.649–0.669)	0.710 (0.707–0.714)	r = 0.917
SE-UNet	0.675 (0.666–0.685)	0.722 (0.719–0.726)	r = 0.933
Human–human reliability	0.744 (0.738–0.751)	—	—

Automated segmentation remained significantly below inter-rater human agreement (P=0.031 for UNet vs human), yet automated and manual WMH volumes correlated at r≈0.93 with concordance correlation coefficients of 0.841–0.956 across external datasets (Kim 2024, PMID 39013565). The two facts are compatible and important: voxel-level agreement is mediocre while volume-level agreement is excellent, so automated pipelines are adequate for the burden measure most studies actually use and inadequate for lesion-boundary questions. The study also produced a per-patient uncertainty index from Kullback–Leibler divergence; in 86% of external cases it fell below 0.35, and those cases averaged Dice 0.744 — human-equivalent. Flagging the uncertain 14% for human review is the most practical quality-control mechanism yet demonstrated for large-scale WMH work.

Machine learning on vascular markers: performance is good, method reporting is not

A systematic review and meta-analysis of 75 studies (43 diagnosis, 27 prognosis, 5 both) using SVD neuroimaging markers in machine-learning models found strong pooled discrimination — AUC 0.88 (95% CI 0.85–0.92) separating healthy controls from Alzheimer dementia and AUC 0.84 (95% CI 0.74–0.95) for cognitive impairment — with nearly 60% of all studies published in the preceding two years (Lohner 2025, PMID 40775365). The methodological picture is worse than the performance picture: only 16 of 75 studies could be meta-analysed because of inconsistent reporting, only five tested generalisability on an external dataset, and six lacked clear diagnostic criteria. An AUC of 0.88 against healthy controls is also the easiest possible comparison; the clinically relevant question is discrimination among people already in a memory or stroke clinic, which almost none of these studies addressed.

SPRINT MRI (n=449 with paired scans) showed intensive SBP targeting reduced WMH growth by 0.54 cm³ (95% CI 0.20–0.87) over ~4 years versus standard targeting, with a 3.7 cm³ greater brain-volume decrease (SPRINT MIND Investigators 2019, PMID 31408137). PRESERVE (n=82) found intensive BP improved DTI network efficiency without changing conventional WMH/volume metrics (Pflanz 2022, PMID 35977831). These trials show that imaging can move with treatment; they do not yet prove that the imaging change is a cognitive surrogate.

Perfusion, oxygenation and vessel function

Structural markers record damage already done; the functional sequences attempt to image the process. Three are far enough along to quote numbers.

Cerebrovascular reactivity (CVR). In 182 patients with lacunar or cortical ischaemic stroke, CVR measured with 3T MRI under CO₂ challenge fell with every SVD feature: per 10-fold increase in WMH volume (as % intracranial volume) B = −0.0073 %/mm Hg (95% CI −0.0133 to −0.0014) in normal-appearing white matter, and per additional lacune −0.00129 (−0.00215 to −0.00043), per microbleed −0.00083 (−0.00130 to −0.00036), per basal-ganglia perivascular-space point −0.0034 (−0.0066 to −0.0002), per total SVD-score point −0.0048 (−0.0075 to −0.0021). The association with MoCA was in the expected direction but did not reach significance (B = 0.00065, 95% CI −0.00007 to 0.00137) — a reminder that a mechanistic marker tracking lesion burden is not thereby a cognitive marker (Sleight 2023, PMID 37814956).

Are the vascular dysfunctions one thing or several? INVESTIGATE-SVDs measured blood–brain barrier permeability, blood plasma volume, pulsatility and CVR concurrently in 77 patients (45 sporadic SVD, 32 CADASIL) at three sites — the first study to do all of them in the same people. Worse WMH burden predicted lower CVR (B = −1.78, 95% CI −3.30 to −0.27) and lower plasma volume fraction (B = −0.594, −0.987 to −0.202), and CVR was worse inside WMH than in normal-appearing white matter (B = −0.048, −0.079 to −0.017). Two negative results carry as much weight: after adjusting for WMH severity, CADASIL and sporadic SVD did not differ (CVR B = 0.0169, 95% CI −0.0247 to 0.0584), and the vascular functions were not interrelated with each other (permeability~CVR B = −0.85, −4.72 to 3.02) (Stringer 2025, PMID 39552538). "Endothelial dysfunction" is therefore not a single measurable quantity, and a trial that moves one of these functions should not be assumed to have moved the others.

Perfusion heterogeneity as a candidate marker. In 368 memory-clinic patients with three annual neuropsychological assessments and MRI, the spatial coefficient of variation of grey-matter ASL signal — proposed as a proxy for cerebrovascular insufficiency rather than for flow itself — predicted 3-year decline in the memory domain (specifically in participants without dementia), WMH progression, incident microbleeds, and incident vascular events (29 events, 7.8%, over a mean 3 years) (Gyanwali 2022, PMID 35871110). The study supports spatial heterogeneity as a candidate physiological measure but does not establish superiority over mean perfusion.

Perforating-artery pulsatility at 7T. Phase-contrast 7T MRI of the lenticulostriate arteries and middle cerebral artery in 28 patients with essential hypertension and 25 age- and sex-matched controls (mean age 63.4) found a higher pulsatility index in the lenticulostriate arteries and a lower damping factor in hypertensives (both P=0.015), with no difference in the middle cerebral artery; higher systolic and mean arterial pressure tracked higher LSA pulsatility, and adjusting for small-vessel-disease score did not alter the relationships (van den Kerkhof 2023, PMID 36722349). Persistence after adjustment for visible SVD is consistent with, but does not prove, an upstream exposure. The discordant lenticulostriate and middle-cerebral-artery results also show that the measures are not interchangeable.

Oxygen extraction. Perfusion-only models treat hypoperfusion as the injury. In 195 patients with SVD and 178 controls, cerebral blood flow was reduced throughout the white matter as expected, but oxygen extraction fraction — from quantitative susceptibility mapping plus qBOLD — was increased in normal-appearing white matter and decreased inside WMH, and across burden strata rose in mild-to-moderate disease before falling in the most severe. Adding OEF to CBF improved prediction of WMH and free-water progression over a mean 2.6 years, and higher baseline CBF and OEF both predicted slower free-water accumulation (Zhang 2025, PMID 40323889). The increase-then-decrease pattern reframes chronic hypoperfusion as a compensated state until compensation fails, which is a different trial target from flow augmentation alone.

The wider advanced-MRI menu — quantitative iron and myelin imaging, DCE-derived permeability, and 7T imaging of perforating-artery flow velocity and pulsatility — is reviewed with its trial applications and its harmonization obstacles in van den Brink 2023 (PMID 35311609).

Amyloid PET and mixed disease

Amyloid PET can establish cerebral amyloid burden but does not by itself separate vascular CAA from parenchymal plaque contribution or quantify vascular causality. Its most useful role in VCI research may be stratifying mixed pathology. Alzheimer-specific PET interpretation belongs in the companion Alzheimer's disease.

The base rates matter for anyone reading a scan in a patient labelled vascular dementia. In an individual-participant meta-analysis of 1,359 clinically diagnosed Alzheimer and 538 non-Alzheimer dementia cases, amyloid positivity in vascular dementia rose with age and APOE ε4: 7% (95% CI 3–18) at 60 years to 29% (17–43) at 80 in ε4 non-carriers (n=77), and 25% (9–52) to 64% (49–77) over the same range in carriers (n=30) (Ossenkoppele 2015, PMID 25988463). A Chinese memory-clinic series of 1,193 patients scanned with ¹¹C-PIB or ¹⁸F-AV45 found 6 of 29 vascular dementia cases amyloid-positive (20.7%), against 86.8% of Alzheimer and 5.6% of frontotemporal cases (Shi 2020, PMID 32385166). A positive amyloid scan in an 80-year-old APOE ε4 carrier with a vascular syndrome is therefore close to uninformative; a negative scan is the more useful result, and its value falls as age rises.

FDG-PET is weaker still for this specific boundary. The EANM–EAN Delphi exercise graded the empirical evidence for FDG-PET as good for separating dementia with Lewy bodies from Alzheimer disease, fair for frontotemporal lobar degeneration, and lacking for Alzheimer disease versus vascular dementia — the panel nonetheless supported its use in all scenarios on negative-predictive-value grounds, which is a consensus recommendation resting explicitly on absent evidence (Nestor 2018, PMID 29736698).

Reporting template

• Clinical question and interval from stroke.
• Scanner/field strength and sequences.
• Acute and chronic infarcts.
• SVD markers using STRIVE terms.
• Haemorrhagic distribution.
• Atrophy pattern.
• Large-vessel findings where imaged.
• Change from prior study.
• Explicit uncertainty and relevant alternatives.

Open questions

• If interrater agreement for WMH progression is κ 0.19–0.39 while cross-sectional agreement is κ 0.59–0.78, should visual scales be barred from progression endpoints outright? (Kapeller 2003, PMID 12574557)
• Is spatial variability of perfusion (ASL sCoV) a better vascular-cognitive marker than mean cerebral blood flow, and does it replicate outside memory clinics? (Gyanwali 2022, PMID 35871110)
• If hypertension raises pulsatility in lenticulostriate but not middle cerebral arteries, are middle-cerebral pulsatility indices measuring the relevant exposure at all? (van den Kerkhof 2023, PMID 36722349)
• Does visual medial-temporal-atrophy rating retain its accuracy advantage over volumetry when the comparison is vascular versus Alzheimer dementia rather than dementia versus no dementia? (Wahlund 2000, PMID 11032615)
• Which MRI progression measure is a valid surrogate endpoint? (Duering 2023, PMID 37236211)
• Can automated tools generalize across vendors and populations? (Jiang 2022, PMID 35914668)
• What imaging combination best apportions vascular and Alzheimer contributions? (Frantellizzi 2020, PMID 31929166)
• Can low-cost CT-based scores support equitable VCI diagnosis where MRI is unavailable? (Razek 2021, PMID 33434866)
• Is voxel-level Dice the right benchmark for automated WMH segmentation when volume-level agreement already reaches r≈0.93? (Kim 2024, PMID 39013565)
• Should per-scan uncertainty indices become a required output of automated pipelines, with human review of flagged cases? (Kim 2024, PMID 39013565)
• Do machine-learning models built on vascular markers retain AUC ≈ 0.85 when the comparison is clinic differentials rather than healthy controls? (Lohner 2025, PMID 40775365)
• Should WMH visual grades be reported with the scale named, given that between-scale concordance falls to 0.4% for periventricular ratings? (Mäntylä 1997, PMID 9259759)
• Are reliability estimates derived in heavily affected cohorts (Fazekas κ 0.89) valid for prevention trials in early disease, where κ fell to 0.20? (Wardlaw 2004, PMID 15164192)
• Do cortical microinfarcts index a mechanism the conventional SVD panel misses, given that they correlate with none of its markers? (Jolink 2025, PMID 40537072)
• Does the increase-then-decrease pattern of oxygen extraction identify a window in which flow-augmenting therapy could still work? (Zhang 2025, PMID 40323889)
• If permeability, pulsatility and reactivity are not interrelated, which one should an early-phase trial move? (Stringer 2025, PMID 39552538)
• Is PSMD an endpoint or a mediator — and can it be both without becoming uninterpretable as a cause? (Hu 2025, PMID 40585818; Zanon Zotin 2023, PMID 36692402)

Related pages

• Cerebral small-vessel disease — marker biology.
• CAA — haemorrhagic pattern.
• Biomarkers — multimodal measures.
• Nosology — causal attribution.

References

1. Duering M, et al. STRIVE-2. Lancet Neurol. 2023. PMID 37236211
2. Wardlaw JM, et al. STRIVE. Lancet Neurol. 2013. PMID 23867200
3. Razek AAKA, et al. Imaging of vascular cognitive impairment. Clin Imaging. 2021. PMID 33434866
4. Charidimou A, et al. Boston criteria v2.0 for CAA. Lancet Neurol. 2022. PMID 35841910
5. Jiang J, et al. Computer-aided extraction of select MRI markers of cerebral small vessel disease. Neuroimage. 2022. PMID 35914668
6. Frantellizzi V, et al. Neuroimaging in VCI and dementia. J Alzheimers Dis. 2020. PMID 31929166
7. Switzer AR, et al. Boston Criteria v2.0 for CAA without hemorrhage. Neurology. 2024;102:e209386. PMID 38710005
8. SPRINT MIND Investigators. Intensive vs standard BP control and cerebral white matter lesions. JAMA. 2019;322:524-534. PMID 31408137
9. Pflanz CP, et al. BP-lowering intensity and white-matter network integrity. Neurology. 2022;99:e1945-e1953. PMID 35977831
10. Kim H, et al. Automated segmentation of MRI white matter hyperintensities in 8421 patients with acute ischemic stroke. AJNR Am J Neuroradiol. 2024;45:1885-1894. PMID 39013565
11. Lohner V, et al. Machine learning applications in vascular neuroimaging for the diagnosis and prognosis of cognitive impairment and dementia: a systematic review and meta-analysis. Alzheimers Res Ther. 2025;17:183. PMID 40775365
12. Rodrigues MA, et al. Association between the Edinburgh CT and genetic diagnostic criteria for cerebral amyloid angiopathy-associated lobar intracerebral haemorrhage and recurrent intracerebral haemorrhage. Lancet Neurol. 2025;24:828-839. PMID 40975099
13. Mäntylä R, et al. Variable agreement between visual rating scales for white matter hyperintensities on MRI. Comparison of 13 rating scales in a poststroke cohort. Stroke. 1997;28:1614-23. PMID 9259759
14. Wardlaw JM, et al. White matter hyperintensities and rating scales—observer reliability varies with lesion load. J Neurol. 2004;251:584-90. PMID 15164192
15. Ferguson KJ, et al. Visual rating scales of white matter hyperintensities and atrophy: comparison of computed tomography and magnetic resonance imaging. J Stroke Cerebrovasc Dis. 2018;27:1815-1821. PMID 29576397
16. Jolink WMT, et al. Cortical cerebral microinfarcts in spontaneous intracerebral hemorrhage on 3T and 7T MRI. Neurology. 2025;105:e213642. PMID 40537072
17. Huang J, et al. Synergistic effect between cortical cerebral microinfarcts and brain atrophy on cognitive decline. Brain. 2025;148:3924-3934. PMID 40796328
18. Zanon Zotin MC, et al. Peak width of skeletonized mean diffusivity: a neuroimaging marker for white matter injury. Radiology. 2023;306:e212780. PMID 36692402
19. Hu Q, et al. Peak width of skeletonized mean diffusivity mediates the relationship between cerebral small vessel disease burden and cognitive impairment in community-dwelling older adults. Brain Commun. 2025;7:fcaf233. PMID 40585818
20. Sleight E, et al. Cerebrovascular reactivity in patients with small vessel disease: a cross-sectional study. Stroke. 2023;54:2776-2784. PMID 37814956
21. Stringer MS, et al. Cerebrovascular function in sporadic and genetic cerebral small vessel disease. Ann Neurol. 2025;97:483-498. PMID 39552538
22. Zhang R, et al. Oxygen extraction fraction in small vessel disease: relationship to disease burden and progression. Brain. 2025;148:1950-1962. PMID 40323889
23. van den Brink H, et al. Advanced MRI in cerebral small vessel disease. Int J Stroke. 2023;18:28-35. PMID 35311609
24. Ossenkoppele R, et al. Prevalence of amyloid PET positivity in dementia syndromes: a meta-analysis. JAMA. 2015;313:1939-49. PMID 25988463
25. Shi Z, et al. Amyloid PET in dementia syndromes: a Chinese multicenter study. J Nucl Med. 2020;61:1814-1819. PMID 32385166
26. Nestor PJ, et al. Clinical utility of FDG-PET for the differential diagnosis among the main forms of dementia. Eur J Nucl Med Mol Imaging. 2018;45:1509-1525. PMID 29736698
27. Kapeller P, et al. Visual rating of age-related white matter changes on magnetic resonance imaging: scale comparison, interrater agreement, and correlations with quantitative measurements. Stroke. 2003;34:441-5. PMID 12574557
28. Wahlund LO, et al. Visual rating and volumetry of the medial temporal lobe on magnetic resonance imaging in dementia: a comparative study. J Neurol Neurosurg Psychiatry. 2000;69:630-5. PMID 11032615
29. Gyanwali B, et al. Association of arterial spin labeling parameters with cognitive decline, vascular events, and mortality in a memory-clinic sample. Am J Geriatr Psychiatry. 2022;30:1298-1309. PMID 35871110
30. van den Kerkhof M, et al. Hypertension correlates with stronger blood flow pulsatility in small perforating cerebral arteries assessed with 7 Tesla magnetic resonance imaging. Hypertension. 2023;80:802-810. PMID 36722349