Rapid Automated Quantification of Cerebral Leukoaraiosis on CT Images: A Multicenter Validation Study.

Purpose To validate a random forest method for segmenting cerebral white matter lesions (WMLs) on computed tomographic (CT) images in a multicenter cohort of patients with acute ischemic stroke, by comparison with fluid-attenuated recovery (FLAIR) magnetic resonance (MR) images and expert consensus. Materials and Methods A retrospective sample of 1082 acute ischemic stroke cases was obtained that was composed of unselected patients who were treated with thrombolysis or who were undergoing contemporaneous MR imaging and CT, and a subset of International Stroke Thrombolysis-3 trial participants. Automated delineations of WML on images were validated relative to experts' manual tracings on CT images, and co-registered FLAIR MR imaging, and ratings were performed by using two conventional ordinal scales. Analyses included correlations between CT and MR imaging volumes, and agreements between automated and expert ratings. Results Automated WML volumes correlated strongly with expert-delineated WML volumes at MR imaging and CT (r2 = 0.85 and 0.71 respectively; P < .001). Spatial-similarity of automated maps, relative to WML MR imaging, was not significantly different to that of expert WML tracings on CT images. Individual expert WML volumes at CT correlated well with each other (r2 = 0.85), but varied widely (range, 91% of mean estimate; median estimate, 11 mL; range of estimated ranges, 0.2-68 mL). Agreements (κ) between automated ratings and consensus ratings were 0.60 (Wahlund system) and 0.64 (van Swieten system) compared with agreements between individual pairs of experts of 0.51 and 0.67, respectively, for the two rating systems (P < .01 for Wahlund system comparison of agreements). Accuracy was unaffected by established infarction, acute ischemic changes, or atrophy (P > .05). Automated preprocessing failure rate was 4%; rating errors occurred in a further 4%. Total automated processing time averaged 109 seconds (range, 79-140 seconds). Conclusion An automated method for quantifying CT cerebral white matter lesions achieves a similar accuracy to experts in unselected and multicenter cohorts.

Prediction of stroke thrombolysis outcome using CT brain machine learning.

A critical decision-step in the emergency treatment of ischemic stroke is whether or not to administer thrombolysis - a treatment that can result in good recovery, or deterioration due to symptomatic intracranial haemorrhage (SICH). Certain imaging features based upon early computerized tomography (CT), in combination with clinical variables, have been found to predict SICH, albeit with modest accuracy. In this proof-of-concept study, we determine whether machine learning of CT images can predict which patients receiving tPA will develop SICH as opposed to showing clinical improvement with no haemorrhage. Clinical records and CT brains of 116 acute ischemic stroke patients treated with intravenous thrombolysis were collected retrospectively (including 16 who developed SICH). The sample was split into training (n = 106) and test sets (n = 10), repeatedly for 1760 different combinations. CT brain images acted as inputs into a support vector machine (SVM), along with clinical severity. Performance of the SVM was compared with established prognostication tools (SEDAN and HAT scores; original, or after adaptation to our cohort). Predictive performance, assessed as area under receiver-operating-characteristic curve (AUC), of the SVM (0.744) compared favourably with that of prognostic scores (original and adapted versions: 0.626-0.720; p < 0.01). The SVM also identified 9 out of 16 SICHs, as opposed to 1-5 using prognostic scores, assuming a 10% SICH frequency (p < 0.001). In summary, machine learning methods applied to acute stroke CT images offer automation, and potentially improved performance, for prediction of SICH following thrombolysis. Larger-scale cohorts, and incorporation of advanced imaging, should be tested with such methods.