Deep Learning Approaches to Identify Patients within the Thrombolytic Treatment Window

Abstract

1.AbstractBackgroundTreatment of acute ischemic stroke is heavily contingent upon time, as there is a strong relationship between time clock and tissue progression. We sought to a develop a deep learning algorithm for classifying time since stroke (TSS) from MR images by comparison to neuroradiologist assessments of imaging signal mismatch and evaluation on external data.MethodsThis retrospective study involved patients who underwent MRI from 2011-2019. Models were trained to classify TSS within 4.5 hours; performance metrics with confidence intervals were reported on both internal and external evaluation sets.ResultsA total of 772 patients (66 ± 9 years, 319 women) were used for model development and evaluation. Three board-certified neuroradiologists’ assessments, based on majority vote, yielded a sensitivity of 0.62, a specificity of 0.86, and a Fleiss’ kappa of 0.46. The deep learning method performed similarly to radiologists and outperformed previously reported methods, with the best model achieving an average evaluation accuracy, sensitivity, and specificity of 0.726, 0.712, and 0.741, on an internal cohort and 0.724, 0.757, and 0.679, respectively, on an external, unseen evaluation cohort from another institution.ConclusionThis model achieved higher generalization performance on external evaluation datasets than the current state of the art for TSS classification.