Deep-learning tool for early identification of non-traumatic intracranial hemorrhage etiology based on CT scan v1

Objective To develop anartificial intelligence system that can accurately identify acute non-traumaticintracranial hemorrhage (ICH) etiology based on non-contrast CT (NCCT) scansand investigate whether clinicians can benefit from it in a diagnostic setting. Methods The deep learning modelwas developed with 1868 eligible NCCT scans with non-traumatic ICH collectedbetween January 2011 and April 2018. We tested the model on two independentdatasets (TT200 and SD 98) collected after April 2018. The model’s diagnosticperformance was compared with clinicians’ performance. We further designed asimulated study to compare the clinicians’ performance with and without thedeep learning system augmentation. Results The proposed deeplearning system achieved area under the receiver operating curve of0.986 (95% CI 0.967–1.000) on aneurysms, 0.952 (0.917–0.987) on hypertensivehemorrhage, 0.950 (0.860–1.000) on arteriovenous malformation (AVM), 0.749(0.586–0.912) on Moyamoya disease (MMD), 0.837 (0.704–0.969) on cavernousmalformation (CM), and 0.839 (0.722–0.959) on other causes in TT200 dataset.Given a 90% specificity level, the sensitivities of our model were 97.1% and90.9% for aneurysm and AVM diagnosis, respectively. The model also shows animpressive generalizability in an independent dataset SD98. The cliniciansachieve significant improvements in the sensitivity, specificity, and accuracyof diagnoses of certain hemorrhage etiologies with proposed systemaugmentation. Conclusions The proposed deep learning algorithms can bean effective tool for early identification of hemorrhage etiologies based onNCCT scans. It may also provide more information forclinicians for triage and further imaging examination selection.