Visual Alberta stroke program early computed tomography score versus RAPID-AI perfusion in predicting outcome after late-window thrombectomy

Abstract

Purpose

To evaluate the prognostic utility of visual Alberta stroke program early computed tomography score (ASPECTS) and perfusion parameters obtained from automated RAPID-AI software in patients undergoing mechanical thrombectomy (MT) beyond 6 hours from stroke onset.

Methods

We retrospectively analyzed 86 patients with anterior circulation large vessel occlusion who underwent non-enhanced computed tomography (NECT), multiphase computed tomography angiography, and computed tomography perfusion within 6–24 hours before thrombectomy. Visual ASPECTS (assessed by junior doctor), RAPID-ASPECTS, and RAPID-CTP parameters (ischemic core volume, penumbra, and mismatch ratio) were recorded. The primary outcome was 90-day functional independence (modified Rankin Score 0-2). Multivariable logistic regression and receiver operating characteristic analysis were used to identify independent predictors.

Results

Visual ASPECTS was significantly associated with a favorable outcome (area under the curve = 0.709; optimal cut-off ≥ 6), while no perfusion-derived parameters reached statistical significance. In multivariable analysis, only visual ASPECTS (OR 0.083, 95% CI: 0.033–0.133; p = 0.001), hypertension (OR 0.252, 95% CI: 0.053–0.452; p = 0.014), and symptomatic intracranial hemorrhage (OR 0.634, 95% CI: 0.303–0.964; p < 0.001) remained independent predictors. Agreement between visual and RAPID-ASPECTS was moderate (intraclass correlation coefficient 0.67; 95% CI: 0.49–0.80; p < 0.001), but poor when dichotomized at the ≥ 6 threshold (Cohen's kappa κ = 0.18, p < 0.001).

Conclusion

Visual ASPECTS outperformed perfusion-derived metrics in predicting clinical outcomes after late-window thrombectomy. These findings support the continued relevance of NECT and expert visual scoring, particularly in settings where perfusion imaging may be limited or inconsistent.

Introduction

Acute ischemic stroke (AIS) due to large vessel occlusion (LVO) remains a leading cause of morbidity and mortality worldwide. ¹ Mechanical thrombectomy (MT) has revolutionized AIS management, especially after the DWI or CTP Assessment with Clinical Mismatch in the Triage of Wake-up and Late Presenting Strokes Undergoing Neurointervention with Trevo (DAWN) trial and the Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke (DEFUSE 3) trials demonstrated its efficacy in carefully selected patients up to 24 hours from last known well.^2,3 Within this extended time window (6–24 hours), neuroimaging plays a critical role in identifying candidates most likely to benefit from reperfusion therapy.

Both DAWN and DEFUSE 3 trials used automated software for processing imaging (RAPID-AI, iSchemaView, Inc., Menlo Park, California) for imaging analysis; however, selection criteria differed: DAWN prioritized clinical-core mismatch, whereas DEFUSE 3 focused on imaging mismatch alone.^2,3 Although perfusion imaging (ischemic core volume, Tmax-defined hypoperfused tissue, and mismatch ratio) provides detailed insights into tissue viability, its availability is often limited by scanner capabilities, time, contrast use, and radiation exposure. In contrast, the Alberta Stroke Program Early CT Score (ASPECTS), based on non-enhanced computed tomography (NECT), remains a widely used and validated tool for patient selection in both early and late MT windows.^2–9

Efforts to refine ASPECTS through artificial intelligence (AI) and initiatives such as the Acute Networks Striving for Excellence in Stroke (Angels Initiative) continue to expand its utility.^10,11 Whether NECT-derived ASPECTS alone can approximate the prognostic value of perfusion imaging in late-window MT remains a critical clinical question.^12–20

While international guidelines (American Heart Association/ American Stroke Association (AHA/ASA) 2019 and European Stroke Organization (ESO) 2023) continue to endorse advanced imaging based on DAWN/DEFUSE 3 criteria,^21,22 practice patterns vary. A large global survey revealed that NECT alone remains common in resource-limited settings, whereas advanced imaging is favored in high-income, comprehensive stroke centers. ²³ Thus, this study aims to evaluate the prognostic performance of visual ASPECTS, automated ASPECTS, and RAPID-AI perfusion metrics in a comprehensive stroke center in a developing country.

Methods

Study design and population

This single-center cohort study was conducted at a leading comprehensive stroke center in a developing country, which participates in the Angle Initiative. A total of 86 consecutive patients were included from January 2022 to May 2023, based on data from our prospective stroke registry. Ethical approval was granted by the Institutional Review Board, and the study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for observational studies (see the Supplemental Material).

Inclusion and exclusion criteria

Eligible patients met the following criteria: (1) age ≥ 18 years, (2) significant neurological deficits (National Institutes of Health Stroke Scale (NIHSS) score ≥ 6), (3) time from last known well to imaging between 6 and 24 hours, (4) acute ischemic stroke (AIS) due to anterior large vessel occlusion (aLVO) confirmed by NECT, computed tomography (CT) perfusion (CTP) and multiphase CT angiography (CTA), (5) ASPECTS calculated by visual and automated methods and CTP processed automatically by RAPID-AI, and (6) MT selected mostly based on DAWN and DEFUSE 3 criteria.

Exclusion criteria included pre-stroke mRS > 2, motion artifacts compromising RAPID-AI analysis (as determined by neurointerventionists with > 4 years of stroke intervention), missing three-month mRS data, or refusal to participate in the study.

Imaging protocol and evaluation

All patients underwent multimodal imaging using a Siemens SOMATOM Definition Edge—128—slide CT scanner, including (1) NECT, (2) multiphase CTA, and (3) CTP.

Main technical parameters of NECT included axial acquisition with a pitch of 0.55, 1 mm slice thickness reformatted to 5 mm, tube voltage of 120 kVp, and automatic exposure control (Casedose 4D). Visual ASPECTS was assessed by a junior physician with 2 years of stroke imaging (TTN), using a stroke window (C = 8, W = 32) at the level of the putamen and above, following a standardized method ²⁴ (Figure 1(a) and (b)). In contrast, RAPID-AI derived ASPECTS detects ischemic regions based on a reduction of 8–10 HU in attenuation compared to the contralateral hemisphere (Figure 1(d)).

Figure 1.

Evaluation by visual ASPECTS, RAPID-ASPECTS, and RAPID-CTP. Visual ASPECTS score of 5 points showing ischemic changes in the left insular cortex, putamen, and cortical regions (M4–M6) (a and b). Automated detection of left proximal M1 occlusion on CTA by RAPID (c). RAPID-ASPECTS score of 5 with involvement of the left insular cortex, putamen, internal capsule, caudate head, and M1 region (d). RAPID-CTP showing ischemic core (CBF <30%), penumbra (or mismatch volume) (Tmax > 6 s, CBF ≥ 30%, and mismatch ratio (e).

ASPECTS: Alberta stroke program early computed tomography score; CTP: computed tomography perf; CTA: computed tomography angiography; CBF: cerebral blood flow.

Multiphase CTA was acquired in three phases. ²⁵ A single 80–100 mL bolus of contrast was injected at 5 mL/s. Phase 1 (arterial phase) spanned the aortic arch to the vertex with occlusion site detected by RAPID-AI (Figure 1(c)). Phases 2 and 3 were obtained 8 and 16 seconds later, respectively, covering the skull base to vertex with no additional contrast. Parameters included low mAs, helical mode with table toggling, and full brain coverage.

CTP was performed subsequently with a delay of at least 60 seconds after CTA, covering a 10 cm region of the anterior circulation. Iodinated contrast (Ultravist 300 (Bayer Healthcare Pharmaceuticals Inc) or Xenetix 300 (Guerbet)) was administered at 1 mL/kg body weight via an 18 G intravenous catheter at 5 mL/s using the Illumena^® Néo injector (Guerbet, 93420 Villepinte, Paris, France). Scanning began 5 seconds before contrast injection and lasted ∼ 60 seconds with 20–30 dynamic acquisitions. RAPID-AI output included ischemic core volume (cerebral blood flow—CBF < 30%), penumbra (Tmax > 6 seconds and CBF ≥ 30%), and mismatch ratio (Figure 1(e)).

Mechanical thrombectomy

All procedures were performed under monoplane (Philips Allura Xper FD20) or biplane (Philips Allura Xper FD10/10) DSA systems by a senior neurointerventionalist (> 4 years’ experience) with a junior doctor. Anesthesia (conscious sedation or general) was selected based on clinical status and operator preference. A long-sheath catheter (Neuron Max 088, Penumbra) was positioned in the carotid artery for initial angiography. MT was performed using a second-generation stent retriever (Solitaire, Medtronic) and/or aspiration catheter (e.g. Sofia Plus, ACE, Jet 7, React) at the interventionist's discretion. Rescue therapies (balloon angioplasty and/or stenting) and antiplatelet agents were used as needed in atherosclerotic occlusions. Hemostasis was achieved via Angio-Seal 8F (Terumo) or manual compression.

Research variables

Clinical variables included age, sex, vascular risk factors (atrial fibrillation, hypertension, and diabetes), baseline NIHSS score, time from last known well to hospital admission, time from admission to imaging, time from imaging to groin puncture, and time from last known well to groin puncture.

Imaging variables included baseline ASPECTS on NECT (visual and RAPID-AI), CTP parameters (ischemic core, penumbra volume, and mismatch ratio) generated by RAPID-AI, as well as the occlusion site (internal carotid artery (ICA), M1, or proximal M2) and collateral status identified by multiphase CTA.

Interventional variables included pretreatment of intravenous recombinant tissue plasminogen activator (IV rtPA), and final modified treatment in cerebral infarction (mTICI) score.

Outcome variables included mRS at 90 days, dichotomized as favorable (mRS 0-2) or unfavorable (mRS 3-6), and presence of asymptomatic and symptomatic intracranial hemorrhage (sICH).

Statistical analysis

Statistical analyses were conducted using SPSS version 26.0 (SPSS Inc., Chicago, IL, USA). Missing data were retrieved from medical records when possible. Continuous variables are presented as means ± standard deviation (SD) or medians with interquartile ranges (IQR), and categorical variables as frequencies and percentages.

Multivariable logistic regression was performed to identify independent predictors of a favorable outcome. ROC curve analysis was used to evaluate the predictive value of ASPECTS, ischemic core, penumbra, and mismatch ratio, with area under the curve (AUC) reported. Agreement between visual and automated ASPECTS was assessed using the intraclass correlation coefficient (ICC) and Cohen's kappa. A p-value <0.05 was considered statistically significant, and 95% confidence intervals (CI) were reported.

Results

Patient characteristics and outcomes

Between January 2022 to May 2023, 86 patients were included in the study. The cohort comprised 48 males (55.8%) and 38 females (44.2%), with a median age of 65 years (IQR 58.8–73; range, 37–88 years). Vascular risk factors were common, with arterial hypertension in 57 patients (66.3%), followed by diabetes in 39 (45.3%) and atrial fibrillation in 30 (34.9%). All patients presented moderate to severe stroke, with a median NIHSS score of 13 (IQR 10–15; range, 6–20) and a median Glasgow Coma Scale score of 14 (IQR 13–15; range, 8–15). The mean time from last known well to hospital admission was 7.1 ± 3.5 hours (range, 6.3–19.3 hours), and the mean delay from admission to imaging was 51.1 ± 18.7 minutes (range, 25–90 minutes).

Regarding occlusion location, 24 patients (27.9%) had ICA occlusions, 49 (57.0%) had M1 occlusions, five (5.8%) had M2 occlusions, and eight patients (9.3%) had tandem occlusion. The median baseline visual ASPECTS was 7 (IQR 6–8; range, 3–10). Automated ASPECTS generated by RAPID was available for 58 patients, with a median of 6 (IQR 5–9; range, 0–10). Perfusion imaging showed a median infarct core volume of 14.5 mL (IQR 3.8–35.2; range, 0–62 mL) and a median penumbra volume of 113 mL (IQR 78.8–144; range, 4–248 mL), yielding a median mismatch ratio of 8.8 (IQR 4.3–24.6). Collateral status was rated as good in 18 patients (20.9%), moderate in 31 (36.0%), and poor in 37 patients (43.1%).

MT was successfully performed in all patients. Twenty-eight patients (32.6%) received intravenous thrombolysis with rtPA prior to thrombectomy. The mean time from stroke onset to groin puncture was 9.3 ± 3.6 hours (range, 6.1–21 hours), and the mean imaging-to-puncture time was 86.32 ± 33.8 minutes (range, 15–165 minutes). Overall, successful reperfusion (mTICI 2b-3) was achieved in 77/86 patients (89.5%). Post-procedure ICH occurred in 33/86 patients (38.4%), including 6 patients (7.0%) of sICH. A favorable functional outcome (mRS 0-2 at 90 days) was observed in 51/86 patients (59.3%), with a mortality rate of 8.1% (7/86).

Prognostic factors for a favorable outcome

Table 1 presents statistically significant differences in age, baseline Glasgow score, presence of atrial fibrillation, hypertension, visual ASPECTS, collateral status, and sICH between patients with and without favorable outcomes.

Table 1.

Factors associated with a favorable outcome based on the modified Rankin score.

Outcome	mRS 0-2 (n = 51)	mRS 3-6 (n = 35)	p value
Age, year, median (IQR)	61 (55–68)	69 (63–77)	0.002†*
Sex, male, n (%)	28 (54.9)	20 (57.1)	0.838§
Baseline NIHSS, median (IQR)	12 (10–15)	13 (11–15)	0.109†
Baseline Glasgow, median (IQR)	15 (14–15)	14 (13–15)	0.016†*
Atrial fibrillation, n (%)	12 (23.5)	18 (54.1)	0.008§*
Hypertension, n (%)	27 (52.9)	30 (85.7)	0.002§*
Diabetes mellitus, n (%)	23 (45.1)	16 (45.7)	0.955§
Pre-rtPA treatment, n (%)	16 (31.4)	12 (34.3)	0.778§
Time from onset to hospital admission, minutes, mean (SD)	378.2 (152.2)	485.3 (262.2)	0.092†
Time from hospital admission to imaging, minutes, mean (SD)	49.2 (17.8)	53.7 (19.7)	0.443†
Time from imaging to groin puncture, minutes, mean (SD)	90.4 (36.4)	80.8 (29.5)	0.200†
Time from onset to groin puncture, minutes, mean (SD)	517.8 (159.7)	619.8 (266.2)	0.207†
ICA occlusion, n (%)	15 (29.4)	9 (25.7)
M1 occlusion, n (%)	30 (58.8)	19 (54.3)
M2 occlusion, n (%)	3 (5.9)	2 (5.7)	0.450§
Tandem occlusion, n (%)	3 (5.9)	5 (14.3)
Baseline visual ASPECTS, median (IQR)	8 (6–9)	6 (4–8)	0.001†*
Baseline RAPID-ASPECTS, median (IQR)	6 (5–9)	6 (3–8)	0.263†
Ischemic core volume, mL, median (IQR)	8 (2–35)	21 (5–44)	0.105†
Penumbra volume, mL, median (IQR)	109 (78–131)	117 (82–171)	0.099†
Mismatch ratio, median (IQR)	10.8 (4.3–35)	6.7 (4.3–17.8)	0.566†
Good collateral status, n (%)	16 (31.4)	2 (5.7)	0.004§*
Moderate collateral status, n (%)	22 (43.1)	9 (25.7)	0.100§
Bad collateral status, n (%)	13 (25.5)	24 (68.6)	<0.001§*
Successful reperfusion (mTICI 2b-3), n (%)	48 (94.1)	28 (80)	0.096§
Intracranial hemorrhage, n (%)	17 (33.3)	16 (45.7)	0.249§
Symptomatic intracranial hemorrhage, n (%)	0 (0)	6 (17.1)	0.002§*

IQR: interquartile range; NIHSS: National Institutes of Health Stroke Scale; rTPA: recombinant tissue plasminogen activator; ASPECTS: Alberta stroke program early computed tomography score; mTICT: modified treatment in cerebral infarction.

^†p was calculated using the Student t-test; ^§ p was calculated using the Mann–Whitney U test; * p < 0.05.

Univariate analysis identified younger age, higher baseline Glasgow, absence of atrial fibrillation, absence of hypertension, higher visual ASPECTS, good collateral status, and absence of sICH as associated with favorable outcomes. In multivariate analysis, only three factors remained independently significant: absence of hypertension (OR 0.252; 95% CI: 0.053–0.452, p = 0.014), higher visual ASPECTS (OR 0.083; 95% CI: 0.033–0.133; p = 0.001), and absence of sICH (OR 0.634; 95% CI: 0.303–0.964; p < 0.001) (Table 2). Variance inflation factor (VIF) values ranged from 1.051 to 1.319, indicating low multicollinearity. When dichotomized, patients with initial visual ASPECTS ≥ 6 were 12 times more likely to achieve favorable outcomes than those with ASPECTS < 6 (70.6% vs. 16.7%; OR 12.0; 95% CI: 3.13–46.05; p < 0.001).

Table 2.

Logistic regression analysis of factors associated with a favorable outcome.

	Unadjusted OR	95% CI	p value	Adjusted OR	95% CI	p value	VIF
Age, per one-year decrease	0.014	0.006–0.023	0.002	0.007	−0.001–0.015	0.085	NA
Baseline NIHSS, per one-point decrease	0.022	−0.007–0.051	0.132	NA	NA	NA	NA
Baseline Glasgow, per one-point decrease	0.108	0.029–0.188	0.008	0.028	−0.040–0.097	0.417	NA
Absence of atrial fibrillation	0.291	0.115–0.734	0.009	0.161	−0.034–0.357	0.105	NA
Absence of hypertension	0.188	0.063–0.560	0.003	0.252	0.053–0.452	0.014	1.319
Baseline visual ASPECTS, per one-point decrease	0.106	0.051–0.160	<0.001	0.083	0.033–0.133	0.001	1.193
Baseline RAPID-ASPECTS, per one-point decrease	0.027	−0.021–0.076	0.263	NA	NA	NA	NA
Ischemic core volume, per 1-mL decrease	0.004	−0.001–0.010	0.105	NA	NA	NA	NA
Penumbra volume, per 1-mL decrease	0.002	0.000–0.004	0.099	NA	NA	NA	NA
Mismatch ratio, per 1-point decrease	−0.001	−0.003–0.002	0.566	NA	NA	NA	NA
Presence of good collateral status	7.543	1.609–35.360	0.010	−0.074	−0.296–0.148	0.508	NA
Successful reperfusion (mTICI 2b-3)	3.310	0.768–14.262	0.108	NA	NA	NA	NA
Absence of intracranial hemorrhage	0.594	0.245–1.437	0.248	NA	NA	NA	NA
Absence of symptomatic ICH	0.637	0.243–1.032	0.002	0.634	0.303–0.964	< 0.001	1.051

OR: odds ratio; CI: confidence interval; VIF: variance inflation factor; NA: not applicable; NIHSS: National Institutes of Health Stroke Scale; ASPECTS, Alberta stroke program early computed tomography score; mTICI: modified treatment in cerebral infarction.

ROC analysis of ASPECTS and perfusion parameters

Among imaging metrics, only visual ASPECTS demonstrated statistically significant predictive value for a favorable outcome, with an AUC of 0.709 (Figure 2). Based on the Youden index, the optimal ASPECTS threshold was 5.5, yielding a sensitivity of 94.1% and specificity of 42.9%.

Figure 2.

ROC curves. Illustration of the optimal cutoff values of ASPECTS and CTP parameters for predicting clinical outcomes in acute ischemic stroke.

ROC: receiver operating characteristic; ASPECTS: Alberta stroke program early computed tomography score; CTP: computed tomography perfusion.

Agreement between visual and automated ASPECTS

Moderate agreement was observed between visual and RAPID-ASPECTS scores (ICC = 0.67; 95% CI: 0.49–0.80; p < 0.001). When dichotomized at the clinical threshold of ≥ 6, agreement was poor (Cohen's κ = 0.18, p < 0.001).

Discussion

In this retrospective cohort of patients undergoing thrombectomy beyond 6 hours from symptom onset, we found that higher visual ASPECTS, absence of hypertension, and the absence of sICH were independent predictors of a favorable 90-day functional outcome. Among all imaging metrics assessed-including ASPECTS on NECT (visual and RAPID-delivered) and perfusion parameters delivered by RAPID-only, visual ASPECTS showed a significant correlation with outcome, with an AUC of 0.709 and a cut-off value of ≥ 6.

These findings support previous literature suggesting that visual ASPECTS remains a reliable tool for assessing ischemic burden and functional prognosis, especially when perfusion imaging is inconclusive or unavailable.^{12,13,15–18} Our results align with the threshold proposed by Shi et al. ¹³ (ASPECTS ≥ 6), but differ from the ASPECTS ≥ 4 threshold reported by Shchehlov et al. ¹² This discrepancy likely reflects variability in visual detection of early ischemic changes on NECT, which depend on reader experience and image quality. ²⁴ Difficulties encountered when scoring ASPECTS on NECT were documented with subtle hypodensity in insular ribbon (κ = 0.27), ²⁶ borderline infarction, deep white matter changes, influence of rater experience, ²⁴ and technical limitations of CT. ²⁷ The challenge to determine ASPECTS on NECT still existed even with the aid of AI software (such as RAPID-AI), in which only moderate agreement was observed (ICC = 0.67; 95% CI: 0.49–0.80; p < 0.001). When dichotomized at the clinically used cutoff of ≥ 6, the agreement was poor (Cohen's κ = 0.18; p < 0.001). This discrepancy suggests that while RAPID-ASPECTS provides standardized and rapid assessments, it may not always align with expert visual interpretation. ¹⁰ These findings highlight the potential limitations of relying exclusively on automated scores in acute decision-making, especially in cases with low image quality or borderline infarcts, and encourage the ASPECTS assessed through practice and training, like the Angles Initiative. ¹¹

Notably, none of the core perfusion metrics, including ischemic core volume, penumbra, or mismatch ratio, were independently associated with outcome in our study. This may reflect real-world limitations of perfusion imaging, ²³ including motion artifacts, contrast timing variability, or physiological differences in collateral circulation beyond 6 hours or in the case of tandem occlusion. While perfusion imaging remains essential for patient selection in randomized trials such as DAWN and DEFUSE 3, its generalizability to broader clinical practice may be limited by these factors. ¹⁸ Note well that the average core volume in our cohort was 14.5 mL (IQR 3.8–35.3 mL), which is larger than the DAWN study (7.6 mL, IQR 2.0–18.0 mL) and DEFUSE3 (9.4 mL, IQR 2.3–25.6 mL). This suggests that the population selection in our cohort had more extensive tissue damage compared to pivotal studies in the extended time window like DAWN and DEFUSE 3, but a proportion of patients with ASPECTS < 6 had a favorable functional outcome, was align with recent MT trials on large cores.^28–33

Regardless of the imaging modality used for patient selection, it is essential to recognize that treatment decisions must be individualized. As demonstrated in our study, other important variables—such as age, ³⁴ baseline Glasgow Coma Scale score, history of hypertension or atrial fibrillation, ³⁴ good collateral status on CTA, ¹² and a high likely of successful recanalization without symptomatic intracranial hemorrhage ³⁴ —also play a critical role in determining functional outcome (Table 1). Taken together, our findings underscore the continuing clinical utility of non-contrast CT and visual ASPECTS coring in the evaluation of patients for thrombectomy in the extended window. Particularly in resource-limited settings where perfusion imaging or RAPID (or other AI) software may be unavailable or technically suboptimal, visual ASPECTS remains a practical and informative tool for prognostication.

We are aware that our research may have several limitations. The retrospective design and single-center setting may limit generalizability. Visual ASPECTS was assessed by only junior doctors; inter-rater variability was not formally measured. Additionally, while we used RAPID software for perfusion analysis, the impact of motion artifacts or suboptimal injection timing could not be fully controlled. The relatively small sample size also limits statistical power for subgroup analysis.

Conclusions

In a real-world cohort of patients treated by thrombectomy in the late time window, visual ASPECTS was the only imaging-derived metric that independently predicted 90-day outcome, outperforming perfusion parameters obtained via automated software. These results support the role of non-contrast CT and expert visual assessment as reliable tools in late-window stroke care, particularly when perfusion imaging may be inaccessible or unreliable.