Vessel Wall MR Imaging Biomarkers of Symptomatic Intracranial Atherosclerosis: A Meta-Analysis

Abstract

Background and Purpose:

Intracranial atherosclerotic disease is a common cause of stroke worldwide. Intracranial vessel wall MR imaging (VWI) may be able to identify imaging biomarkers of symptomatic plaque. We performed a meta-analysis to evaluate the strength of association of imaging features of symptomatic plaque leading to downstream ischemic events. Effects on the strength of association were also assessed accounting for possible sources of bias and variability related to study design and MR parameters.

Methods:

PubMed, Scopus, Web of Science, EMBASE and Cochrane databases were searched up to October 2019. Two independent reviewers extracted data on study design, VWI techniques, and imaging endpoints. Per-lesion odds ratios (OR) were calculated and pooled using a bivariate random-effects model. Subgroup analyses, sensitivity analysis and evaluation of publication bias were also performed.

Results:

Twenty-one articles met inclusion criteria (1,750 lesions; 1,542 subjects). Plaque enhancement (OR 7.42, 95% CI 3.35–16.43), positive remodeling (OR 5.60, 95% CI 2.23–14.03), T1 hyperintensity (OR 2.05, 95% CI 1.27–3.32) and surface irregularity (OR 4.50, 95% CI 1.39–8.57) were significantly associated with downstream ischemic events. T2 signal intensity was not significant (p=0.59). Plaque enhancement was significantly associated with downstream ischemic events in all subgroup analyses and showed stronger associations when measured in retrospectively designed studies (p=0.02), by a radiologist as a rater (p<0.001), and on lower VWI spatial resolution sequences (p=0.02).

Conclusions:

Plaque enhancement, positive remodeling, T1 hyperintensity and surface irregularity emerged as strong imaging biomarkers of symptomatic plaque in patients with ischemic events. Plaque enhancement remained significant accounting for sources of bias and variability in both study design and instrument. Future studies evaluating plaque enhancement as a predictive marker for stroke recurrence with larger sample sizes would be valuable.

Introduction

Intracranial atherosclerotic disease (ICAD) is one of the most common causes of stroke worldwide with an estimated prevalence of 20–40 persons per 100,000 population.¹ In high risk groups, ICAD accounts for up to 10% of transient ischemic attacks (TIAs), 30–50% of ischemic strokes, and has an annual risk of 20% of recurrent stroke.^2,3 It is most prevalent in Asian, Hispanic and African-American populations, highlighting a cardiovascular disease that needs attention to reduce health disparities in prevention and treatment.

The high incidence and burden of ICAD led to a consensus conference in 2009 dedicated to identifying principles of management and research priorities for ICAD.⁴ Recognized research priorities included identifying new vascular imaging modalities to identify vulnerable plaques and identification of surrogate endpoints to validate plaque stability and regression to direct treatment strategies.⁴ Over a decade later, these research priorities remain consequential yet unresolved.

Intracranial vessel wall MR imaging (VWI) has become recognized as a potential noninvasive vascular imaging modality for studying intracranial plaque.^5,6 VWI can image beyond the lumen to visualize specific plaque features, which may be surrogate imaging endpoints for future clinical trials. As many of the studies using VWI for ICAD comprise of small sample sizes and are underpowered, we performed this meta-analysis to quantitatively pool data and evaluate the strengths of association of commonly investigated imaging features of symptomatic plaque. We also examined the robustness of these potential imaging biomarkers of plaque accounting for different study designs. Studies with methodological shortcomings can overestimate diagnostic accuracy measures^7–9 and may be reasons for failure of bench to bedside translation of stroke therapies.¹⁰ Moreover, as there is widespread technical variability among VWI protocols used in the literature,¹¹ instrument variability and methods of image acquisition may also be sources of bias. Finally, we also investigated whether these potential imaging biomarkers were consistent between patients with stroke versus TIA, as identifying patients with high risk TIA would be clinically useful.

Methods

Search strategy and study selection

The data that support the findings of this study are available from the corresponding author upon reasonable request. A systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. PubMed, Scopus, EMBASE, Web of Science and Cochrane databases were searched up to October 31, 2019 with no language restrictions. Supplemental Table I shows the search strategy. A manual review of the citations and other reviews was also performed. Two reviewers independently screened studies with discrepancies resolved by consensus. Inclusion criteria were (a) evaluation of ICAD associated with acute or subacute ischemic events; (b) humans; and (c) VWI on 1.5 or 3 Tesla. Exclusion criteria were (a) insufficient exclusion of other ischemic stroke etiologies; (b) < 10 subjects/plaques or insufficient raw data; (c) conference abstracts; (d) histology studies; and (e) subject overlap with same imaging endpoints. If plaque quadrant was the only reported imaging endpoint, this was considered insufficient data as pooling this imaging endpoint would not be generalizable across all intracranial arteries. Corresponding authors were contacted for more data as needed.

Data extraction

Two reviewers (JWS, AP) independently extracted data with discrepancies resolved by consensus. Data were collected about study design (e.g., publication year, subject inclusion criteria, prospective or retrospective enrollment, demographic data, prevalence of stroke risk factors), MR technology (e.g., magnet strength, coil, vessel wall MR protocols and pulse sequence parameters), and image analysis (e.g., rater characteristics, imaging endpoint criteria, measurement technique, assessed artery and results). Demographic and imaging endpoint data were extracted on a per-patient and per-lesion basis, respectively. Methods are further expanded upon in Supplemental Methods.

Risk of bias was independently assessed by 2 reviewers (JWS, AP) using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool with discrepancies resolved by consensus and results tabulated for graphical presentation.¹²

Statistical Analysis

Categorical variables are expressed in counts and percentages; continuous variables are expressed in means, weighted means, or medians. Agreement was calculated with an unweighted Cohen’s κ. Odds ratios (ORs) with corresponding 95% confidence intervals (CIs) were calculated for a per-lesion analysis to determine the association between the ischemic event and VWI plaque feature. Pooled ORs were estimated with a bivariate random-effects model using the DerSimonian-Laird method based on the assumption that different, yet related, effects were estimated between-studies.¹³ Forest plots were generated if there were at least 3 studies that could be quantitatively pooled. Data from studies that compared different types of infarcts due to ICAD (e.g., large vessel occlusion, branch atheromatous disease) were combined as one arm.^14–16 A continuity correction of 0.5 was applied for studies with one arm. SPSS (v19 IBM, Chicago) and STATA 16 (StataCorp, College Station, TX) were used for statistical analyses.

Details for subgroup analysis, sensitivity analysis, and publication bias are described in Supplemental Methods.

Results

Literature Search

From 883 studies, 666 were screened for eligibility by 2 raters (Cohen’s κ=0.62, 95% CI 0.55–0.70, p<0.001). A manual citation search yielded an additional 39 studies. Full-text review was performed on 136 studies from which 21 studies were included (Figure 1).

Figure 1: Literature search.

Quality Assessment

The QUADAS-2 tool was used to review risk of bias and applicability (Supplemental Figure I). Among the 4 domains, patient selection (26%) and reference standard (39%) domains were major sources of concern with high risk. Only 38% of studies reported at least 1 (n=7 studies)^17–22 or 2 (n=1 study)²³ atherosclerotic risk factors including age as inclusion criteria.

Study Characteristics

Study design, demographic data and prevalence of risk factors of the 21 studies are summarized in Table 1.

Table 1:

Study Characteristics

				Subject Inclusion Criteria			Demographic Data
Study	Study design	Country	Number of subjects (vessel segments)	Stenosis (modality)	Event to VWI interval	Infarct or TIA	Age (yrs)	Male (%)	HTN (%)	DM (%)	Dyslipidemia (%)	Smoking (%)	Event to VWI interval
Xu et al 2010.17	Pro	China	61 (61)	≥50% (TCD, MRA)	4wks	Both	62.4±11.6	63.9	68.9	29.5	31.1	36.1	14±6 d
Chung et al 2012.24	Pro	S. Korea	30 (30)	≥50% (MRA)	7d	Both	63/66*	63.3	56.7	46.7	70	43.3	NR
Shi et al 2012.25	Pro	China	36 (36)	≥50% (TCD, MRA)	3d	Both	57.5±8.7	63.9	52.8	41.7	63.9	61.1	2.2±0.4 d
Xu et al 2012.26	Retro	China	104 (109)	>70% (MRA)	1mo	Infarct	56.7±12.8	81.3	67.3	30.8	44.9	61.7	13±10 d
Ryoo et al 2015.15	Pro	S. Korea	80 (77)†	Any stenosis (MRA)	7d	Infarct	64.5± 13.7	61.3	67.5	41.3	51.3	27.5	NR
Zhao et al 2015.27	Pro	China	51 (51)	>30% (MRA)	1wk	Both	68.39±8.6	58.8	64.7	35.3	NR‡	31.4	NR
Kim et al 2016.28	Pro	S. Korea	138 (138)	n.s.	7d	Both	61.7±15.1	55.8	63	31.9	NR §	27.5/24.6 §§	NR
Peng et al 2016.19	Pro	China	111 (131)	>0% (NR)	NR	Both	57.5±10.5	72.1	NR	NR	NR	NR	NR
Bae et al 2017.14	Retro	S. Korea	59 (145)∥	n.s.	7d	Infarct	62.5	60	63.4	34.5	20	42.8	NR
Zhang et al 2017.23	Pro	China	33 (33)	>30% (MRA)	1wk	Infarct	68.1±11.8	81.8	81.8	51.5	NR‡	57.6	54.5±18.8 hrs
Wang et al 2017.29	Pro	China	57 (57)	>50% (CTA, DSA)	4wks	Infarct	59.4±8.1	77.2	63.2	52.6	21.1	49.1	13.4 d
Lu et al 2018a.20	Retro	China	46 (92)	< 50% (3D T1 VFA-TSE)	2wks	Both	60.68±12.6	69.6	76.1	39.1	34.8	32.6	4.3±3.3 d
Shan et al 2018.30	Retro	China	31 (62)	>0% (NR)	7d	Both	59.4±11.8	80.6	64.5	41.9	35.5	54.8	NR
Wu et al 2018a.18	Pro	China	52 (178)	≥50% (MRA, CTA or DSA)	1mo	Infarct	49.4±11.6	67.3	NR #	NR	NR‡	48.1	11.9±7.3 d
Wu et al 2018b.16	Pro	China	74 (74)	Any stenosis (MRA, CTA or DSA)	2wks	Infarct	54.7±12.1	79.7	75.7	21.6	40.5	52.7 (current)	8.3±3.9 d
Lu et al 2018b.21	Retro	China	46 (46)	>50% (MRA, 3D BB MRI)	1wk	Both	56.1±15.2	67.4	76.1	28.3	47.8	28.3 (current)	NR
Alexander et al 2019.31	Retro	USA	54 (54)	n.s.	14d	Infarct	62.8±16.2	61.1	NR	NR	NR	NR	7 d
Choi et al 2019.32	Pro	S. Korea	144 (144)	≥30% (conventional or MRA)	7d	Both	55.5±13.5	49.7	48.6	27.1	61.1	NR	NR
Liang et al 2019.22	Pro	China	47 (47)	>0% (MRA)	NR	Both	58.5	44.7	66	63.8	38.3	29.8	NR
Meng et al 2019.33**	Retro	China	192 (196)	>50% (MRA)	NR	Infarct	57.0±12.9	69.4	66.8	30.1	44.9	34.7	NR
Yu et al 2019.34	Retro	China	88 (108)	≥50% (T2-FSE)	7d	Infarct	58±15	0.75	65	26	43	NR	NR

Abbreviations: pro, prospective; retro, retrospective; S. Korea, South Korea; CV, cardiovascular; TIA, transient ischemic attack; HTN, hypertension; DM, diabetes mellitus; CAD, coronary artery disease; TCD, transcranial Doppler; BB MRI, black blood MRI; SD, standard deviation; hrs, hours; d, day; wks, weeks; mo, month; yrs, years; n.s., not inclusion screening criterion; VWI, vessel wall MR imaging; NR, not reported; TOF MRA, time of flight magnetic resonance angiography; CTA, computed tomographic angiography; DSA, digital subtraction angiography; FSE, fast spin echo; VFA-TSE, variable flip angle turbo spin echo

^*Median age for symptomatic (63years) and asymptomatic (66years)

^†Table in Ryoo et al¹⁵ reports 77 vessel segments.

^‡Lipid panel measures reported instead.

^§§27.5% current and 24.6% previous smokers

^∥Acute or chronic infarctions (n=145); acute infarction associated with plaque (n=59); demographic data reported for 145 subjects

^#Blood pressure ranges reported.

^**Construction cohort; per-plaque details for demographic data reported.

From 1,542 subjects, 1,750 vessel segments were pooled. Most studies were cross-sectional (n=20). The middle cerebral artery was most commonly evaluated as the sole artery or in combination with other vessel segments (95%). MR technical details are summarized in Supplemental Table II. Intracranial VWI protocols included both single-contrast (48%) and multi-contrast (52%) protocols. VWI protocols used 2D turbo/fast spin echo (n=11), 3D variable-flip angle turbo spin echo (n=7)^{18,23–27,33}, or both 2D and 3D sequences (n=3).^15,21,27 In-plane spatial resolutions ranged from 0.3 × 0.4 mm² to 0.7 × 0.9 mm².^19,27

VWI Features of Symptomatic Plaque

Forest plots for plaque enhancement, positive wall remodeling, T1 hyperintensity, and plaque surface irregularity are shown in Figure 2. Symptomatic plaque was significantly associated with enhancement (OR 7.4, 95% CI 3.4–16.4, p<0.001), positive remodeling (OR 5.6, 95% CI 2.2–14.0, p<0.001), T1 hyperintensity (OR 2.1, 95% CI 1.3–3.3, p<0.001) and surface irregularity (OR 4.5, 95% CI 1.4–8.6, p=0.01). Examples of plaque imaging features are shown in Figure 3. T2 signal intensity did not reach significance (OR 0.9, 95% CI 0.5–1.5, p=0.59) (Supplemental Figure II). Seven studies assessed plaque eccentricity but due to significant heterogeneity (I²=89.4%, p<0.001), results were not pooled (Supplemental Figure III).

Figure 2: Forest plots of imaging biomarkers of symptomatic plaque.

Forest plots of pooled data are graphically shown for 4 imaging endpoints. The vertical line represents an odds ratio of 1 (no effect). Squares represent point estimates of a study’s effect size; square sizes are proportional to the inverse of the variance of the estimate. Horizontal lines represent 95% CIs. Diamonds represent pooled estimates with the diamond width representing 95% CIs.

*Mean/weighted mean ± standard deviation or median [interquartile range]; if measured stenosis not reported, stenosis inclusion criterion noted in parentheses.

Abbreviations: PD, proton density; MCA, middle cerebral artery; BA, basilar artery; ICA, internal carotid artery; COW, circle of Willis; NR, not reported

Figure 3: VWI imaging features of symptomatic plaque.

(A) Coronal postcontrast VWI shows MCA plaque enhancement (arrowhead; inset: pre-contrast VWI). (B) Axial precontrast VWI shows positive wall remodeling (arrowhead) associated with a MCA plaque. Note the absence of severe luminal narrowing. (C) Coronal precontrast VWI shows T1 hyperintensity (arrowhead) of a left MCA plaque. (D) Sagittal precontrast VWI image of a basilar artery shows plaque surface irregularity (arrowheads along long-axis of basilar artery).

Subgroup Analyses

Subgroup analyses were performed to examine study design, MR technique, and subject inclusion criteria (Supplemental Table III). Plaque enhancement was the most frequently studied imaging biomarker (n=11). For enhancement, retrospective studies (OR 32.3, 95% CI 7.4–141.0) had significantly higher odds ratios compared to prospective studies (OR 4.5, 95% CI 2.4–8.3, p=0.02). A significantly higher odds ratio for enhancement emerged when at least 1 radiologist was a rater (OR 12.6, 95% CI 2.5–16.1) compared to if no radiologist was involved (OR 2.5, 95% CI 1.2–5.0, p<0.001). Finally, blinding to clinical history (p=0.39) or conventional imaging (p=0.74) did not significantly impact the association of symptomatic plaque enhancement in subjects with downstream ischemia. Notably, enhancement remained significantly associated with ischemic events despite differences in these study designs.

We examined whether MR sequence type or spatial resolution influenced the results (Supplemental Table IV). No significant difference emerged between 2D versus 3D isotropic sequences for plaque enhancement (p=0.87). However, a significantly higher odds ratio emerged with enhancement when sequences with lower spatial resolution (≥0.55m²) (OR 30.4, 95% CI 7.0–132.4) were used compared to higher spatial resolutions (<0.54mm²) (OR 4.6, 95% CI 2.4–8.6, p=0.02).

The association of plaque enhancement and ischemic event did not significantly differ between studies that included subjects with acute (≤7 days from event to VWI) versus subacute (7 days to 4 weeks) ischemic events (p=0.73). No significant difference emerged between studies that required inclusion of at least 1 atherosclerotic risk factor versus studies that did not have this inclusion criterion (Supplemental Table V).

TIA versus cerebral infarct

Only plaque enhancement was able to be extracted from enough studies to compare subjects with TIA versus cerebral infarcts (Figure 4). A nonsignificant difference between the 2 groups emerged with enhancement being significantly associated with cerebral ischemia (n=372/688 plaques; OR 3.0, 95% CI 1.7–5.2) but not in TIAs (n=105/192 plaques; OR 1.5, 95% CI 0.7–3.2; group difference, p=0.15).

Figure 4: Plaque enhancement in subjects with transient ischemic attack versus cerebral infarct.

Forest plots of plaque enhancement by group. Dotted diamonds represent respective group pooled statistics. Solid black diamond represents overall group (TIA and cerebral infarct) pooled statistic.

Sensitivity analysis

Based on the QUADAS-2 assessment, 3 studies were identified as having at least 2 domains with high risk of bias.^15,21,34 A sensitivity analysis excluding these 3 studies showed a similar direction of the main results (Supplemental Table VI).

Publication bias

A funnel plot for plaque enhancement (n=11 studies) and an Egger’s test (p=0.24) supported the absence of publication bias (Supplemental Figure IV).

Discussion

Increasing diagnostic confidence that an identified intracranial atherosclerotic plaque is indeed the cause of a patient’s ischemic stroke is important in reducing unnecessary diagnostic work up, guiding therapy, and reducing the risk of recurrence. The results show plaque enhancement, positive wall remodeling, T1 hyperintensity, and plaque surface irregularity are significantly associated with downstream ischemia. Our results add to the VWI literature and provide additional data to 2 previously published meta-analyses.^35,36 First, we present an updated literature search with inclusion of 6 new studies since 2018 enabling a comprehensive analysis of 6 potential plaque biomarkers, whereas Gupta et al³⁵ evaluated only plaque enhancement and Lee et al³⁶ omitted T2 signal intensity. To reduce heterogeneity, we also limit our analysis to acute and subacute ischemic events and exclude chronic infarcts, given imaging features likely change with time in chronic infarcts. Finally, we statistically address sources of study design bias and variability, which has not been shown before in the VWI literature, and report a strong and robust association between plaque enhancement and downstream ischemia.

Plaque enhancement may be due to neovascularization, inflammation, and endothelial dysfunction leading to leakage of gadolinium.³⁷ Recent work suggests vessel wall changes, such as enhancement, may be more important than luminal changes in predicting stroke occurrence.³⁸ In a longitudinal study, Kim et al reported a 30.3% one-year event rate of stroke recurrence in the presence of plaque enhancement, compared to 6.8% in the non-enhancing plaque group.²⁸ Identifying patients at greatest risk for recurrence with a predictive marker would be invaluable to help implement more aggressive therapies or continue them for a longer duration.

Although our analysis comparing TIA and infarcts did not reach significance, there was an attenuated association between plaque enhancement and symptomatic plaque in patients with TIA relative to those with ischemic stroke. This likely reflects the difficulty in determining which vessel is symptomatic in the absence of an acute infarct and the nebulous nature of TIA, where some transient events described as TIAs are not in fact vascular. Nevertheless, the analysis raises questions whether patients with a TIA and enhancing plaque could be at higher-risk for recurrence. Among a small sample of subjects with TIAs and capsular warning syndrome, plaque enhancement was significantly associated with recurrent infarct within 1 week.³⁹ Features that stratify stroke risk after TIA are critical for early management decisions,⁴⁰ and larger longitudinal studies to evaluate this association as a predictive marker are needed.

The effect of differences in study design was best assessed for plaque enhancement and revealed it to be a robust imaging biomarker. Retrospective studies showed a significantly higher association than prospectively designed studies, which may be due to selection bias. Notably only 10 studies reported using a consecutive or random enrollment method. Presence of a radiologist as one of the raters also showed a significantly higher strength of association. Possible explanations include longer length of image interpretation training and understanding of MR physics and artifacts. Finally, our results did not show significant effects due to blinding to clinical history or conventional imaging. When a criterion is objective, such as evaluating for presence/absence of enhancement, blinding may be less of an effect compared to more subjective assessments.⁷

Spatial resolution may also impact the ability to visualize specific plaque features. In-plane spatial resolutions up to 0.77 mm² did not significantly impact the association seen with positive wall remodeling. Enhancement, however, showed a stronger association at lower spatial resolutions, possibly due to partial volume averaging which may increase enhancement conspicuity. For smaller distal arteries, lower spatial resolutions could be considered an advantage.⁴¹ No effect was seen comparing 2D versus 3D sequences on plaque enhancement or T1 hyperintensity. The effect of 2D versus 3D sequences on positive remodeling could not be assessed due to the limited number of studies. However, given the inability to evaluate tortuous intracranial vessels without multiplanar reformats on 2D sequences, it is conceivable that sequence selection could influence measuring positive remodeling.

Other imaging features such as positive remodeling, T1 hyperintensity, and surface irregularity were also significantly associated with symptomatic plaque. These plaque features are also associated with vulnerable plaque in carotid and coronary arteries. For example, arterial wall remodeling is related to plaque burden and independently associated with ischemia in carotid and coronary arteries.^42,43 Also, intraplaque hemorrhage and rupture of the fibrous cap, manifesting as T1 hyperintensity and surface irregularity, have been histologically validated in carotid endarterectomy specimens.⁴⁴ While histologic correlates of in vivo imaged intracranial plaque are sparse,^45,46 it is promising that the same imaging features are significantly associated with intracranial symptomatic plaque.

Presence of T2 signal intensity was not seen to be significantly associated with symptomatic plaque. While T2w imaging as part of a multicontrast VWI protocol may help distinguish between ICAD and other vasculopathies,⁴⁷ the presence of T2 signal alone may not indicate a high-risk feature. In fact, a range of T2 signal intensities has been described to reflect fibrous caps, intraplaque hemorrhage, or thrombus.^48,49 Thus, a T2 signal intensity threshold may be more specific for detecting high-risk plaque components and warrants further investigation.³⁴

There are several limitations to this study. Although some between-study heterogeneity was present, our pooled analyses all had I² < 50% indicating less than moderate heterogeneity. Second, several imaging features were not able to be assessed in subgroup analyses due to an insufficient number of studies partly related to limited reporting. Two attempts were made to reach authors to request more data. Third, the wide confidence intervals suggest low statistical power leading to imprecise estimates of effects, which may increase the likelihood of not identifying clinically important associations. Fourth, measurement methods varied; for example, positive remodeling was scored as a remodeling index ≥1.05 in all studies except one, which used a threshold of ≥1.2.¹⁵ Similarly, T2 signal intensity measurement methods varied among investigators ranging from “heterogeneous”²⁸ and “hyperintense” signal.²¹ Fifth, we were not able to adjust for stenosis degree. Although most studies used stenosis as an inclusion criterion, this measurement ranged from any stenosis to >70% stenosis. Whether stenosis grade together with specific plaque imaging features could further strengthen diagnostic confidence in identifying symptomatic plaque is an area of future research. It is also worth noting that in the absence of stenosis and presence of positive remodeling, VWI may add value in evaluating cryptogenic strokes.

Summary

Several VWI imaging features are promising as imaging biomarkers for symptomatic intracranial plaque. Future prospective studies of patients with ischemia with follow-up should be undertaken to assess the strength of these imaging features as predictive markers. Such studies may also identify high-risk groups for recurrence and inform the need for more aggressive secondary stroke measures including prolonged dual antiplatelet therapy and selecting patients for revascularization interventions.

Nonstandard abbreviations and acronyms:

VWI

vessel wall MR imaging

ICAD

intracranial atherosclerotic disease

TIA

transient ischemic attack