Abstract WMP14: e-ASPECTS Improves Sensitivity to Early Ischemic Injury on Acute Computed Tomography Scans

Abstract

Background: ASPECTS (Alberta Stroke Program Early CT Score) is a validated scoring system for assessment of early ischemic change (EIC) on CT head scans, which can be used to guide patient management and improve diagnostic accuracy. Detection of EIC can be challenging particularly for less experienced clinicians. e-ASPECTS software uses machine learning algorithms to support physicians in detecting EIC, which can be quantified using the ASPECTS score. Hypothesis: e-ASPECTS shortens time for CT scan assessment and improves agreement with reference standard ASPECTS when compared to blinded assessment. Methods: 26 clinicians (including 11 radiologists, 6 junior and 7 consultant stroke physicians, and 2 non-specialist physicians) independently scored 2560 ASPECTS regions from 64 patients for signs of EIC on non-contrast CT brain scans. These were acquired within 4.5 hours of stroke onset. A familiarization training set of 5 patients was used prior to scoring. Images were randomized to manual or software assistance. After two weeks images were rescored using the alternative method. Scorers were blinded to clinical symptoms. Reference standard scores were defined by an independent neuroradiologist with information on clinical symptoms, access to 24h follow-up, and with CT perfusion or MRI scans when available. Results: Mean NIHSS was 11. Mean time to score scans fell by 34% (45s, 2:12 to 1:27, mm:ss) using e-ASPECTS assistance. Rater agreement with ground truth was greatest in the radiologist cohort, but performance improved across all clinician categories using e-APSECTS assistance (radiology kappa: 0.26 to 0.38). Sensitivity to EIC improved by a factor of two across all clinician groups using e-ASPECTS assistance, and this was most marked for less experienced physicians. Conclusion: In acute ischemic stroke e-ASPECTS assisted analysis increased accuracy and reduced time for detection of EIC. Routine assistance of non-contrast CT interpretation has the potential to reduce treatment times and improve accuracy across clinicians and sites.

Artificial intelligence to diagnose ischemic stroke and identify large vessel occlusions: a systematic review

“…The 10 studies reviewed that use RFL show that the AI often out-performs single radiologist ASPECTS and is non-inferior or even better than consensus ASPECTS. The reported sensitivity of AI algorithm ASPECTS range from 45% to 98%, mean 68%, and specificity ranges from 57% to 95%, mean 81% 9 10 14 16 17 22–27. Use of a convolutional neural network (CNN) for a combined asymmetric middle cerebral artery territory hypodensity and dense vessel detection may have higher performance; however, only area under the curve (AUC) metrics are reported (receiver operating characteristic AUC 92–96%) 28.…”

Artificial intelligence applied in acute ischemic stroke: from child to elderly

“…When using e-Aspects, the average time to score was reduced by 34%. All groups of clinicians who used e-ASPECTS assistance saw a twofold increase in their sensitivity to early ischemic changes, with the effects being more pronounced for less experienced clinicians [ 35 ]. The detection of LVOs has been the subject of other studies.…”

“…As was already mentioned, current commercial software focuses on the detection of AIS and estimation of the key characteristics that are essential to understand during the emergency phase, such as collaterals, ASPECT, and perfusion parameters [ 30 – 35 ].…”