Evaluating artificial intelligence software for delineating hemorrhage extent on CT brain imaging in stroke: AI delineation of ICH on CT.

BACKGROUND: The extent and distribution of intracranial hemorrhage (ICH) directly affects clinical management. Artificial intelligence (AI) software can detect and may delineate ICH extent on brain CT. We evaluated e-ASPECTS software (Brainomix Ltd.) performance for ICH delineation. METHODS: We qualitatively assessed software delineation of ICH on CT using patients from six stroke trials. We assessed hemorrhage delineation in five compartments: lobar, deep, posterior fossa, intraventricular, extra-axial. We categorized delineation as excellent, good, moderate, or poor. We assessed quality of software delineation with number of affected compartments in univariate analysis (Kruskall-Wallis test) and ICH location using logistic regression (dependent variable: dichotomous delineation categories 'excellent-good' versus 'moderate-poor'), and report odds ratios (OR) and 95 % confidence intervals (95 %CI). RESULTS: From 651 patients with ICH (median age 75 years, 53 % male), we included 628 with assessable CTs. Software delineation of ICH extent was 'excellent' in 189/628 (30 %), 'good' in 255/628 (41 %), 'moderate' in 127/628 (20 %), and 'poor' in 57/628 cases (9 %). The quality of software delineation of ICH was better when fewer compartments were affected (Z = 3.61-6.27; p = 0.0063). Software delineation of ICH extent was more likely to be 'excellent-good' quality when lobar alone (OR = 1.56, 95 %CI = 0.97-2.53) but 'moderate-poor' with any intraventricular (OR = 0.56, 95 %CI = 0.39-0.81, p = 0.002) or any extra-axial (OR = 0.41, 95 %CI = 0.27-0.62, p<0.001) extension. CONCLUSIONS: Delineation of ICH extent on stroke CT scans by AI software was excellent or good in 71 % of cases but was more likely to over- or under-estimate extent when ICH was either more extensive, intraventricular, or extra-axial.

Accuracy of artificial intelligence software for CT angiography in stroke.

OBJECTIVE: Software developed using artificial intelligence may automatically identify arterial occlusion and provide collateral vessel scoring on CT angiography (CTA) performed acutely for ischemic stroke. We aimed to assess the diagnostic accuracy of e-CTA by Brainomix™ Ltd by large-scale independent testing using expert reading as the reference standard. METHODS: We identified a large clinically representative sample of baseline CTA from 6 studies that recruited patients with acute stroke symptoms involving any arterial territory. We compared e-CTA results with masked expert interpretation of the same scans for the presence and location of laterality-matched arterial occlusion and/or abnormal collateral score combined into a single measure of arterial abnormality. We tested the diagnostic accuracy of e-CTA for identifying any arterial abnormality (and in a sensitivity analysis compliant with the manufacturer's guidance that software only be used to assess the anterior circulation). RESULTS: We include CTA from 668 patients (50% female; median: age 71 years, NIHSS 9, 2.3 h from stroke onset). Experts identified arterial occlusion in 365 patients (55%); most (343, 94%) involved the anterior circulation. Software successfully processed 545/668 (82%) CTAs. The sensitivity, specificity and diagnostic accuracy of e-CTA for detecting arterial abnormality were each 72% (95% CI = 66-77%). Diagnostic accuracy was non-significantly improved in a sensitivity analysis excluding occlusions from outside the anterior circulation (76%, 95% CI = 72-80%). INTERPRETATION: Compared to experts, the diagnostic accuracy of e-CTA for identifying acute arterial abnormality was 72-76%. Users of e-CTA should be competent in CTA interpretation to ensure all potential thrombectomy candidates are identified.

External Validation of e-ASPECTS Software for Interpreting Brain CT in Stroke.

OBJECTIVE: The purpose of this study was to test e-ASPECTS software in patients with stroke. Marketed as a decision-support tool, e-ASPECTS may detect features of ischemia or hemorrhage on computed tomography (CT) imaging and quantify ischemic extent using Alberta Stroke Program Early CT Score (ASPECTS). METHODS: Using CT from 9 stroke studies, we compared software with masked experts. As per indications for software use, we assessed e-ASPECTS results for patients with/without middle cerebral artery (MCA) ischemia but no other cause of stroke. In an analysis outside the intended use of the software, we enriched our dataset with non-MCA ischemia, hemorrhage, and mimics to simulate a representative "front door" hospital population. With final diagnosis as the reference standard, we tested the diagnostic accuracy of e-ASPECTS for identifying stroke features (ischemia, hyperattenuated arteries, and hemorrhage) in the representative population. RESULTS: We included 4,100 patients (51% women, median age = 78 years, National Institutes of Health Stroke Scale [NIHSS] = 10, onset to scan = 2.5 hours). Final diagnosis was ischemia (78%), hemorrhage (14%), or mimic (8%). From 3,035 CTs with expert-rated ASPECTS, most (2084/3035, 69%) e-ASPECTS results were within one point of experts. In the representative population, the diagnostic accuracy of e-ASPECTS was 71% (95% confidence interval [CI] = 70-72%) for detecting ischemic features, 85% (83-86%) for hemorrhage. Software identified more false positive ischemia (12% vs 2%) and hemorrhage (14% vs <1%) than experts. INTERPRETATION: On independent testing, e-ASPECTS provided moderate agreement with experts and overcalled stroke features. Therefore, future prospective trials testing impacts of artificial intelligence (AI) software on patient care and outcome are required before widespread implementation of stroke decision-support software. ANN NEUROL 2022;92:943-957.

Accuracy of Automated Computer-Aided Diagnosis for Stroke Imaging: A Critical Evaluation of Current Evidence.

There is increasing interest in computer applications, using artificial intelligence methodologies, to perform health care tasks previously performed by humans, particularly in medical imaging for diagnosis. In stroke, there are now commercial artificial intelligence software for use with computed tomography or MR imaging to identify acute ischemic brain tissue pathology, arterial obstruction on computed tomography angiography or as hyperattenuated arteries on computed tomography, brain hemorrhage, or size of perfusion defects. A rapid, accurate diagnosis may aid treatment decisions for individual patients and could improve outcome if it leads to effective and safe treatment; or conversely, to disaster if a delayed or incorrect diagnosis results in inappropriate treatment. Despite this potential clinical impact, diagnostic tools including artificial intelligence methods are not subjected to the same clinical evaluation standards as are mandatory for drugs. Here, we provide an evidence-based review of the pros and cons of commercially available automated methods for medical imaging diagnosis, including those based on artificial intelligence, to diagnose acute brain pathology on computed tomography or magnetic resonance imaging in patients with stroke.

The SITS Open Study: A Prospective, Open Label Blinded Evaluation Study of Thrombectomy in Clinical Practice.

BACKGROUND AND PURPOSE: We designed SITS (Safe Implementation of Treatment in Stroke) Open to determine benefit and safety of thrombectomy in clinical practice for large artery occlusion stroke, using selected stent retrievers plus standard care versus standard care alone. METHODS: SITS Open was a prospective, open, blinded evaluation, international, multicenter, controlled, nonrandomized registry study. Centers lacking access to thrombectomy contributed controls. Primary end point was categorical shift in modified Rankin Scale score at 3 months in the per protocol (PP) population. Principal secondary outcomes were symptomatic intracranial hemorrhage, functional independency (modified Rankin Scale score 0-2) and death at 3 months. Patients independently evaluated by video-recorded modified Rankin Scale interviews blinded to treatment or center identity by central core laboratory were regarded as PP population. Propensity score matching with covariate adjusted analysis was performed. RESULTS: During 2014 to 2017, 293 patients (257 thrombectomy, 36 control) from 26 centers in 10 countries fulfilled intention-to-treat and 200 (170 thrombectomy, 30 control) PP criteria; enrollment of controls was limited by rapid uptake of thrombectomy. In PP analysis, median age was 71 versus 71 years, and baseline National Institutes of Health Stroke Scale 17 versus 17 in the thrombectomy and control arms, respectively. The propensity score matching analysis for PP showed a significant shift for modified Rankin Scale at 3 months favoring the thrombectomy group (odds ratio, 3.8 [95% CI, 1.61-8.95]; P=0.002). Regarding safety, there were 4 cases of symptomatic intracranial hemorrhage in the thrombectomy group (2.4%) and none in the control group. CONCLUSIONS: In clinical practice, thrombectomy for patients with large artery occlusion stroke is superior to standard of care in our study. Registration: URL: https://www.clinicaltrials.gov. Unique Identifier: NCT02326428.

Osmotherapy for malignant cerebral edema in a phase 2 prospective, double blind, randomized, placebo-controlled study of IV glibenclamide.

BACKGROUND/OBJECTIVE: Malignant edema can be a life-threatening complication of large hemispheric infarction (LHI), and is often treated with osmotherapy. In this exploratory analysis of data from the GAMES-RP study, we hypothesized that patients receiving osmotherapy had symptomatic cerebral edema, and that treatment with intravenous (IV) glibenclamide would modify osmotherapy use as compared with placebo. METHODS: GAMES-RP was a phase 2 multi-center prospective, double blind, randomized, placebo-controlled study in LHI. Patients were randomized to IV glibenclamide (e.g. IV glyburide) or placebo. Cerebral edema therapies included osmotherapy and/or decompressive craniectomy at the discretion of the treating team. Total bolus osmotherapy dosing was quantified by "osmolar load". Radiographic edema was defined by dichotomizing midline shift at 24 h. Clinical changes were defined as any increase in NIHSS1a. RESULTS: Osmotherapy was administered to 40 of the 77 patients at a median of 39 [27-55] h after stroke onset. The median baseline DWI lesion volume was significantly larger in the osmotherapy treated group (167 [146-211] mL v. 139 [112-170] mL; P=0.046). Adjudicated malignant edema (75% v. 16%; P<0.001) was more common in the osmotherapy treated group. There were no differences in the proportion of patients receiving osmotherapy or the median total osmolar load between treatment arms. Most patients (76%) had a decrease in consciousness (NIHSS item 1A ≥1) on the day they began receiving osmotherapy. CONCLUSIONS: In the GAMES-RP trial, osmolar therapies were most often administered in response to clinical symptoms of decreased consciousness. However, the optimal timing of administration and impact on outcome after LHI have yet to be defined.

Real-world Independent Testing of e-ASPECTS Software (RITeS): statistical analysis plan.

Background: Artificial intelligence-based software may automatically detect ischaemic stroke lesions and provide an Alberta Stroke Program Early CT score (ASPECTS) on CT, and identify arterial occlusion and provide a collateral score on CTA. Large-scale independent testing will inform clinical use, but is lacking. We aim to test e-ASPECTS and e-CTA (Brainomix, Oxford UK) using CT scans obtained from a range of clinical studies. Methods: Using prospectively collected baseline CT and CTA scans from 10 national/international clinical stroke trials or registries (total >6600 patients), we will select a large clinically representative sample for testing e-ASPECTS and e-CTA compared to previously acquired independent expert human interpretation (reference standard). Our primary aims are to test agreement between software-derived and masked human expert ASPECTS, and the diagnostic accuracy of e-ASPECTS for identifying all causes of stroke symptoms using follow-up imaging and final clinical opinion as diagnostic ground truth. Our secondary aims are to test when and why e-ASPECTS is more or less accurate, or succeeds/fails to produce results, agreement between e-CTA and human expert CTA interpretation, and repeatability of e-ASPECTS/e-CTA results. All testing will be conducted on an intention-to-analyse basis. We will assess agreement between software and expert-human ratings and test the diagnostic accuracy of software.  Conclusions: RITeS will provide comprehensive, robust and representative testing of e-ASPECTS and e-CTA against the current gold-standard, expert-human interpretation.

Intravenous Glibenclamide Reduces Lesional Water Uptake in Large Hemispheric Infarction.

Background and Purpose- Prior studies have shown a linear relationship between computed tomography (CT)-derived radiodensity and water uptake, or brain edema, within stroke lesions. To test the hypothesis that intravenous glibenclamide (glyburide; BIIB093) reduces ischemic brain water uptake, we quantified the lesional net water uptake (NWU) on serial CT scans from patients enrolled in the phase 2 GAMES-RP Trial (Glyburide Advantage in Malignant Edema and Stroke). Methods- This was a post hoc exploratory analysis of the GAMES-RP study. Noncontrast CT scans performed between admission and day 7 (n=264) were analyzed in the GAMES-RP modified intention-to-treat sample. Quantitative change in CT radiodensity (ie, NWU) and midline shift (MLS) was measured. The gray and white matter NWU were also examined separately. Repeated-measures mixed-effects models were used to assess the effect of intravenous glibenclamide on MLS or NWU. Results- A median of 3 CT scans (interquartile range, 2-4) were performed per patient during the first 7 days after stroke. In a repeated-measures regression model, greater NWU was associated with increased MLS (ß=0.23; 95% CI, 0.20-0.26; P<0.001). Treatment with intravenous glibenclamide was associated with reduced NWU (ß=-2.80; 95% CI, -5.07 to -0.53; P=0.016) and reduced MLS (ß=-1.50; 95% CI, -2.71 to -0.28; P=0.016). Treatment with intravenous glibenclamide reduced both gray and white matter water uptake. In mediation analysis, gray matter NWU (ß=0.15; 95% CI, 0.11-0.20; P<0.001) contributed to a greater proportion of MLS mass effect, as compared with white matter NWU (ß=0.08; 95% CI, 0.03-0.13; P=0.001). Conclusions- In this phase 2 post hoc analysis, intravenous glibenclamide reduced both water accumulation and mass effect after large hemispheric infarction. This study demonstrates NWU is a quantitative and modifiable biomarker of ischemic brain edema accumulation. Clinical Trial Registration- URL: https://www.clinicaltrials.gov. Unique identifier: NCT01794182.

Effect of IV glyburide on adjudicated edema endpoints in the GAMES-RP Trial.

OBJECTIVE: In this secondary analysis of the Glyburide Advantage in Malignant Edema and Stroke (GAMES-RP) Trial, we report the effect of IV glyburide on adjudicated, edema-related endpoints. METHODS: Blinded adjudicators assigned designations for hemorrhagic transformation, neurologic deterioration, malignant edema, and edema-related death to patients from the GAMES-RP phase II randomized controlled trial of IV glyburide for large hemispheric infarct. Rates of these endpoints were compared between treatment arms in the per-protocol sample. In those participants with malignant edema, the effects of treatment on additional markers of edema and clinical deterioration were examined. RESULTS: In the per-protocol sample, 41 patients received glyburide and 36 received placebo. There was no difference in the frequency of hemorrhagic transformation (n = 24 [58.5%] in IV glyburide vs n = 23 [63.9%] in placebo, p = 0.91) or the incidence of malignant edema (n = 19 [46%] in IV glyburide vs n = 17 [47%] in placebo, p = 0.94). However, treatment with IV glyburide was associated with a reduced proportion of deaths attributed to cerebral edema (n = 1 [2.4%] with IV glyburide vs n = 8 [22.2%] with placebo, p = 0.01). In the subset of patients with malignant edema, those treated with IV glyburide had less midline shift (p < 0.01) and reduced MMP-9 (matrix metalloproteinase 9) levels (p < 0.01). The glyburide treatment group had lower rate of NIH Stroke Scale (NIHSS) increase of ≥4 during the infusion period (n = 7 [37%] in IV glyburide vs n = 12 [71%] in placebo, p = 0.043), and of change in level of alertness (NIHSS subscore 1a; n = 11 [58%] vs n = 15 [94%], p = 0.016). CONCLUSION: IV glyburide was associated with improvements in midline shift, level of alertness, and NIHSS, and there were fewer deaths attributed to edema. Additional studies of IV glyburide in large hemispheric infarction are warranted to corroborate these findings. CLINICALTRIALSGOV IDENTIFIER: NCT01794182. LEVEL OF EVIDENCE: This study provides Class II evidence that for patients with large hemispheric infarction, IV glyburide improves some edema-related endpoints.

Effect of IV alteplase on the ischemic brain lesion at 24-48 hours after ischemic stroke.

OBJECTIVE: To determine whether alteplase alters the development of ischemic lesions on brain imaging after stroke. METHODS: The Third International Stroke Trial (IST-3) was a randomized controlled trial of IV alteplase for ischemic stroke. We assessed CT or brain MRI at baseline (pretreatment) and 24 to 48 hours posttreatment for acute lesion visibility, extent, and swelling, masked to all other data. We analyzed associations between treatment allocation, change in brain tissue appearances between baseline and follow-up imaging, and 6-month functional outcome in IST-3. We performed a meta-analysis of randomized trials of alteplase vs control with pre- and postrandomization imaging. RESULTS: Of 3,035 patients recruited in IST-3, 2,916 had baseline and follow-up brain imaging. Progression in either lesion extent or swelling independently predicted poorer 6-month outcome (adjusted odds ratio [OR] = 0.92, 95% confidence interval [CI] 0.88-0.96, p < 0.001; OR = 0.73, 95% CI 0.66-0.79, p < 0.001, respectively). Patients allocated alteplase were less likely than controls to develop increased lesion visibility at follow-up (OR = 0.77, 95% CI 0.67-0.89, p < 0.001), but there was no evidence that alteplase reduced progression of lesion extent or swelling. In meta-analysis of 6 trials including IST-3 (n = 4,757), allocation to alteplase was associated with a reduction in ischemic lesion extent on follow-up imaging (OR = 0.85, 95% CI 0.76-0.95, p = 0.004). CONCLUSION: Alteplase was associated with reduced short-term progression in lesion visibility. In meta-analysis, alteplase reduced lesion extent. These findings may indicate that alteplase improves functional outcome by reducing tissue damage. CLASSIFICATION OF EVIDENCE: This study provides Class II evidence that IV alteplase impedes the progression of ischemic brain lesions on imaging after stroke.

Blood pressure variability and leukoaraiosis in acute ischemic stroke.

Higher blood pressure, blood pressure variability, and leukoaraiosis are risk factors for early adverse events and poor functional outcome after ischemic stroke, but prior studies differed on whether leukoaraiosis was associated with blood pressure variability, including in ischemic stroke. In the Third International Stroke Trial, blood pressure was measured in the acute phase of ischemic stroke immediately prior to randomization, and at 0.5, 1, and 24 h after randomization. Masked neuroradiologists rated index infarct, leukoaraiosis, and atrophy on CT using validated methods. We characterized blood pressure variation by coefficient of variance and three other standard methods. We measured associations between blood pressure, blood pressure variability, and leukoaraiosis using generalized estimating equations, adjusting for age, and a number of covariates related to treatment and stroke type/severity. Among 3017 patients, mean (±SD) systolic and diastolic blood pressure decreased from 155(±24)/82(±15) mmHg pre-randomization to 146(±23)/78(±14) mmHg 24 h later ( P < 0.005). Mean within-subject coefficient of variance was 0.09 ± 0.05 for systolic and 0.11 ± 0.06 for diastolic blood pressure. Patients with most leukoaraiosis were older and had higher blood pressure than those with least ( P < 0.0001). Although statistically significant in simple pairwise comparisons, no measures of blood pressure variability were associated with leukoaraiosis when adjusting for confounding variables ( P > 0.05), e.g. age. Our results suggest that blood pressure variability is not a potential mechanism to explain the association between leukoaraiosis and poor outcome after acute stroke.