Beyond plaque morphology: a multimodal imaging model combining structural vulnerability and functional reserve predicts stroke recurrence in symptomatic intracranial atherosclerosis

Introduction

Intracranial atherosclerotic disease (ICAD) represents a predominant etiology of ischemic stroke globally, especially within Asian populations, and imposes a significant healthcare burden (1). Despite intensive guideline-directed medical therapy, patients with symptomatic ICAD continue to exhibit a substantial long-term risk of stroke recurrence, exceeding 25% within two years (2). This persistent clinical challenge highlights the urgent need for more precise tools to identify individuals at the highest risk for recurrent ischemic events and to unravel the underlying pathophysiology.

High-resolution vessel wall imaging (HR-VWI) has revolutionized the assessment of ICAD by enabling detailed visualization and quantification of plaque morphology beyond mere luminal stenosis. Quantitative markers, including plaque burden and enhancement ratio, have emerged as robust imaging biomarkers closely associated with inflammatory activity, intraplaque neovascularization, and overall plaque instability (3-5). In addition, HR-VWI has been increasingly applied to characterize the relationship between intracranial plaque burden and cerebral small vessel disease manifestations, such as white matter hyperintensity, further highlighting its value in comprehensive cerebrovascular risk assessment (6). Recent investigations confirm that these plaque characteristics independently predict recurrent stroke (4,7,8). However, their standalone predictive performance remains suboptimal. Not all patients harboring high-risk plaque features experience recurrence, whereas some with less pronounced plaque abnormalities suffer recurrent events (9,10). This observation strongly suggests that plaque morphology alone—while critical—captures only the structural dimension of risk, leaving a significant portion of recurrence mechanisms unexplained (9,11,12).

Compromised cerebral hemodynamics constitutes another pivotal pathway to recurrent ischemia (13-15). Previous studies have assessed collateral circulation or perfusion deficits using isolated modalities such as computed tomography angiography (CTA), CT perfusion (CTP), or magnetic resonance (MR) perfusion (16,17). However, these approaches are often fragmented and single-modal, focusing separately on arterial collaterals, tissue perfusion, or venous drainage. Such evaluations fail to capture the integrated, multi-tiered nature of cerebral hemodynamic compensation. Recently, the cerebral collateral cascade (CCC) model has been proposed as a comprehensive framework that unifies arterial inflow, tissue-level perfusion efficiency, and venous outflow into a holistic assessment of circulatory reserve (18). Although validated in acute ischemic stroke for predicting infarct progression and clinical outcome (19,20), its application to long-term recurrence risk in chronic symptomatic ICAD remains largely unexplored.

Critically, no study to date has integrated HR-VWI-assessed plaque vulnerability with a multidimensional hemodynamic model like the CCC to predict long-term stroke recurrence in ICAD. Existing prediction models rely predominantly on either plaque characteristics (21) or isolated perfusion parameters (22,23), thereby neglecting the potential synergistic or antagonistic interplay between local structural instability and global functional compensatory capacity. Furthermore, robust evidence focusing specifically on long-term recurrence beyond 12 months remains scarce, limiting the applicability of current imaging biomarkers to chronic risk stratification.

We therefore propose and aim to test a “dual-hit” pathophysiological hypothesis: stroke recurrence in ICAD may result from an initial insult from a vulnerable plaque (structural risk), the clinical impact of which may be influenced by the brain’s integrated hemodynamic reserve (functional vulnerability). When robust CCC exists, the brain can tolerate embolic showers or perfusion fluctuations; when CCC is depleted, even minor insults may precipitate infarction.

Consequently, we hypothesized that the synergistic integration of HR-VWI-derived plaque features reflecting local structural instability and the CCC framework reflecting integrated hemodynamic reserve could provide complementary and additive information to improve prediction of long-term stroke recurrence. This study aimed to develop a novel multimodal imaging model integrating structural and functional risk domains within a unified framework. In addition, we sought to evaluate whether this approach could enhance risk stratification and provide in vivo imaging evidence supporting an integrated structure-function paradigm in symptomatic ICAD. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0716/rc).

Methods

General information

This was a retrospective, multicenter, observational cohort study. Consecutive patients with symptomatic ICAD admitted between September 2018 and October 2024 at the First and Fourth Affiliated Hospitals of Soochow University were screened for inclusion. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The First Affiliated Hospital of Soochow University, Suzhou, China (Approval No. 2025-819), and the Ethics Committee of The Fourth Affiliated Hospital of Soochow University, Suzhou, China (Approval No. 2025-251270). Given the retrospective nature of the study, the requirement for informed consent was waived by the Ethics Committees.

Inclusion criteria were: (I) age >18 years; (II) first-ever acute ischemic stroke in the unilateral anterior circulation confirmed by DWI; (III) ≥50% atherosclerotic stenosis of the culprit intracranial artery (intracranial internal carotid artery or M1/M2 segment of the middle cerebral artery) confirmed by CTA or MRA; (IV) completion of a standardized baseline multimodal CT protocol at admission, including non-contrast CT (NCCT), CTA, and CTP; (V) adequate CTA image quality permitting reliable evaluation of collateral circulation and venous outflow, with uniform visualization of the superior sagittal, transverse, and sigmoid sinuses; (VI) performance of three-dimensional high-resolution vessel wall imaging (3D HR-VWI) within 7 days after the index ischemic event; (VII) receipt of standardized secondary prevention medical therapy (antiplatelet agents and statins) during follow-up, with no history of endovascular intervention or extracranial/intracranial bypass surgery.

Exclusion criteria comprised: (I) non-atherosclerotic vascular lesions (e.g., aneurysm, arterial dissection, vasculitis, Moyamoya disease); (II) coexisting ≥50% stenosis in other intracranial or extracranial carotid arteries; (III) evidence of a potential cardiogenic embolism source(e.g., atrial fibrillation, patent foramen ovale); (IV) incomplete clinical or imaging data; (V) poor image quality precluding reliable quantitative or qualitative analysis; (VI) complete occlusion of the target vessel.

Baseline demographic information, vascular risk factors, admission National Institutes of Health Stroke Scale (NIHSS) score, key laboratory parameters, and in-hospital medication data were systematically extracted from the electronic medical record system of each participating institution. The study workflow, detailing patient screening and selection, is presented in Figure 1.

Figure 1 Flowchart of this study population. CTA, computed tomography angiography; CTP, computed tomography perfusion; HR-VWI high-resolution vessel wall imaging.

Imaging protocol

All patients underwent standardized multimodal CT and HR-MRI examinations to ensure consistency and comparability of data.

CT imaging protocol: a one-stop multimodal CT examination was performed using a 256-slice CT scanner. The protocol included: Non-contrast CT: 1.25-mm slice thickness; CT perfusion: Whole-brain coverage, tube voltage 80 kV, tube current 120 mA; CT angiography: coverage from the aortic arch to the cranial vault, tube voltage 120 kV, tube current 250 mA, with intravenous contrast administered at a rate of 5.0 mL/s.

MRI protocol: MRI was conducted on a 3.0 T scanner. The protocol included DWI, TOF-MRA, and pre- and post-contrast 3D T1-weighted HR-VWI. HR-VWI was acquired with an isotropic voxel size of 0.6 mm × 0.6 mm × 0.6 mm. Post-contrast imaging was initiated 5 minutes after intravenous administration of a gadolinium-based contrast agent.

Detailed acquisition parameters for all sequences are provided in Table S1.

Imaging analysis

All imaging analyses were performed independently by two physicians, each with over five years of experience in neuroimaging. Both raters were blinded to clinical outcomes and group assignments. Discrepancies were resolved by consensus. The reliability of measurements was assessed using intraclass correlation coefficients (ICC), with ICC >0.80 considered indicative of good agreement.

Plaque characteristic analysis (HR-VWI)

The culprit plaque was defined as the atherosclerotic lesion responsible for the highest degree of stenosis within the vascular territory corresponding to the acute infarct. Quantitative plaque metrics, including percentage stenosis, plaque burden, plaque volume, plaque enhancement ratio, remodeling index, and eccentricity index, were measured using dedicated imaging software (RadiAnt DICOM Viewer; 3D Slicer). Detailed calculation formulas and definitions for all plaque parameters are provided in the Appendix 1.

CCC assessment

CCC was evaluated according to the integrated framework proposed by Faizy et al. (18), encompassing three dimensions: arterial collaterals, tissue-level perfusion, and venous outflow. Arterial collaterals were assessed on CTA source images using the modified Tan score. A score ≥2 was defined as good collateral flow. Tissue-level perfusion was quantified using the hypoperfusion intensity ratio (HIR), automatically generated by uAI Discover Cerebral CTP (United Imaging Intelligence, Shanghai, China), HIR ≤0.4 indicated good tissue-level perfusion (24,25). Venous outflow was assessed on venous-phase CTA images based on the opacification of three major cortical veins—the vein of Labbé, the sphenoparietal sinus, and the superficial middle cerebral vein. Each vein was scored from 0 to 2, yielding a total score of 0–6. A total score ≥3 was considered good venous outflow (26). Patients were subsequently classified into three CCC categories based on this three-tiered assessment: CCC+ (all three components good), CCC− (all three components poor), and CCCmixed (any other combination).

Perfusion parameters

Quantitative perfusion parameters were automatically processed using uAI Discover Cerebral CTP. The ischemic core volume was defined as the tissue volume with a relative cerebral blood flow (rCBF) <30%. The hypoperfusion volume was defined as the tissue volume with a Tmax >6 seconds.

Outcome assessment and follow-up

The primary study outcome was recurrent ischemic stroke during follow-up. Recurrent ischemic stroke was defined as a new focal neurological deficit occurring ≥24 hours after the index event and confirmed by neuroimaging demonstrating a new cerebral infarction within the vascular territory of the affected artery or in a different vascular territory (27). Progression of the index infarction was not considered a recurrent event. All patients were required to complete a minimum of 12 months of follow-up. All endpoint events were adjudicated independently by two neurologists blinded to the baseline imaging characteristics.

Statistical analysis

Continuous variables were expressed as mean ± standard deviation (SD) or median (IQR), and categorical variables as frequency (percentage). Between-group comparisons were performed using the t-test, Mann–Whitney U test, Chi-square test, or Fisher’s exact test, as appropriate. Univariate Cox proportional hazards regression was performed to identify candidate predictors of stroke recurrence. To address the issue of sparse events and potential bias in maximum likelihood estimation, all subsequent multivariable analyses were conducted using Firth’s penalized-likelihood Cox regression, which provides more stable hazard ratio (HR) estimates and confidence intervals (CI) in small-to-moderate sample sizes with limited outcomes. Variables showing an association with recurrence at a liberal threshold of P<0.10 in univariate analysis were considered for inclusion in the initial multivariate model. Internal validation was conducted by calculating shrinkage factors and performing 1,000 bootstrap resamples to evaluate and correct for model overfitting. Results were reported as HR with 95% CI. The discrimination performance of prediction models was evaluated using receiver operating characteristic (ROC) curves analysis and quantified by the area under the curve (AUC). We compared the performance of models using the DeLong test. The performance of single-modality (plaque-only or CCC-only) and integrated multimodal prediction models was evaluated using ROC curves, clinical decision curves, calibration curves, and DeLong’s test. Differences in mRS scores among the three CCC groups were assessed using the Kruskal-Wallis test. All statistical analyses were conducted using R software (version 4.4.2). A two-sided P<0.05 was considered statistically significant.

Results

Baseline characteristics of the study population

During the study period, 235 patients with acute ischemic stroke attributable to anterior circulation ICAD were screened. Of these, 89 patients met all inclusion criteria and completed the required follow-up, forming the final analysis cohort (Figure 1). The mean age of the cohort was 57.1±11.8 years, and 61.8% (55/89) were male. The median follow-up duration was 30.0 months. During this period, 20 patients experienced a recurrent ischemic stroke, resulting in a cumulative recurrence rate of 22.5% (20/89). No significant differences were observed between the recurrence and non-recurrence groups in age, sex, vascular risk factors, admission NIHSS scores, or in-hospital medication profiles (Table 1, all P>0.05).

Imaging features of culprit plaques and cerebral perfusion

Quantitative analysis of culprit plaques using HR-VWI revealed significant differences in vulnerability markers between groups (Table 2). Patients with recurrent stroke demonstrated a significantly higher median plaque burden (89.63% vs. 83.76%, P=0.004) and a markedly elevated median plaque enhancement ratio (2.16 vs. 1.65, P=0.002) compared to the non-recurrence group. Representative examples of plaque characteristics are shown in Figure 2. In contrast, no significant between-group differences were found in other morphological characteristics, including degree of stenosis, plaque volume, remodeling index, or eccentricity index (all P>0.05).

Figure 2 Neuroimaging findings in a representative patient with recurrent ischemic stroke. A patient presented with left-sided weakness for 3 days. (A) Diffusion-weighted imaging reveals a hyperintense acute infarct in the right middle cerebral artery territory. (B) On the 3D-TOF MRA, severe stenosis was demonstrated in the right M1 segment of the MCA (white arrow). (C,D) High-resolution vessel wall imaging identifies the culprit plaque (white arrow) with significant enhancement on postcontrast scans. (E) CT angiography source images show poor pial arterial collaterals (red arrows) in the affected hemisphere. (F) Good venous outflow (green arrow) is observed in the unaffected hemisphere, contrasted with poor venous outflow (red arrow) in the symptomatic hemisphere. (G) CT perfusion map indicates unfavorable tissue-level collaterals, evidenced by an elevated hypoperfusion intensity ratio of 0.42. (H) Diffusion-weighted imaging shows ipsilateral ischemic stroke recurrence. 3D-TOF MRA, three-dimensional time-of-flight magnetic resonance angiography; CT, computed tomography; CTA, computed tomography angiography; MCA, middle cerebral artery.

Assessment of baseline cerebral perfusion parameters indicated more severe hemodynamic compromise in patients who later experienced a recurrence. The median ischemic core volume (rCBF <30%) was significantly larger in the recurrence group (0.69 vs. 0.00 mL, P<0.001). Similarly, the median hypoperfusion volume (Tmax >6 s), reflecting the burden of critically hypoperfused tissue, was substantially greater in the recurrence group (88.82 vs. 3.02 mL, P<0.001) (Table 2).

CCC profile and clinical functional outcomes

A comprehensive, three-dimensional evaluation of the CCC revealed profound deficiencies at every level in patients who suffered a recurrence (Table 2). Arterial collaterals: only 25.0% (5/20) of patients in the recurrence group had well-developed arterial collaterals (modified Tan score ≥2), compared to 65.2% (45/69) in the non-recurrence group (P=0.003). Tissue-level perfusion: the median HIR, a marker of tissue-level collateral efficiency, was significantly higher in the recurrence group (0.38 vs. 0.00, P<0.001), indicating poorer microcirculatory compensation. Venous outflow: venous drainage was also impaired, with the median cortical venous opacification score being significantly lower in the recurrence group (2.00 vs. 4.00, P<0.001). When integrated into the CCC framework, the recurrence rate differed dramatically across CCC categories (P<0.001). Patients classified as CCC− had the highest recurrence rate (64.3%, 9/14), whereas those with a CCC+ profile had the lowest rate (5.3%, 2/38). Furthermore, a worse integrated CCC profile at baseline was strongly associated with poorer 90-day functional outcomes, as evidenced by significantly higher mRS scores in the CCC− group (P<0.001, Figure 3).

Figure 3 Functional outcomes at 90 days stratified by cerebral collateral cascade status. The cerebral collateral cascade framework classified 38 patients as CCC+, 37 as CCCmixed, and 14 as CCC−. A progressive deterioration in CCC status was associated with a shift toward higher (less favorable) mRS scores, indicating poorer functional recovery. CCC, cerebral collateral cascade; mRS, modified Rankin scale.

Intra- and inter-observer agreement for all quantitative imaging measurements was excellent, with ICCs exceeding 0.80 for all assessed metrics, confirming the reliability of our imaging analysis (Table S2).

Construction and validation of the stroke recurrence prediction model

In univariate Cox regression analysis, plaque enhancement ratio, plaque burden, ischemic core volume, Tmax >6 s hypoperfusion volume, and CCC− profile were all significantly associated with stroke recurrence (all P<0.10) and were carried forward for multivariable modeling. The final parsimonious multivariable Firth-corrected Cox model identified three independent predictors of stroke recurrence (Table 3). Each 1-unit increase in plaque enhancement ratio was associated with a 152% higher risk of recurrence [adjusted hazard ratio (aHR) =2.52, 95% CI: 1.09–4.26, P=0.005]. Each 1% increase in plaque burden corresponded to a 7% increase in recurrence risk (aHR =1.07, 95% CI: 1.02–1.15, P=0.013). Patients with a CCC− status had a 3.39-fold higher risk of recurrence compared to those with CCC+ or CCCmixed profiles (aHR =3.39, 95% CI: 1.22–10.54, P=0.005).

Concurrently, a multivariable risk model incorporating CTA-assessed collateral circulation status was developed (Table S3), in which this variable was no longer identified as an independent predictor (P>0.05). Based on these findings, the CCC model was ultimately selected as an integrated indicator of collateral functional status for the construction of the combined prediction model. Compared with CTA-based arterial collateral assessment, the CCC framework provided additional prognostic information for stroke recurrence prediction.

Model performance, calibration

The discriminative performance of different prediction strategies was evaluated using ROC analysis (Figure 4). The AUC for plaque burden, plaque enhancement ratio, and CCC− profile as independent predictors were 0.711, 0.728, and 0.689, respectively. The multimodal integrated model, combining all three independent predictors, demonstrated improved discriminative performance, with an AUC of 0.829 (95% CI: 0.728–0.930). The DeLong test confirmed that this integrated model outperformed the best single-modality model (plaque enhancement ratio + plaque burden, AUC =0.735, P<0.05).

Figure 4 Predictive performance of key variables for stroke recurrence. (A,B) Comparison of culprit plaque enhancement ratio and plaque burden between patients with and without recurrent stroke. Data are presented as median with interquartile range; P<0.05. (C) ROC curves demonstrate the discriminatory ability of individual features and their combination for predicting stroke recurrence. The combined model (red curve) integrating plaque burden, plaque enhancement ratio, and poor cerebral collateral cascade status achieved the highest predictive performance. *, P<0.05. ROC, receiver operating characteristic.

Internal validation via 1000 bootstrap resamples yielded a shrinkage factor of 0.807 and an optimism-corrected AUC of 0.802, suggesting satisfactory internal performance after bootstrap correction. Calibration plots for 1-, 2-, and 3-year recurrence probabilities showed good agreement between predicted and observed event rates across the spectrum of risk (Figure 5A-5C). Decision curve analysis (Figure 5D-5F) demonstrated that, across a wide range of clinically relevant risk thresholds (10–50%), the use of the multimodal prediction model to guide clinical decisions provided a greater net benefit than strategies of treating all patients or treating none.

Figure 5 Calibration and decision curves for the predictive model. (A-C) Calibration curves illustrating the agreement between predicted and observed probabilities of stroke recurrence at 1, 2, and 3 years. The curves demonstrate good overall calibration across all time points. (D-F) DCA show that, within the clinically relevant threshold range, the prediction model provides a greater net clinical benefit compared with the treat-all and treat-none strategies. *, P<0.05. DCA, decision curve analysis.

Discussion

This study establishes a novel neuroimaging paradigm that moves beyond the isolated assessment of either local plaque pathology or global brain perfusion. By innovatively integrating quantitative plaque vulnerability from HR-VWI with the comprehensive hemodynamic reserve profile from the CCC model, we developed a multimodal prediction model with favorable performance following internal validation.

Our findings reinforce and extend the established role of HR-VWI in risk stratification. The strong predictive value of the plaque enhancement ratio, a direct imaging correlate of intraplaque inflammation and neovascularization, aligns with prospective data linking plaque enhancement to adverse outcomes (28) and recent meta-analytic conclusions (29). Similarly, the independent association of plaque burden with recurrence risk corroborates prior work (30), underscoring its role as a morphological marker of atherosclerotic disease volume and local hemodynamic disturbance. Critically, however, our results also demonstrate the inherent limitations of a purely structural approach. As seen in our cohort and reported elsewhere (21), even robust identification of “high-risk” plaques yields imperfect predictive performance. This suggests that reliance on a single structural biomarker cannot fully capture the multifactorial nature of ICAD-related recurrence, particularly given the complexity of systemic hemodynamic regulation and differences in tissue tolerance. These limitations provide the rationale for incorporating functional compensation assessment into the predictive framework.

To address this critical gap, our study is the first to apply the integrated CCC framework to long-term outcomes prediction in chronic ICAD. Unlike conventional, fragmentary assessments (13), the CCC model provides a holistic, multi-tiered evaluation of the entire compensatory pathway (18). We demonstrate that a poor CCC profile was independently associated with recurrence risk (aHR =3.39). However, the relatively wide confidence interval (95% CI: 1.22–10.54) indicates uncertainty regarding the precise magnitude of this association, likely reflecting the limited number of outcome events. From a hemodynamic standpoint, impaired CCC signifies a depletion of circulatory reserve across its key components: deficient arterial inflow limits blood flow redistribution, while inefficient tissue-level perfusion (reflected by a high HIR) jeopardizes tissue viability and reflects impaired hemodynamic reserve (24). Although HIR was initially developed as a perfusion marker in the acute stroke setting, it may also serve as an indicator of tissue-level compensatory capacity. Within the CCC framework, an elevated HIR reflects persistent hypoperfusion despite collateral recruitment and may therefore identify patients with increased vulnerability to future ischemic events. Furthermore, compromised venous outflow disrupts metabolic clearance and perfusion pressure homeostasis. This systemic hemodynamic frailty creates a state of heightened vulnerability where even minor embolic showers or physiological blood pressure fluctuations can precipitate infarction (31,32). Our findings thus provide direct, in vivo imaging evidence elucidating the mechanisms of hemodynamic stroke (15) and explain why some patients remain prone to recurrence despite standard medical therapy, suggesting that the underlying vulnerability lies in systemic compensatory failure rather than merely in focal plaque pathology.

The central contribution of this work lies in the quantitative evaluation of the synergistic value derived from integrating structural and functional imaging domains. While previous studies have established the independent relevance of either plaque (4,29) or perfusion (11,24) biomarkers, our analysis demonstrates—through formal model comparison and reclassification metrics—that their combination yields a substantial and significant improvement in predictive accuracy (AUC increase >0.09). Although a formal test for multiplicative interaction was not significant, pointing to an additive model in our cohort, this marked improvement in discriminative performance is consistent with a pathophysiologically coherent “dual-hit” model of recurrence. In this paradigm, the first hit originates from the vulnerable plaque (artery-to-artery embolism or flow-limiting stenosis), while the second, decisive hit is the failure of the CCC to effectively clear emboli or compensate for reduced perfusion. When these mechanisms occur simultaneously, recurrence risk increases exponentially. This provides in vivo imaging evidence supporting Caplan and Henneric’s “obstacle to embolic clearance” hypothesis (32) and reinforces the recently emphasized interactive mechanism linking embolism and hypoperfusion (12). Our predictive model operationalizes this concept, translating the theoretical “obstacle to embolic clearance” hypothesis into a quantifiable clinical tool. Nevertheless, the present model was evaluated using internal bootstrap validation only. Therefore, its predictive performance and generalizability require confirmation in larger independent prospective cohorts before broader clinical application.

This refined risk stratification has direct clinical implications for personalized medicine in ICAD. A single, coordinated multimodal imaging evaluation (HR-VWI + CTP/CTA) can now simultaneously inform on both the aggressiveness of the plaque and the resilience of the brain. For patients identified as “dual high-risk” standard medical therapy may be inadequate. This subgroup may warrant more tailored management strategies, such as intensified antithrombotic regimens, meticulous blood pressure optimization tailored to individual perfusion thresholds, or prioritization for emerging therapies aimed at augmenting collateral flow or plaque stabilization.

Several limitations of this study should be acknowledged. First, this was a retrospective observational study and is therefore inherently susceptible to selection bias, information bias, and residual confounding. Inclusion required complete HR-VWI, CTA, and CTP examinations, adequate image quality, and longitudinal follow-up, which may have preferentially selected patients who were clinically stable enough to undergo comprehensive imaging evaluation. Consequently, patients with severe neurological deterioration, incomplete imaging studies, poor image quality, or inability to complete follow-up may have been underrepresented. Although consecutive patient enrollment from two independent stroke centers, predefined eligibility criteria, blinded imaging assessment, and multivariable adjustment were applied to minimize bias, these measures cannot completely eliminate the inherent limitations of retrospective research. Second, despite the use of Firth penalized Cox regression and bootstrap internal validation, the relatively limited sample size and number of recurrent events may still have resulted in some degree of model optimism. To reduce the risk of overfitting, the final multivariable model was intentionally restricted to a small number of independent predictors. Nevertheless, the predictive performance of the proposed model should be interpreted cautiously until confirmed in larger cohorts. Third, although the CCC framework was originally designed as a three-tier classification system, a binary categorization was adopted in the final multivariable analysis to improve model stability given the limited number of outcome events. While this approach reduced sparse-data bias, it may also have resulted in some loss of prognostic information. Future studies with larger sample sizes should evaluate the full three-tier CCC framework. Finally, perfusion parameters incorporated into the CCC framework were originally developed and validated primarily for acute ischemic stroke assessment rather than long-term recurrence prediction. Although these metrics were used in the present study as surrogate markers of cerebral hemodynamic reserve and collateral efficiency, their prognostic value for chronic recurrence risk warrants further investigation.

Conclusions

This study developed a multimodal imaging prediction model that integrates plaque vulnerability and cerebral hemodynamic reserve. By providing a unified structure-function assessment, the model demonstrated improved performance for long-term stroke recurrence risk stratification compared with plaque-based assessment alone and provided in vivo support for an integrated “dual-hit” pathophysiological framework in symptomatic ICAD. These findings advocate for the adoption of comprehensive multimodal neuroimaging to guide precision risk assessment and may facilitate more individualized risk assessment in this high-risk population.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0716/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0716/dss

Funding: This work was supported by the Science and Technology Program of Suzhou (No. 2025C224).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0716/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The First Affiliated Hospital of Soochow University, Suzhou, China (Approval No. 2025-819), and the Ethics Committee of The Fourth Affiliated Hospital of Soochow University, Suzhou, China (Approval No. 2025-251270). Given the retrospective nature of the study, the requirement for informed consent was waived by the Ethics Committees.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Arenillas JF. Intracranial atherosclerosis: current concepts. Stroke 2011;42:S20-3. [Crossref] [PubMed]
2. Chen LH, Spagnolo-Allende A, Yang D, Qiao Y, Gutierrez J. Epidemiology, Pathophysiology, and Imaging of Atherosclerotic Intracranial Disease. Stroke 2024;55:311-23. [Crossref] [PubMed]
3. Ran Y, Wang Y, Zhu M, Wu X, Malhotra A, Lei X, Zhang F, Wang X, Xie S, Zhou J, Zhu J, Cheng J, Zhu C. Higher Plaque Burden of Middle Cerebral Artery Is Associated With Recurrent Ischemic Stroke: A Quantitative Magnetic Resonance Imaging Study. Stroke 2020;51:659-62. [Crossref] [PubMed]
4. Lv Y, Ma X, Zhao W, Ju J, Yan P, Li S, Xue Y, Sui Y, Shao S, Sun Q, Qiu C. Association of plaque characteristics with long-term stroke recurrence in patients with intracranial atherosclerotic disease: a 3D high-resolution MRI-based cohort study. Eur Radiol 2024;34:3022-31. [Crossref] [PubMed]
5. Shi Z, Li J, Zhao M, Peng W, Meddings Z, Jiang T, Liu Q, Teng Z, Lu J. Quantitative Histogram Analysis on Intracranial Atherosclerotic Plaques: A High-Resolution Magnetic Resonance Imaging Study. Stroke 2020;51:2161-9. [Crossref] [PubMed]
6. Li M, Song X, Wei Q, Wu J, Wang S, Liu X, Guo C, Gao Q, Zhou X, Niu Y, Guo X, Zhao X, Chen L. The relationship between intracranial atherosclerosis and white matter hyperintensity in ischemic stroke patients: a retrospective cross-sectional study using high-resolution magnetic resonance vessel wall imaging. Quant Imaging Med Surg 2024;14:6002-14. [Crossref] [PubMed]
7. Sun B, Wang L, Li X, Zhang J, Zhang J, Liu X, Wu H, Mossa-Basha M, Xu J, Zhao B, Zhao H, Zhou Y, Zhu C. Intracranial Atherosclerotic Plaque Characteristics and Burden Associated With Recurrent Acute Stroke: A 3D Quantitative Vessel Wall MRI Study. Front Aging Neurosci 2021;13:706544. [Crossref] [PubMed]
8. Zhao J, Zhao X, Peng Z, Liu J. A model based on high-resolution magnetic resonance vessel wall imaging for predicting stroke recurrence in patients with symptomatic intracranial atherosclerosis. Quant Imaging Med Surg 2025;15:10422-32. [Crossref] [PubMed]
9. Lan L, Leng X, Ip V, Soo Y, Abrigo J, Liu H, Fan F, Ma SH, Ma K, Ip BY, Chan KL, Mok VC, Liebeskind DS, Wong KS, Leung TW. Sustaining cerebral perfusion in intracranial atherosclerotic stenosis: The roles of antegrade residual flow and leptomeningeal collateral flow. J Cereb Blood Flow Metab 2020;40:126-34. [Crossref] [PubMed]
10. Zhao Y, Yang J, Dai Y, Ma Y, Song Z, Hu C, Zhang Y. Quantitative plaque characteristics based on 3D high-resolution vessel wall imaging predict long-term recurrence in intracranial atherosclerotic stroke. Neuroradiology 2025;67:2309-19. [Crossref] [PubMed]
11. de Havenon A, Khatri P, Prabhakaran S, Yeatts SD, Peterson C, Sacchetti D, Alexander M, Cutting S, Mac Grory B, Furie K, Liebeskind DS, Yaghi S. Hypoperfusion Distal to Anterior Circulation Intracranial Atherosclerosis is Associated with Recurrent Stroke. J Neuroimaging 2020;30:468-70. [Crossref] [PubMed]
12. Feng X, Fang H, Ip BYM, Chan KL, Li S, Tian X, Zheng L, Liu Y, Lan L, Liu H, Abrigo J, Ma SH, Fan FSY, Ip VHL, Soo YOY, Mok VCT, Song B, Leung TW, Xu Y, Leng X. Cerebral Hemodynamics Underlying Artery-to-Artery Embolism in Symptomatic Intracranial Atherosclerotic Disease. Transl Stroke Res 2024;15:572-9. [Crossref] [PubMed]
13. Maguida G, Shuaib A. Collateral Circulation in Ischemic Stroke: An Updated Review. J Stroke 2023;25:179-98. [Crossref] [PubMed]
14. Liu Y, Li S, Tian X, Leung TW, Liu L, Liebeskind DS, Leng X. Cerebral haemodynamics in symptomatic intracranial atherosclerotic disease: a narrative review of the assessment methods and clinical implications. Stroke Vasc Neurol 2023;8:521-30. [Crossref] [PubMed]
15. Wegener S, Baron JC, Derdeyn CP, Fierstra J, Fromm A, Klijn CJM, van Niftrik CHB, Schaafsma JD. Hemodynamic Stroke: Emerging Concepts, Risk Estimation, and Treatment. Stroke 2024;55:1940-50. [Crossref] [PubMed]
16. Varga A, Di Leo G, Banga PV, Csobay-Novák C, Kolossváry M, Maurovich-Horvat P, Hüttl K. Multidetector CT angiography of the Circle of Willis: association of its variants with carotid artery disease and brain ischemia. Eur Radiol 2019;29:46-56. [Crossref] [PubMed]
17. Wang Y, Zhao X, Liu L, Soo YO, Pu Y, Pan Y, Wang Y, Zou X, Leung TW, Cai Y, Bai Q, Wu Y, Wang C, Pan X, Luo B, Wong KSCICAS Study Group. Prevalence and outcomes of symptomatic intracranial large artery stenoses and occlusions in China: the Chinese Intracranial Atherosclerosis (CICAS) Study. Stroke 2014;45:663-9. [Crossref] [PubMed]
18. Faizy TD, Mlynash M, Kabiri R, Christensen S, Kuraitis GM, Mader MM, Flottmann F, Broocks G, Lansberg MG, Albers GW, Marks MP, Fiehler J, Wintermark M, Heit JJ. The Cerebral Collateral Cascade: Comprehensive Blood Flow in Ischemic Stroke. Neurology 2022;98:e2296-306. [Crossref] [PubMed]
19. Huang L, Zhang H, Li W, Gong C, Jiang S, Li Z, Yuan J, Xu T, Chen Y, Zhang L, Wang Y. Cerebral collateral cascade associated with infarct growth rate in ischemic stroke patients undergoing endovascular treatment. Eur J Radiol 2025;193:112418. [Crossref] [PubMed]
20. Huang L, Jiang S, Gong C, Wu G, Guo J, Liu J, Yuan J, Wang Y, Xu T, Liu C, Chen S, Chen Y. Favorable Cerebral Collateral Cascades Improve Futile Recanalization by Reducing Ischemic Core Volume in Acute Ischemic Stroke Patients Undergoing Endovascular Treatment. Transl Stroke Res 2025;16:1689-97. [Crossref] [PubMed]
21. Yao W, Chen H, Huang K, Peng W, Zhang X, Yang D, Teng Z, Shen J, Yang J, Cheng X, Han Y, Zhu W, Wang J, Du J, Liu X. Atherosclerotic plaque evolution predicts cerebral ischemic events in patients with intracranial atherosclerosis: a multicentre longitudinal study using high-resolution MRI. Eur Radiol 2025;35:3238-48. [Crossref] [PubMed]
22. Sacchetti DC, Cutting SM, McTaggart RA, Chang AD, Hemendinger M, Mac Grory B, Siket MS, Burton T, Thompson B, Rostanski SK, Prabhakaran S, Willey JZ, Marshall RS, Elkind MS, Khatri P, Furie KL, Jayaraman MV, Yaghi S. Perfusion imaging and recurrent cerebrovascular events in intracranial atherosclerotic disease or carotid occlusion. Int J Stroke 2018;13:592-9. [Crossref] [PubMed]
23. Yaghi S, Khan F, Lewis S, Stipanovich A, Choi R, Baker R, et al. Impaired Perfusion and Early Ischemic Stroke Recurrence in Symptomatic Intracranial Atherosclerosis: BIORISK ICAS Study. Stroke 2026;57:96-106. [Crossref] [PubMed]
24. Olivot JM, Mlynash M, Inoue M, Marks MP, Wheeler HM, Kemp S, Straka M, Zaharchuk G, Bammer R, Lansberg MG, Albers GW. DEFUSE 2 Investigators. Hypoperfusion intensity ratio predicts infarct progression and functional outcome in the DEFUSE 2 Cohort. Stroke 2014;45:1018-23. [Crossref] [PubMed]
25. Faizy TD, Kabiri R, Christensen S, Mlynash M, Kuraitis GM, Broocks G, Flottmann F, Marks MP, Lansberg MG, Albers GW, Fiehler J, Wintermark M, Heit JJ. Favorable Venous Outflow Profiles Correlate With Favorable Tissue-Level Collaterals and Clinical Outcome. Stroke 2021;52:1761-7. [Crossref] [PubMed]
26. Jansen IGH, van Vuuren AB, van Zwam WH, van den Wijngaard IR, Berkhemer OA, Lingsma HF, Slump CH, van Oostenbrugge RJ, Treurniet KM, Dippel DWJ, van Walderveen MAA, van der Lugt A, Roos YBWEM, Marquering HA, Majoie CBLM, van den Berg R. MR CLEAN Trial Investigators. Absence of Cortical Vein Opacification Is Associated with Lack of Intra-arterial Therapy Benefit in Stroke. Radiology 2018;286:643-50. [Crossref] [PubMed]
27. Coull AJ, Rothwell PM. Underestimation of the early risk of recurrent stroke: evidence of the need for a standard definition. Stroke 2004;35:1925-9. [Crossref] [PubMed]
28. Zheng W, Yan H, Tian D, Li Y, Wang L, Lei Y, Wang B, Wang Y, Yang G, Wang X. Plaque enhancement of middle cerebral artery and pre-stroke diet are associated with prognosis of subacute ischemic stroke: A prospective high-resolution MR vessel wall imaging study. Eur J Radiol 2024;180:111693. [Crossref] [PubMed]
29. Zhang J, Sun B, Wang H, Shi H, Wen Q, Mossa-Basha M, Zhu C. Intracranial atherosclerotic plaque features on vessel wall imaging predict first ever and recurrence of stroke: a meta-analysis. Eur Radiol 2025;35:5017-26. [Crossref] [PubMed]
30. Vergallo R, Park SJ, Stone GW, Erlinge D, Porto I, Waksman R, et al. Vulnerable or High-Risk Plaque: A JACC: Cardiovascular Imaging Position Statement. JACC Cardiovasc Imaging 2025;18:709-40. [Crossref] [PubMed]
31. Grubb RL Jr, Raichle ME, Eichling JO, Ter-Pogossian MM. The effects of changes in PaCO2 on cerebral blood volume, blood flow, and vascular mean transit time. Stroke 1974;5:630-9. [Crossref] [PubMed]
32. Caplan LR, Hennerici M. Impaired clearance of emboli (washout) is an important link between hypoperfusion, embolism, and ischemic stroke. Arch Neurol 1998;55:1475-82. [Crossref] [PubMed]