Stroke Scoring Under the Microscope: Review Finds DWI-ASPECTS Agreement Varies Widely

When a large stroke strikes, one small number can decide whether a patient is wheeled into the angiography suite for a clot-retrieving thrombectomy or treated with medication alone. That number, the ASPECT score, is supposed to summarize how much brain tissue has already been lost on an early brain scan. A new systematic review now reports that when clinicians read the same score from diffusion-weighted MRI scans — a version of the scale known as DWI-ASPECTS — their agreement with one another ranges all the way from poor to excellent, and the evidence base behind the score is far thinner than its everyday clinical use would suggest.

The review, published in Frontiers in Neurology by a team from The Ottawa Hospital and the Ottawa Hospital Research Institute, set out to answer a deceptively simple question: how consistently do clinicians actually score DWI-ASPECTS in patients with acute ischemic stroke of the anterior circulation, the large territory of the brain supplied by the carotid arteries where most thrombectomy decisions are made? The authors searched MEDLINE, EMBASE and EMBASE Classic, EBM Reviews and CINAHL for articles published between January 1, 2000 and September 9, 2026, and screened 3,658 titles and abstracts after duplicate removal. Thirty-eight articles went forward to full-text review, and one more emerged from screening the reference lists, giving 39 candidates. Twenty were excluded for not meeting the criteria, leaving nineteen studies that together contributed thirty-seven separate measures of agreement.

The scale itself has a short but consequential history. The Alberta Stroke Program Early CT Score, first described in 2000, is a ten-point categorical scale designed to quantify early signs of ischemia on computed tomography scans of acute stroke patients. It was created to help guide acute stroke management and predict prognosis in patients treated with intravenous thrombolysis, and it went on to become a vital patient selection criterion for thrombectomy trials. It is now embedded in national stroke guidelines in hyperacute treatment algorithms that help clinicians decide who should undergo mechanical clot removal. The same scoring template can be applied to magnetic resonance imaging, and specifically to the diffusion-weighted sequence, which visualizes hyperacute ischemia with more sensitivity than CT. DWI-ASPECTS has been associated with lower mortality at ninety days compared with CT-based scoring, is widely used in France as the initial diagnostic tool for acute ischemic stroke, and is endorsed as an initial diagnostic modality by the American Heart Association, the American Stroke Association and the European Stroke Organisation and European Society for Minimally Invasive Neurological Therapy guidelines.

What the review found was not a single reliability figure but a spread. For interrater agreement — how well different clinicians score the same scans — full-scale intraclass correlation coefficients ranged from 0.72 to 0.96, which the authors classify as strong to excellent agreement. Unweighted kappa statistics ranged from 0.17 to 0.76, spanning poor to strong agreement, and weighted kappa values ranged from 0.441 to 0.985, from moderate to excellent. When the score was collapsed into two categories, kappa values ranged from 0.62 to 0.75, and for three-category versions from 0.64 to 0.68, both classified as strong agreement. Intrarater agreement — how consistently the same clinician scores the same scans twice — was reported by only two studies. There, full-scale ICC values ranged from 0.862 to 0.93, rated excellent, while kappa values ranged from 0.47 to 0.53, rated moderate.

The heterogeneity behind those numbers is itself a central finding. The included studies varied in populations, methods, analyses and results so widely that the authors could not pool them in a meta-analysis. The mean number of patients per study was about 130, ranging from 30 to 309, and the mean number of raters was about five, ranging from two to thirty-four, though only three studies had more than five raters. Six of the interrater agreement studies used more than two raters. The imaging hardware differed as well: eight studies used 1.5 Tesla MRI scanners, four used 3 Tesla, one used both, and six did not disclose the magnetic field strength at all. Reporting within the included articles was often incomplete, the authors note, which compounds the difficulty of drawing a single conclusion about how reliable the score really is.

The statistical spread also reflects a deeper mathematical tension in the scale itself. DWI-ASPECTS is an ordinal, categorical instrument that stands in for something continuous: the volume of infarcted brain tissue. Studies that treated the score as a continuous variable and measured agreement with the intraclass correlation coefficient mostly found good to excellent agreement. But the same numerical score can correspond to very different infarct volumes, a point demonstrated in the HERMES meta-analysis, which showed that points subtracted from different, unequally weighted brain regions can produce the same total ASPECTS. It is not clearly defined how much volume loss is required in a specific region for that region to register as a point on the score. Because of this, the authors argue, weighted kappa — which treats disagreements between adjacent scores as less serious than large ones — is the more clinically appropriate measure, since a one-point disagreement matters far less in some regions and score ranges than in others. Unweighted kappa, which treats every disagreement equally, is correspondingly less relevant, and it was indeed where the review observed its highest heterogeneity.

The clinical stakes concentrate at the low end of the scale. Traditionally, stroke guidelines have recommended thrombectomy for patients with a large vessel occlusion and an initial ASPECTS of six or higher. But recent randomized controlled trials have begun to demonstrate improved functional outcomes with thrombectomy in patients with larger infarct burdens — DWI-ASPECTS of three to five — opening the door to treating patients with bigger strokes. That shift makes the reliability of the score in exactly this range urgent, and it is precisely where the evidence is weakest. Only four of the reviewed studies offered interrater agreement values for DWI-ASPECTS below six, and of the limited studies that compared reliability across dichotomized or trichotomized subgroups of more severe strokes, none looked explicitly at the reliability of DWI-ASPECTS three to five versus other categories. On the intrarater side, none of the sparse existing data offered information on dichotomized or trichotomized values for more severe strokes at all.

The review also highlights a subtler problem with reading too much into any single agreed-upon number. DWI-ASPECTS is an ordinal scale built from multiple categorical decisions about whether ischemia is present at particular locations, compressed into one numerical value. Two raters can therefore produce the same total score while disagreeing about exactly where the ischemia lies. And the regions are not equivalent: in patients with low DWI-ASPECTS from zero to five who received thrombectomy, involvement of the M6 region in right-sided strokes and of the internal capsule in left-sided strokes has been associated with increased disability. This inherent heterogeneity among the scored regions adds a further layer of complexity when the scale is applied at the bedside, where a single number is asked to carry prognostic and treatment weight.

The authors place their findings in a broader context of imaging-based stroke scoring. A previous systematic review and meta-analysis demonstrated low interrater agreement for CT-ASPECTS, and while no comparable review exists for perfusion-based CTP-ASPECTS, individual studies have found higher degrees of interrater and intrarater agreement for perfusion deficits on CT perfusion compared with CT-ASPECTS. In large strokes with ASPECTS below six, an infarct volume of seventy milliliters or more was found to be a better predictor of poor outcome than a CT-ASPECTS of five or below alone in patients undergoing thrombectomy within the first six hours — results that, the authors suggest, point toward a need to better understand how infarct volume correlates with ASPECT scores in future work.

The overall conclusion is a call for better evidence rather than a verdict against the score. Data on the reliability of DWI-ASPECTS in acute ischemic stroke are scant, the methodologies and results of the available studies are heterogeneous, and more robust studies are required — particularly for strokes with DWI-ASPECTS below six, and especially in the three-to-five range now being targeted by new thrombectomy trials. As DWI-ASPECTS takes on a larger role in selecting patients for increasingly aggressive treatment of larger infarcts, the review suggests that clinicians and trialists should be aware that the number guiding those decisions rests on a surprisingly thin and inconsistent reliability literature, and that variations in agreement measures are directly tied to the mathematical limitations of the score itself.

Subject of Research: Interrater and intrarater reliability of the DWI-ASPECTS stroke imaging score in anterior circulation acute ischemic stroke

Article Title: The reliability conundrum of DWI-ASPECTS in stroke assessment: a systematic review of agreement on DWI-ASPECTS in anterior circulation stroke

Article References: Jamal, M., Brissette, V., Zitikyte, G., Shamy, M., & Fahed, R. (2026). The reliability conundrum of DWI-ASPECTS in stroke assessment: a systematic review of agreement on DWI-ASPECTS in anterior circulation stroke. Frontiers in Neurology, 17, Article 1900599. https://doi.org/10.3389/fneur.2026.1900599

Image Credits: AI Generated

DOI: 10.3389/fneur.2026.1900599

Keywords: DWI-ASPECTS, stroke, thrombectomy, systematic review, interrater agreement, intrarater agreement, MRI, acute ischemic stroke, ASPECT score, reliability, large vessel occlusion, infarct volume