Head CT deep learning model is highly accurate for early infarct estimation

Abstract

Non-contrast head CT (NCCT) is extremely insensitive for early (< 3–6 h) acute infarct identification. We developed a deep learning model that detects and delineates suspected early acute infarcts on NCCT, using diffusion MRI as ground truth (3566 NCCT/MRI training patient pairs). The model substantially outperformed 3 expert neuroradiologists on a test set of 150 CT scans of patients who were potential candidates for thrombectomy (60 stroke-negative, 90 stroke-positive middle cerebral artery territory only infarcts), with sensitivity 96% (specificity 72%) for the model versus 61–66% (specificity 90–92%) for the experts; model infarct volume estimates also strongly correlated with those of diffusion MRI (r2 > 0.98). When this 150 CT test set was expanded to include a total of 364 CT scans with a more heterogeneous distribution of infarct locations (94 stroke-negative, 270 stroke-positive mixed territory infarcts), model sensitivity was 97%, specificity 99%, for detection of infarcts larger than the 70 mL volume threshold used for patient selection in several major randomized controlled trials of thrombectomy treatment.

Priorities for Advancements in Neuroimaging in the Diagnostic Workup of Acute Stroke

“…AI is currently being extensively tested regarding its performance in early stroke detection. An AI model outperformed expert readers in detecting early ischemic changes on NCCT in a recent study, 75 and a systematic review including 11 studies and 1976 cases revealed that AI-based ASPECTS performed similar or better than radiologists in identifying early stroke changes on NCCT. 76 Moreover, AI-based NCCT-ASPECTS was reported as good or better as human rating for posterior circulation stroke.…”

Incorporating algorithmic uncertainty into a clinical machine deep learning algorithm for urgent head CTs

“…This is not unexpected given that the ML algorithm is derived from a system that was initially developed to detect acute intracranial hemorrhage. In addition, early ischemia may not create CT scan abnormalities that are detectable by even experienced neuroradiologists, even though ischemic stroke may produce severe neurological deficits [21,22]. It is thus possible that cases with early acute ischemia did not result in sufficient CT abnormalities that could be detected by the ML algorithm.…”

“…However, progress has been made to train an algorithm specifically to detect subtle CT changes from early cerebral ischemia. Our group has created and validated such an algorithm [22]. We foresee aggregating these algorithms to provide a more thorough assessment of patients who present with symptoms that may suggest a stroke.…”