Deep Learning-based Detection Of Intracranial Artery Stenosis And Occlusion In MR Angiography

Background and purpose: Stroke is the second leading cause of mortality globally.Intracranial atherosclerotic stenosis is the common cause of ischemic stroke.With the steadily increasing demand for radiologic images,we aim to develop and evaluate a computer-aided detection system based on deep learning for automated detection and localization of intracranial arterial stenosis and occlusion from time-of-flight MR angiography.Materials and methods: In a retrospective study,MR images with radiological reports of intracranial arterial stenosis and occlusion were extracted.All lesions were reinterpreted and measured according to the method from The Warfarin-Aspirin Symptomatic Intracranial Disease Study(WASID),and manual annotation with a bounding box labeled “moderate stenosis”,“severe stenosis”,“occlusion”,and “absence of signal” was considered as ground truth.The images were randomly divided into a training set and a test set.A deep learning algorithm based on you only look once version 5(YOLOv5)detection mode was developed with the training set,and its sensitivity and positive predictive value to detect lesions were evaluated in the test set.Results: Dataset of 200 examinations consisted of a total of 411 lesions,including 242 moderate stenosis,84 severe stenosis,70 occlusion and 15 absence of signal.MR images contained 291 lesions in training set,120 lesions in test set.The sensitivity and positive predictive value were 64.2% and 83.7%,respectively.The detection sensitivity in relation to location was greatest in internal carotid artery(86.2%).The false negatives cases from the algorithm were mainly due to moderate stenosis(30/43).Conclusions: Deep learning algorithm has feasibility and great potential for automated detection of intracranial arterial steno-occlusive lesions from TOF-MRA,providing with exact position for lesions.It is expected to be involved in clinical diagnosis routine in the future.