Plaque Enhancement Grade on High-Resolution MRI and Recurrent Ischemic Events in Intracranial Stenosis: A Retrospective Cohort Study

Summary

In 381 medically managed patients with 30%–69% intracranial stenosis, higher plaque enhancement grade on high-resolution magnetic resonance imaging predicted recurrent ischemic stroke or transient ischemic attack.

Abstract

Recurrent ischemic events may occur in patients with mild-to-moderate intracranial atherosclerotic stenosis despite standardized medication; stenosis rate alone is insufficient for risk stratification, and plaque activity imaging markers are needed. Because plaque enhancement may reflect plaque activity beyond luminal narrowing, we hypothesized that higher enhancement grade would predict recurrent ischemic events of stenosis rate. This study evaluated the association between high-resolution magnetic resonance imaging (HR-MRI) plaque enhancement grading and recurrent ischemic events and examined its incremental value beyond stenosis rate. In a retrospective medication-treated cohort, patients aged 18–85 years with ischemic stroke or TIA and responsible intracranial arterial stenosis of 30%–69% completed contrast-enhanced HR-MRI within 14 days of onset and received standardized secondary prevention. Enhancement referenced to the pituitary stalk was graded 0–2 with blinded reading. The primary outcome was recurrent ischemic events (stroke or TIA) in the responsible vascular territory. Time-to-event analysis used left truncation. Kaplan–Meier and Cox models adjusted for stenosis rate, LDL-C, age, and imaging interval. Among 381 patients, loss to follow-up was 4.72%. Recurrence occurred in 52 (13.65%) patients; median event time was 171.84 days. Recurrence-free survival differed by enhancement grade (log-rank P = 0.002). Versus grade 0, grade 2 was associated with higher recurrence risk (simplified adjusted HR = 2.50, 95% CI 1.30–4.82; fully adjusted HR = 2.31, 95% CI 1.17–4.56). Per 1-grade increase, risk increased (fully adjusted HR = 1.33, 95% CI 1.07–1.65; trend P = 0.01); grade 1 vs grade 0 was not significant. For stroke recurrence, grade 2 remained associated with higher risk (aHR = 2.63, 95% CI 1.03–6.69). Higher-grade HR-MRI plaque enhancement independently predicts recurrent ischemic events in mild-to-moderate intracranial arterial stenosis and provides incremental risk stratification beyond stenosis rate, supporting the study hypothesis and identifying patients requiring intensified follow-up and risk-factor target achievement.

Introduction

Ischemic stroke remains one of the main causes of death and disability¹, and intracranial atherosclerotic stenosis is more common in Asian populations and is an important mechanism of ischemic stroke and recurrence². Standardized antiplatelet and statin therapy can reduce the risk of recurrence³, but in clinical practice, patients with mild-to-moderate stenosis may still develop recurrent events in the responsible vascular territory; such populations usually do not meet indications for revascularization⁴. Treatment strategies mainly rely on medication and risk factor management, and earlier and more accurate risk stratification tools are needed. Traditional stenosis rate assessment based on lumen imaging focuses on structural burden and cannot reflect biological instability such as intraplaque inflammatory activity, neovascularization of the intima, and impairment of the endothelial barrier, and these processes are closer to triggering links of recurrence such as artery-to-artery embolism and local perfusion decompensation⁵. High-resolution MRI vessel wall imaging makes it possible to evaluate intracranial plaque phenotype in vivo, and post-contrast plaque enhancement is associated with inflammation and vulnerable phenotype⁶. Previous studies have mostly focused on symptomatic differentiation or populations with severe stenosis; direct evidence is still lacking regarding the predictive value for recurrence in mild-to-moderate stenosis under a background of uniform medication treatment. Meanwhile, non-uniform evaluation standards for enhancement and time bias caused by differences in examination time points also limit the generalizability of conclusions⁷. Based on this, using a medication-treated cohort design, we included patients with mild-to-moderate responsible intracranial arterial stenosis who completed HR-MRI early after onset, applied a reproducible three-grade plaque enhancement system, explored the association between enhancement grading and recurrent ischemic events in the responsible vascular territory during follow-up, and controlled for the effects of stenosis rate and key clinical factors within a time-to-event framework. The purpose of this study was to clarify whether enhancement grading can provide incremental risk information beyond stenosis severity, and to provide imaging evidence for high-risk identification and follow-up management in populations with mild-to-moderate intracranial atherosclerosis. We hypothesized that, in medically managed patients with mild-to-moderate intracranial atherosclerotic stenosis, higher plaque enhancement grade on high-resolution magnetic resonance imaging would be associated with an increased risk of recurrent ischemic events in the responsible vascular territory independent of stenosis severity and key clinical factors.

Protocol

This retrospective study involving human participants complied with institutional guidelines and was approved by the Ethics Committee of Quzhou Hospital of Traditional Chinese Medicine (Approval No.: 2023-08-233). The requirement for written informed consent was waived. All included cases were de-identified before statistical analysis.

Study design

This study was a single-center retrospective medication-treated cohort study. Eligible patients who visited the Department of Neurology of our hospital between January 1, 2022 and July 31, 2024 were consecutively enrolled retrospectively, and follow-up ended on August 31, 2025 (data lock date). The study site was the Department of Neurology of this hospital. The core exposure factor was HR-MRI plaque enhancement three-grade classification (Grade 0/1/2), and the primary outcome was recurrent ischemic events in the responsible vascular territory. The overall study design and analytical workflow are summarized in Figure 1 and were implemented according to the following pre-specified steps: retrieve records of visits for ischemic stroke or transient ischemic attack within the study time window in the hospital information system, and verify in the imaging PACS whether intracranial HR-MRI vessel wall imaging and contrast-enhanced scans were completed; form the study cohort according to inclusion and exclusion criteria, and complete determination of the responsible vessel and measurement of stenosis rate; use the date of index event onset (or the date of first visit/triage, whichever was earlier) as the baseline time point to extract clinical data, laboratory indicators, and medication information; and complete HR-MRI image reading and enhancement grading within 14 days after the index event; verify outcome events according to pre-specified follow-up time points and construct time-to-event data; complete statistical analysis according to the pre-specified statistical plan.

Study participants

The study participants were patients who visited our hospital within the study time window, developed ischemic stroke or transient ischemic attack, and completed intracranial HR-MRI vessel wall imaging. A cohort was constructed by consecutive screening and enrolment of eligible patients within the study time window, and the final sample size was determined by the number of eligible cases. To ensure the stability of the multivariable Cox proportional hazards regression model, the number of covariates was pre-specified, and the EPV ≥ 10 principle was referenced⁸; the actual EPV is reported in the results.

Inclusion criteria includes: age 18–85 years; the clinical event was ischemic stroke or transient ischemic attack and could be localized to a single responsible intracranial arterial territory; responsible intracranial arterial stenosis was 30%–69%; HR-MRI vessel wall imaging and contrast-enhanced scans were completed within 14 days after the index event; standardized secondary prevention medication treatment had been initiated at baseline, with both antiplatelet therapy and statin therapy started; stenosis rate, enhancement grading, follow-up outcome information, and the time of last follow-up were complete and available.

Exclusion criteria include: evidence of a definite non-atherosclerotic mechanism for ischemic events, including definite diagnosis of cardioembolic stroke, definite diagnosis of vasculitis, and definite imaging diagnosis of arterial dissection; the responsible artery had previously received interventional treatment or surgical revascularization; HR-MRI images were not readable, defined as inability to identify the inner and outer boundaries of the vessel wall or failure of pre- and post-contrast registration leading to inability to assess enhancement; the responsible vessel could not be determined; key variables were missing, including stenosis rate, enhancement grading, outcome event information, or time of last follow-up.

Determination of the responsible vessel followed uniform rules: When acute-phase DWI showed an acute ischemic lesion in a single vascular territory, the responsible vascular territory was determined according to the correspondence between the anatomical distribution of the ischemic lesion and intracranial arterial territories, and the stenotic segment within the corresponding territory was defined as the responsible vessel. When DWI was negative and the definition of transient ischemic attack was met, the responsible vascular territory was determined according to the anatomical localization of neurological deficits and the distribution of vascular territories, and the stenotic segment within the corresponding territory was defined as the responsible vessel. When multiple stenoses were present within the same territory, the segment that best matched clinical localization and had the most severe stenosis was selected as the responsible stenotic segment.

Stenosis severity was calculated using the WASID method⁹. Stenosis rate (%) = [1 − (lumen diameter at the narrowest stenosis/proximal normal lumen diameter)] × 100%. Mild stenosis was defined as 30%–49%, and moderate stenosis was defined as 50%–69%. Stenosis rate was recorded as a continuous variable, and a mild/moderate stratification variable was also retained for descriptive statistics and model adjustment.

Clinical data and variable definitions

Baseline clinical data and variable definitions were uniformly extracted at the baseline time point and applied using a single standard. Demographic variables were age and sex. Risk factor variables were recorded as binary variables: hypertension was defined as a previous definite diagnosis of hypertension or use of antihypertensive medication at baseline; diabetes was defined as a previous definite diagnosis of diabetes or use of glucose-lowering medication at baseline; smoking was defined as being a current smoker at baseline, with the criterion of continuous smoking within the past 30 d. Baseline laboratory indicators mandatorily included low-density lipoprotein cholesterol (LDL-C). The LDL-C time window was defined as the most recent fasting lipid test result within 48 h after the index event, with the unit uniformly as mmol/L; patients without an LDL-C result within 48 h after the index event were recorded as missing. Medication treatment background variables were used only for confounding adjustment and were uniformly recorded based on the baseline prescription regimen. Antiplatelet therapy was categorized as single antiplatelet therapy (SAPT) and dual antiplatelet therapy (DAPT): SAPT was defined as aspirin 100 mg/day monotherapy or clopidogrel 75 mg/day monotherapy; DAPT was defined as aspirin 100 mg/day combined with clopidogrel 75 mg/day. Statin intensity was categorized as high-intensity and non-high-intensity: high-intensity was defined as atorvastatin 40–80 mg/day or rosuvastatin 20 mg/day; non-high-intensity was defined as statin regimens that did not meet the above dose thresholds.

HR-MRI examination and imaging assessment

All patients completed intracranial vessel wall HR-MRI and were assessed according to a unified workflow¹⁰. Scan protocol and key sequences: the scanner was a 3.0T MRI system using a 32-channel head-neck combined coil; the scan sequences mandatorily included 3D TOF-MRA, 3D T1-weighted vessel wall imaging (pre-contrast), and 3D T1-weighted vessel wall imaging (post-contrast). The contrast agent was gadopentetate dimeglumine, with a dose of 0.1 mmol/kg and an injection rate of 2.0 mL/s; the start time of post-contrast vessel wall scanning was fixed at 5 min after completion of injection. Image reconstruction used isotropic voxels of 0.6 mm, and the scan coverage included the Circle of Willis and the segments related to the responsible artery. Image quality control: the quality control criteria were fixed as clear vessel wall layers, identifiable inner and outer boundaries of the vessel wall, and consistent pre- and post-contrast registration; if any criterion was not met, the images were judged as unreadable and handled according to the exclusion criteria. Plaque identification and matching to the responsible artery: plaque was defined as focal or eccentric thickening of the wall of the responsible artery; the plaque assessment level was fixed at the level corresponding to the narrowest stenosis of the responsible artery, and extended 2 levels proximally and 2 levels distally to confirm plaque continuity; the plaque assessment range was limited to the responsible vessel segment. Plaque enhancement grading: enhancement was graded using a three-grade classification (Grade 0/1/2). Determination was made by comparing post-contrast T1 vessel wall images with pre-contrast T1 vessel wall images, with the reference structure fixed as the enhancement intensity of the pituitary stalk. Grade 0 was defined as plaque enhancement intensity ≤ the intracranial arterial wall without plaque at other sites in the same patient and lower than the pituitary stalk; Grade 1 was defined as plaque enhancement intensity > the intracranial arterial wall without plaque at other sites in the same patient and lower than the pituitary stalk; Grade 2 was defined as plaque enhancement intensity greater than or equal to the pituitary stalk¹¹. Enhancement grading was recorded as an ordinal categorical variable based on the final grade. Image reading process and consistency evaluation: image reading was independently performed by two radiologists experienced in vessel wall imaging, and the readers were blinded to clinical outcomes and follow-up information; the two readers separately provided enhancement grades, and if grades were inconsistent, a third senior physician adjudicated and determined the final grade; consistency evaluation used the weighted Kappa coefficient, and the Kappa value and its 95% CI were reported in the Results section.

Observation indicators and evaluation criteria

Data collection time points and frequency definition

This study used a collection mode of single baseline acquisition and multiple follow-up verifications. The baseline time point was defined as the date of index event onset (or the date of first visit/triage, whichever was earlier). Follow-up was set at fixed verification time points: T1 was defined as 90 ± 15 days after baseline; T2 was defined as 180 ± 15 days after baseline; T3 was defined as 365 ± 30 days after baseline. The follow-up method was fixed as verification of outpatient follow-up records and telephone follow-up verification. The follow-up termination conditions were fixed as occurrence of the primary outcome event, completion of T3 follow-up, or reaching the study end date. Loss to follow-up was defined as inability to obtain outcome information at two consecutive fixed verification time points and no outpatient records.

Time windows and evaluation criteria for each indicator

Plaque enhancement grading was assessed on the day of HR-MRI examination completed within 14 days after the index event and recorded as the baseline exposure variable. The responsible arterial stenosis rate was measured on the day of HR-MRI/TOF-MRA assessment and recorded as the baseline imaging variable. Age, sex, and risk factors (hypertension, diabetes, smoking) were collected once at baseline, sourced from admission medical records and prior diagnoses/medication records. LDL-C was collected within 48 h after the index event, sourced from fasting lipid test results in the laboratory system. The antiplatelet regimen and statin intensity were collected once at baseline.

The primary outcome was recurrent ischemic events in the responsible vascular territory, with the collection time window from baseline to follow-up termination. Outcome event information was verified at fixed time points T1, T2, and T3, and emergency or hospitalization events occurring between two verifications were additionally checked. Ischemic stroke was defined as new focal neurological deficits lasting more than 24 h and confirmed by head MRI or CT showing a new ischemic infarct consistent with symptoms. Transient ischemic attack was defined as focal neurological deficits lasting no more than 24 h and head MRI (preferably DWI) or CT showing no evidence of acute infarction consistent with symptoms¹². The rules for determining recurrence in the responsible vascular territory were fixed as recurrence events being consistent with the baseline responsible arterial territory: for stroke, determination was based on matching the distribution of the new infarct lesion with the vascular territory; for transient ischemic attack, determination was based on matching symptom localization with the vascular territory. The event date was fixed as the earlier of the imaging examination date and the emergency or hospitalization triage date; the time-to-event variable was defined as the number of days from baseline to the event date. For patients without recurrence events, the censoring time was the T3 completion date or the last available follow-up date. The interval (d) from the index event to completion of HR-MRI was also recorded for multivariable model adjustment.

Statistical analysis

Considering that exposure was obtained on the day of HR-MRI, time-to-event analysis used left truncation: the index event date was used as the time scale origin, and the time of entry into the risk set was defined as the HR-MRI completion date¹³. Recurrence occurring before HR-MRI completion was not included in the analysis. For continuous variables, the Shapiro–Wilk test was used to assess normality, and the Levene test was used to assess homogeneity of variance; continuous variables satisfying normality and homogeneity of variance were described as mean ± SD, otherwise as median (interquartile range); categorical variables were described as n and percentage. Baseline characteristics were compared by enhancement grade (Grade 0/1/2). For continuous variables satisfying normality and homogeneity of variance, one-way analysis of variance was used; otherwise, the Kruskal-Wallis test was used. Categorical variables were compared using the chi-square test.

For time-to-event analysis, the Kaplan–Meier method was used to plot recurrence-free survival curves, and the log-rank test was used to compare recurrence differences among enhancement grades. The primary association analysis used the Cox proportional hazards regression model, reporting hazard ratio (HR) and 95% CI. Given the limited number of outcome events, the main analysis was pre-specified to use a simplified adjustment model: adjusting for stenosis rate, LDL-C, the interval from the index event to HR-MRI, and age (enhancement grade entered the model as an ordinal variable). The fully adjusted model further included sex, hypertension, diabetes, smoking, antiplatelet regimen, and statin intensity on this basis as a sensitivity analysis. Using Grade 0 as the reference, HRs for Grade 1 and Grade 2 were estimated, and enhancement grade was treated as an ordinal variable for trend testing. The proportional hazards assumption was evaluated using the Schoenfeld residual test.

Sensitivity analysis was pre-specified as repeating the above Cox regression analysis using only recurrent ischemic stroke as the outcome event, to test the stability of conclusions under a stricter outcome definition. Missing data handling: patients with missing exposure factor (enhancement grade), outcome event information, or time of last follow-up were excluded; missing values of other covariates were handled by multiple imputation to complete the main analysis, and complete-case analysis was used as a sensitivity analysis. Statistical tests were two-sided, with significance level set at P < 0.05. The statistical software was fixed as R.

Results

Study participants and cohort screening

A total of 1879 stroke/TIA (transient ischemic attack) records were retrieved, and 548 patients completed HR-MRI within 14 days after the index event. Patients who did not complete HR-MRI within this 14 days window were not eligible for exposure-based time-to-event analysis. After excluding cases with non-atherosclerotic mechanisms, non-eligible stenosis range, unclear responsible vessel, prior revascularization, unreadable images, and key missing data according to pre-specified criteria, 381 patients entered the analysis and constituted the final cohort (Figure 2). The exclusion categories and cohort construction process are summarized in Figure 2.

Baseline clinical data and HR-MRI imaging assessment

When baseline characteristics were compared by enhancement grade, LDL-C, stenosis rate and stenosis stratification, and the interval to HR-MRI differed significantly (P < 0.001 or P = 0.006), while differences in other variables were not significant (Table 1). Baseline antiplatelet regimen and statin intensity did not differ significantly across enhancement grades and were retained as covariates in the fully adjusted model. The grading consistency assessment showed good agreement, as reflected by the observed agreement and weighted Kappa values reported in Table 2. Representative annotated examples of Grade 0, Grade 1, and Grade 2 plaque enhancement are shown in Supplementary Figure 1.

Follow-up verification and outcome event determination

Follow-up was conducted according to the pre-specified T1/T2/T3 time windows by verifying outcome status through outpatient records and telephone follow-up, with few losses to follow-up. During follow-up, the primary outcome was recorded and determination of consistency with the responsible vascular territory was completed, event dates and censoring times were determined, and time-to-event data were constructed (Table 3). The responsible stenotic segments were the middle cerebral artery in 166 patients (43.6%), intracranial internal carotid artery in 68 (17.8%), basilar artery in 67 (17.6%), vertebral artery in 55 (14.4%), and posterior cerebral artery, anterior cerebral artery, or other segments in 25 (6.6%). Segment-specific recurrence rates ranged from 11.4% to 19.4%, with numerically higher rates in the basilar and vertebral artery groups. Because the number of events within individual vascular subgroups was limited, these comparisons were interpreted descriptively (Supplementary Table 1).

Missing data and multiple imputation

Missingness was limited to current smoking, LDL-C, and statin intensity, with missing rates of 0.52% (2/381), 2.62% (10/381), and 0.26% (1/381), respectively. No missingness was observed in other covariates included in the analysis. According to the pre-specified plan, missing covariates were imputed in the main analysis, and complete-case analysis was performed as a sensitivity analysis to assess robustness (Table 4).

Recurrence-free survival analysis

Absolute recurrence rates stratified by stenosis severity and plaque enhancement grade are reported in Supplementary Table 2. Among patients with 30%–49% stenosis, recurrence rates were 8.2%, 8.5%, and 18.5% for Grade 0, Grade 1, and Grade 2 enhancement, respectively. Among patients with 50%–69% stenosis, the corresponding recurrence rates were 12.3%, 13.4%, and 26.2%. Grade 2 enhancement showed the highest absolute recurrence rate in both stenosis strata. The recurrence-free survival curves for different enhancement grades showed a separation trend, with a higher recurrence risk in the high enhancement group. Between-group differences were estimated using the Kaplan–Meier method and compared using the log-rank test, and the results indicated a significant overall difference (P = 0.002) (Figure 3). These findings supported the subsequent Cox regression analysis, in which stenosis rate was incorporated as an adjustment variable.

Cox-regression analysis for the primary outcome

Under left truncation, enhancement grade showed an increasing association with recurrence risk. Because the number of recurrent events was limited, the simplified adjusted Cox model was used as the primary model. The simplified adjusted Cox model (adjusting for stenosis rate, LDL-C, the interval from index event to HR-MRI, and age) showed that Grade 2 remained associated with a higher recurrence risk, and the trend test for grade was significant (Wald test, P = 0.010); results of the fully adjusted model were consistent and were interpreted as sensitivity support (Table 5).

Sensitivity analysis and model diagnostics

In the simplified primary model, Grade 2 showed increased recurrence risk compared with Grade 0 (HR = 2.50, 95% CI 1.30–4.82). The fully adjusted model yielded a consistent estimate (aHR = 2.31, 95% CI 1.17–4.56, P = 0.016) and was interpreted as sensitivity support because of the limited number of recurrent events. The sensitivity analysis with recurrent ischemic stroke only showed consistent results (aHR = 2.63, 95% CI 1.03–6.69, P = 0.043); the difference for Grade 1 was not significant (P = 0.329/0.491) (Figure 4). The proportional hazards assumption test showed no violation; the EPV of the fully adjusted model was low, and the results were interpreted only as sensitivity (Table 6).

Taken together, these findings support the study hypothesis that higher plaque enhancement grade on HR-MRI is associated with a higher risk of recurrent ischemic events after adjustment for stenosis rate and key clinical factors. Grade 2 enhancement showed the clearest risk separation, whereas Grade 1 did not differ significantly from Grade 0. Because stenosis rate was incorporated into the Cox models, the observed association supports the incremental value of plaque enhancement grading beyond luminal stenosis severity. The consistent direction of the simplified primary model, fully adjusted sensitivity model, and stroke-only sensitivity analysis supports plaque enhancement grading as an incremental imaging marker for risk stratification in mild-to-moderate intracranial stenosis.

DATA AVAILABILITY:

The de-identified raw data supporting this study have been provided as supplementary materials. No identifiable patient information is included.

Figure 1: Overall study design and analytical workflow. The workflow summarizes cohort determination, baseline clinical and vascular assessment, contrast-enhanced high-resolution magnetic resonance imaging within 14 days after the index event, blinded plaque enhancement grading, outcome verification at 90 ± 15, 180 ± 15, and 365 ± 30 days, and left-truncated survival analysis with Cox regression and sensitivity analyses. HR-MRI, high-resolution magnetic resonance imaging; TIA, transient ischemic attack; LDL-C, low-density lipoprotein cholesterol; WASID, Warfarin–Aspirin Symptomatic Intracranial Disease. Please click here to view a larger version of this figure.

Figure 2: Flowchart of cohort screening and inclusion of study participants. A total of 1,879 stroke/TIA records were retrieved, and 548 patients completed intracranial high-resolution magnetic resonance imaging (HR-MRI) within 14 days after the index event. After excluding non-atherosclerotic mechanisms, non-eligible stenosis range, unclear responsible vessel, prior revascularization, unreadable images, and key missing data, 381 patients were included in the final analysis. Please click here to view a larger version of this figure.

Figure 3: Recurrence-free survival curves for ischemic events in patients with different HR-MRI plaque enhancement grades. Left truncation Kaplan–Meier estimation was used, and between-group comparison used the log-rank test; 95% CIs and numbers at risk are shown. Please click here to view a larger version of this figure.

Figure 4: Forest plot of plaque enhancement grade and recurrence risk. Based on the left-truncated Cox models, hazard ratios and 95% CIs for Grade 1/2 relative to Grade 0 are shown; sensitivity analysis used recurrent ischemic stroke as the outcome. Please click here to view a larger version of this figure.

Variable		Grade 0(n = 118)	Grade 1(n = 143)	Grade 2(n = 120)	Statistic	P value
Age (years)		62.01 ± 7.28	62.71 ± 7.60	62.95 ± 7.28	F = 0.521	0.594
Sex (male)		70(59.3%)	97(67.8%)	86(71.7%)	χ² = 4.273	0.118
Hypertension		82(69.5%)	103(72.0%)	91(75.8%)	χ² = 1.218	0.544
Diabetes		33(28.0%)	44(30.8%)	40(33.3%)	χ² = 0.806	0.668
Current smoking		28(23.7%)	45(31.5%)	46(38.3%)	χ² = 5.914	0.052
LDL-C (mmol/L)		2.11 ± 0.25	2.24 ± 0.27	2.34 ± 0.30	F = 21.081	<0.001
LDL-C missing		3(2.5%)	4(2.8%)	3(2.5%)	Fisher	1.000
Antiplatelet regimen: SAPT		68(57.6%)	72(50.3%)	51(42.5%)	χ² = 5.450	0.066
Antiplatelet regimen: DAPT		50(42.4%)	71(49.7%)	69(57.5%)	χ² = 5.450	0.066
Statin intensity: high-intensity		64(54.2%)	82(57.3%)	79(65.8%)	χ² = 3.586	0.166
Statin intensity: non-high-intensity		54(45.8%)	61(42.7%)	41(34.2%)	χ² = 3.586	0.166
Responsible circulation: anterior circulation		92(78.0%)	113(79.0%)	95(79.2%)	χ² = 0.062	0.969
Responsible circulation: posterior circulation		26(22.0%)	30(21.0%)	25(20.8%)	χ² = 0.062	0.969
Stenosis rate (%) (WASID)		49.99 ± 4.33	51.97 ± 4.44	53.93 ± 4.71	F = 22.869	<0.001
Stenosis stratification: 30-49%		63(53.4%)	53(37.1%)	31(25.8%)	χ² = 19.289	<0.001
Stenosis stratification: 50-69%		55(46.6%)	90(62.9%)	89(74.2%)	χ² = 19.289	<0.001
Interval from index event to HR-MRI (d)		7.82 ± 1.07	7.86 ± 1.10	7.46 ± 1.07	F = 5.171	0.006

Table 1: Baseline characteristics grouped by plaque enhancement grade on HR-MRI. One-way analysis of variance was used for continuous variables; χ² test was used for categorical variables; Fisher exact test was used for “LDL-C missing” due to small counts. Continuous variables are presented as mean ± SD.

Indicator	Value
Number of cases consistent between the two readers	308(80.84%)
Number of cases inconsistent between the two readers	73(19.16%)
Inconsistent cases adjudicated by the third reader	73(19.16%)
Observed agreement Po	80.84%
Weighted Kappa κw	0.79
κw SE	0.03
κw 95%CI	0.73-0.85

Table 2: Consistency of imaging reading and adjudication. Two readers independently evaluated images and were blinded to outcome information; discrepant grades were adjudicated by a third senior physician. Observed agreement (Po), weighted Kappa (κw), SE, and 95% CI are reported.

Indicator	Value
Follow-up implementation
T1 (90±15 d after baseline)	378(99.21%)
T2 (180±15 d after baseline)	371(97.38%)
T3 (365±30 d after baseline)	361(94.75%)
Loss to follow-up	18(4.72%)
Follow-up duration (d)	364.87(347.92,377.11)
Outcome events and censoring
Primary outcome events (recurrent stroke or TIA)	52(13.65%)
of which: recurrent ischemic stroke	31(8.14%)
of which: recurrent TIA	21(5.51%)
Consistency with responsible vascular territory: “Yes”	48(92.31%)
Time-to-event among event cases (d)	171.84(97.13,270.69)
Censoring: completed T3 without recurrence	309(81.10%)
Censoring: last available follow-up without recurrence	20(5.25%)

Table 3: Summary of follow-up implementation and outcome event determination. “Verifiable” for T1/T2/T3 refers to obtaining outcome status information within the corresponding time window, including patients with outcomes already occurring. Loss to follow-up was defined as inability to verify outcome status at two consecutive time points and absence of outpatient records; loss to follow-up was censored at the last available follow-up date. The last censoring group included 20 patients, including 18 patients lost to follow-up. Categorical data are presented as n (%), and continuous data are presented as median (interquartile range).

Variable	Missing n(%)	Multiple imputation method
Current smoking	2(0.52%)	Logistic
LDL-C (mmol/L)	10(2.62%)	PMM
Statin intensity (high vs non-high)	1(0.26%)	Logistic

Table 4: Covariate missingness and multiple imputation parameters. This table lists only covariates with missing values. Missing covariates were handled using multiple imputation by chained equations (MICE), with m = 21 imputations. Predictive mean matching (PMM) was used for continuous variables, and logistic regression imputation was used for binary variables. Estimates across imputed datasets were pooled according to Rubin’s rules. The imputation model included enhancement grade, event indicator, and follow-up time; covariates without missing values were included only as predictors in the imputation model.

Variable		Univariate HR (95%CI)	Wald z	P value	Simplified adjusted HR (95%CI) (main analysis)	Fully adjusted HR (95%CI)	Wald z	P value
Enhancement grade: Grade 1 vs Grade 0		1.55(0.77–3.11)	1.231	0.218	1.46(0.72–2.96)	1.42(0.70–2.86)	0.977	0.329
Enhancement grade: Grade 2 vs Grade 0		2.87(1.48–5.61)	3.102	0.002	2.50(1.30–4.82)	2.31(1.17–4.56)	2.413	0.016
Enhancement grade (ordinal variable, per 1-grade increase)		1.47(1.22–1.76)	4.121	<0.001	1.39(1.15–1.68)	1.33(1.07–1.65)	2.581	0.010
Stenosis rate (per 1% increase)		1.04(1.02–1.06)	3.997	<0.001	1.03(1.01–1.05)	1.03(1.01–1.05)	2.983	0.003
Age (per 1-year increase)		1.01(0.99–1.03)	0.985	0.325	1.01(0.99–1.03)	1.01(0.98–1.03)	0.784	0.433
Sex (male vs female)		1.22(0.74–2.01)	0.78	0.435	1.13(0.66–1.93)	1.10(0.63–1.93)	0.334	0.739
Hypertension (yes vs no)		1.18(0.67–2.07)	0.575	0.565	1.17(0.67–2.05)	1.16(0.65–2.09)	0.498	0.618
Diabetes (yes vs no)		1.29(0.78–2.13)	0.994	0.32	1.26(0.76–2.07)	1.21(0.71–2.06)	0.701	0.483
Current smoking (yes vs no)		1.43(0.88–2.33)	1.44	0.15	1.36(0.84–2.21)	1.31(0.78–2.20)	1.021	0.307
Antiplatelet regimen (DAPT vs SAPT)		0.79(0.49–1.26)	–0.978	0.328	NA	0.83(0.50–1.37)	–0.725	0.469
Statin intensity (high-intensity vs non-high-intensity)		0.74(0.46–1.18)	–1.253	0.21	NA	0.78(0.47–1.30)	–0.957	0.338
LDL-C (per 1 mmol/L increase)		1.58(1.21–2.06)	3.37	<0.001	NA	1.41(1.03–1.93)	2.145	0.032
Interval from index event to HR-MRI (per 1 d increase)		0.95(0.87–1.05)	–1.069	0.285	NA	0.96(0.88–1.05)	–0.906	0.365

Table 5: Left truncation Cox regression results for the primary outcome. The Cox model used left truncation, with entry into the risk set defined as the HR-MRI completion date; P values are from two-sided Wald tests. The simplified primary model adjusted for stenosis rate, LDL-C, age, and the interval from index event to HR-MRI. The fully adjusted model additionally adjusted for sex, hypertension, diabetes, smoking, antiplatelet regimen, and statin intensity and was interpreted as sensitivity support. Grade 1/2 vs 0 are results from the categorical model; “per 1-grade increase” is the result from the ordinal-variable model.

Model		Events/k(EPV)	Schoenfeld global χ²(df)	P value	Schoenfeld: enhancement grade χ²(df)	P value
Main multivariable Cox (left truncation)		52/12(4.33)	8.742(12)	0.725	1.604(2)	0.448
Sensitivity multivariable Cox (recurrent ischemic stroke only; left truncation)		31/12(2.58)	10.316(12)	0.588	1.138(2)	0.566

Table 6: Model adequacy and diagnostic results. EPV = events/k, where k indicates the number of model parameters. The Schoenfeld residual test was used to evaluate the proportional hazards assumption, and χ²(df) and P values are reported for the global test and the primary exposure term (overall enhancement grade).

Supplementary Figure 1: Representative examples of plaque enhancement grades on high-resolution magnetic resonance imaging. Representative contrast-enhanced vessel wall images show Grade 0, Grade 1, and Grade 2 plaque enhancement. Grade 0 indicates no enhancement or enhancement not exceeding the non-plaque intracranial arterial wall; Grade 1 indicates enhancement greater than the non-plaque arterial wall but lower than the pituitary stalk; Grade 2 indicates enhancement equal to or greater than the pituitary stalk. Arrows indicate the responsible plaque.Please click here to download this file.

Supplementary Table 1: Distribution of responsible stenotic segments and segment-specific recurrent ischemic events. The responsible stenotic segment was determined according to the vascular territory of the index ischemic event. Other segments included the posterior cerebral artery, anterior cerebral artery, and less frequent responsible intracranial arterial segments. Segment-specific recurrence rates were evaluated descriptively because of limited event numbers. TIA, transient ischemic attack.Please click here to download this file.

Supplementary Table 2: Absolute recurrence rates stratified by stenosis severity and plaque enhancement grade. Absolute recurrence rates were calculated as the number of recurrent ischemic events divided by the total number of patients in each stenosis-by-enhancement stratum. Mild stenosis was defined as 30%–49%, and moderate stenosis was defined as 50%–69%.Please click here to download this file.

Discussion

In the context of mild-to-moderate intracranial arterial stenosis receiving standardized secondary prevention medication treatment, plaque enhancement grade could still stably distinguish the risk of recurrent ischemic events during follow-up. This finding supports the study hypothesis that higher plaque enhancement grade provides recurrence-risk information beyond luminal stenosis severity. After incorporating factors such as stenosis rate, lipid level, age, and the interval between imaging examinations into the model, the direction of the association between enhancement and recurrence remained consistent and showed a gradient trend, and high-grade enhancement provided a clearer risk signal¹⁴. Differences in enhancement were not fully explained by stenosis rate, indicating that luminal geometric changes are insufficient to summarize the mechanisms of recurrence, and the biological activity within the plaque wall is closer to the event-triggering link¹⁵. Enhancement signal may reflect increased plaque permeability and contrast-agent uptake¹⁶, suggesting more active biological processes within the plaque wall. These changes can promote tissue factor expression and platelet adhesion, increasing the probability of focal thrombosis formation and artery-to-artery embolism¹⁷. Therefore, recurrence in the responsible vascular territory may occur even when stenosis remains at a moderate level. Intermediate-grade enhancement did not produce equally clear risk separation, suggesting that enhancement may have a threshold feature, or that the discriminative ability of grading is limited when early inflammatory burden is mild. Previous vessel wall imaging studies have mostly regarded enhancement as a marker of symptomatic plaque activity and associated it with recent events¹⁸, but results are not fully consistent across cohorts with different stenosis spectra and different examination time points¹⁹. Studies evaluating recurrent stroke risk have also suggested that plaque enhancement or enhancement ratio may be associated with subsequent ischemic events, although differences in scan timing, stenosis severity, and outcome attribution limit direct comparison across cohorts. This cohort focused on mild-to-moderate stenosis and handled exposure timing through left-truncated survival analysis, which helps place the predictive meaning of enhancement grading in the most common clinical gray-zone population. This stratification suggests that patients with high-grade enhancement may require closer follow-up and target-achieving management, and is clinically operable.

The incremental value of enhancement grading is also reflected in explaining recurrence heterogeneity among patients with mild-to-moderate stenosis. In baseline characteristics, higher enhancement grade changed in the same direction as lipid level and stenosis burden, suggesting that lipid metabolism and structural burden may jointly participate in plaque activation; however, in multivariable analysis, enhancement could still distinguish risk, indicating that the intramural inflammatory microenvironment and plaque surface stability are not fully determined by stenosis rate. During the mild-to-moderate stenosis stage, positive remodeling and collateral compensation are common²⁰, and the luminal proportion has limited ability to reflect fibrous cap integrity, endothelial dysfunction imbalance, and local coagulation activation. Enhancement is more like an imaging surrogate indicator of these proximal mechanisms. The continued association between lipid level and recurrence suggests that residual risk may be expressed through lipid-driven inflammatory pathways as increased enhancement and translated into embolic risk, leading to events in some patients despite baseline prescription of antiplatelet therapy and statin therapy. The absence of a clear independent signal for baseline medication regimens should be interpreted cautiously. It may reflect convergence of baseline prescriptions, difficulty in quantifying longitudinal adherence, changes in treatment intensity during follow-up, or insufficient statistical power due to limited event numbers, rather than indicating that antiplatelet or statin therapy has no effect on recurrence. Previous studies on recurrence in mild-to-moderate stenosis suggest that stratification based solely on stenosis severity tends to underestimate high-risk individuals²¹. Vessel wall imaging phenotypes can supplement luminal information, and the results of this cohort support using enhancement grading as one of the core bases for clinical communication and follow-up strategies. Stenosis rate remained associated with risk, suggesting that structural burden and plaque activity coexist; in moderate stenosis, recurrence is more inclined toward a microembolism mechanism. Outcomes attributed to the responsible vascular territory can better reflect the value of enhancement and facilitate stratified management.

Translational applicability depends on the reproducibility of exposure measurement and the robustness of the inference chain. Enhancement grading was determined using a uniform reference structure. The semiquantitative three-grade system based on comparison with the pituitary stalk was selected because it can be applied directly in routine clinical HR-MRI without additional post-processing and is consistent with previously reported qualitative or semiquantitative enhancement classifications. Double reading was completed under blinding to outcomes, and reproducibility was ensured through consistency evaluation, reducing the interference of subjective differences on risk estimation²². Follow-up was completed within pre-specified time windows, and outcome determination emphasized consistency with the responsible vascular territory, allowing a more direct correspondence between enhancement as a responsible plaque phenotype and subsequent recurrence, and avoiding dilution of the association by mixing events of different mechanisms. Analytically, the index event was used as the time scale and left truncation was applied, aligning the time of entry into the risk set with the timing of imaging exposure acquisition and reducing time bias introduced by recurrence before imaging completion. The proportional hazards assumption test showed no obvious violation. Missing covariates were handled by multiple imputation and supplemented with complete-case sensitivity analysis. Under a stricter definition of recurrent ischemic stroke, the direction of conclusions remained consistent, supporting the stability of the association between enhancement grading and true ischemic recurrence. Previous vessel wall imaging studies varied substantially in imaging protocols, enhancement criteria, and imaging endpoint definitions, limiting cross-cohort comparability²³. Using standardized grading combined with a rigorous time-to-event strategy helps apply enhancement grading to high-risk identification, follow-up intensity arrangement, and risk communication in patients with mild-to-moderate stenosis, promoting a shift from luminal assessment to intramural phenotype assessment. If grading error is mainly nondifferential, it usually biases the true effect toward the null, and the ability of high-grade enhancement to still separate risk is more convincing. Using the pituitary stalk as the reference can reduce bias introduced by individual differences in enhancement and facilitates reuse under different scanners and sequence conditions²⁴. A quantitative signal intensity ratio may further improve objectivity and reproducibility, especially in multicenter studies. However, this retrospective cohort was not prospectively optimized for quantitative enhancement measurement, and quantitative ratios may be influenced by region-of-interest placement, sequence parameters, coil sensitivity, and signal normalization. Future studies should combine semiquantitative grading with quantitative enhancement ratio, wall thickness, remodeling index, and plaque component analysis.

Limitations mainly arise from the retrospective single-center cohort design and the limited number of events, and selection bias and unmeasured confounding may still exist. Because the study was observational and retrospective, the findings should be interpreted as associations rather than causal effects. The single-center setting may limit generalizability, particularly to non-Asian populations in which the prevalence, vascular distribution, and biological characteristics of intracranial atherosclerosis may differ. Only patients who completed HR-MRI within 14 days after the index event were included, which may have introduced selection bias by excluding patients with very mild symptoms who did not undergo vessel wall imaging and patients with severe conditions who could not complete MRI. Insufficient matching between the number of covariates and the number of events limited more complex stratified and interaction analyses, and some adjusted results are more appropriate to be regarded as sensitivity support. The fully adjusted Cox model included multiple covariates relative to the number of recurrent events and may therefore be susceptible to overfitting; the simplified adjusted model was used as the primary model, whereas the fully adjusted model was interpreted as supportive sensitivity analysis. Enhancement was a semi-quantitative grade and was not combined with dynamic quantitative parameters, wall-thickness remodeling, or finer phenotypes such as plaque components; mechanistic interpretation still relied mainly on indirect inference. Imaging was mostly completed in the early acute phase, and repeated imaging during follow-up was lacking to verify consistency between changes in enhancement and changes in risk. Medication variables were mainly based on baseline prescriptions, and longitudinal antiplatelet adherence, statin dose adjustment or interruption, and LDL-C target achievement during follow-up were not systematically captured, which may affect the relationship between medication factors and recurrence.

In patients with mild-to-moderate intracranial arterial stenosis receiving standardized secondary prevention medication treatment, HR-MRI plaque enhancement grading showed a stable grade-related relationship with the risk of recurrent ischemic events during follow-up. High-grade enhancement indicated a higher recurrence tendency, and this association remained evident after adjustment for stenosis severity and common clinical risk factors. The vessel wall inflammatory activity and intimal microvessel-related changes reflected by enhancement grading may represent an imaging surrogate indicator of plaque vulnerability and embolic tendency, thereby explaining the possible mechanistic basis for recurrence during the mild-to-moderate stenosis stage. This grading system is reproducible and operable and may provide incremental risk stratification beyond luminal assessment for populations with mild-to-moderate intracranial atherosclerosis, helping to identify high-risk individuals who may require more intensive follow-up and more intensified target-achieving management of risk factors. Future multicenter prospective studies are needed, combined with quantitative enhancement and more comprehensive plaque phenotype indicators, to validate its predictive performance and clarify its application value in clinical decision pathways.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
3.0T MRI system	Siemens Healthineers	https://www.siemens-healthineers.com/en-ca/magnetic-resonance-imaging/3t-mri-scanner/magnetom-skyra	MAGNETOM Skyra 3T MRI system; Intracranial high-resolution vessel wall magnetic resonance imaging (MRI)
32-channel head-neck combined coil	Siemens Healthineers	https://medicine.tulane.edu/centers-institutes/taic/facilities	32-channel head/neck coil (Skyra-compatible configuration); Signal acquisition for high-resolution intracranial vessel wall imaging
Disposable syringe set for MR injector	Bayer / MEDRAD	SSQK 65/115vs; https://www.radiology.bayer.com/it/products/medrad-spectris-solaris	Qwik-Fit Syringe Kit; Contrast-agent and saline delivery for MR imaging
Gadopentetate dimeglumine injection	Bayer AG	https://www.bayer.com/sites/default/files/magnevist-smpc-may-2016.pdf	Magnevist 0.5 mmol/mL (469 mg/mL); Contrast-enhanced vessel wall imaging at 0.1 mmol/kg
MR-compatible power injector	Bayer	https://www.radiology.bayer.com/products/medrad-spectris-solaris	MEDRAD Spectris Solaris EP MR Injection System; Automated contrast-agent injection at 2.0 mL/s during contrast-enhanced MRI
Picture archiving and communication system (PACS)	Zhejiang Laida Information Technology Co., Ltd.	https://www.qzszyy.com/home/info/15410?type=003	PACS system; Image retrieval, image verification, and follow-up record checking
R statistical software	R Foundation for Statistical Computing	R version 4.3.0, RRID:SCR_001905 https://cran.r-project.org/bin/windows/base/old/4.3.0/	Statistical analysis