Machine Learning Approach to Identify Stroke Within 4.5 Hours

• DOI: 10.1161/strokeaha.119.027611
• Publication Date: 2020-01-28T10:00:26Z
• Authors: Hyunna Lee; Eun-Jae Lee; Sungwon Ham; Han-Bin Lee; Ji Sung Lee; Sun U. Kwon; Jong S. Kim; Namkug Kim; Dong-Wha Kang
• Keywords: Fluid-attenuated inversion recovery; Stroke

Background and Purpose— We aimed to investigate the ability of machine learning (ML) techniques analyzing diffusion-weighted imaging (DWI) fluid-attenuated inversion recovery (FLAIR) magnetic resonance identify patients within recommended time window for thrombolysis. Methods— analyzed DWI FLAIR images consecutive with acute ischemic stroke 24 hours clear symptom onset by applying automatic image processing approaches. These processes included infarct segmentation, DWI, registration feature extraction. A total 89 vector features from each sequence were captured used in ML. Three ML models developed estimate binary classification (≤4.5 hours): logistic regression, support machine, random forest. To evaluate performance models, sensitivity specificity identifying 4.5 compared human readings DWI-FLAIR mismatch. Results— Data a 355 analyzed. mismatch identified 48.5% 91.3% specificity. algorithms had significantly greater sensitivities than readers (75.8% P =0.020; 72.7% =0.033; 75.8% forest, =0.013) detecting hours, but their specificities comparable (82.6% =0.157; 82.6% =0.157). Conclusions— using multiple feasible even more sensitive