Diagnostic Value of Deep Learning-Based Technology for Obstructive Stenosis of Extracranial and Intracranial Artery
Objective To explore the diagnostic value of deep learning(DL)technology for obstructive stenosis of extracranial and intracranial artery.Methods The study retrospectively included patients suspected with acute ischemic stroke from January 2020 to June 2021,who underwent both CTA and DSA within one month.Degrees of stenosis were classified as normal(0％),mild stenosis(＜50％),moderate stenosis(50-69％),severe stenosis(70-99％)and occlusion(100％)on patient-based and vessel-based analysis.Obstructive stenosis was defined as diameter stenosis＞70％.Diagnostic performance was assessed through AUC,sensitivity and specificity with DSA as reference standard.Results In patient-based analysis,the AUCs of DL technology and radiologists in detecting obstructive stenosis were 0.781[sensitivity and specificity were 0.934 and 0.627]and 0.840 respectively,and there had no statistical significance(P=0.074).In vessel-based analysis,the AUCs of DL technology and radiologists were 0.923[sensitivity and specificity were 0.885 and 0.962]and 0.932 respectively,and there had no statistical significance(P=0.393).The median analysis time of DL technology was 8.67 minutes,which was significantly lower than 29.55 minutes of radiologists(P＜0.001).Conclusion DL technology,with less time-consuming,can accurately assess extracranial and intracranial artery stenosis and will be a promising method to optimize risk stratification and guide treatment strategies.
Deep LearningCTAStenosisExtracranial and Intracranial Artery
NETL
NSTL
万方数据
