MR高分辨率血管壁成像影像组学联合注意力机制预测症状性颅内动脉粥样硬化狭窄患者卒中复发  

MR high-resolution vessel wall imaging radiomics combined with attention mechanism for predicting stroke recurrence in patients with symptomatic intracranial atherosclerosis stenosis

作　　者：高宇 李子昂 魏正琦 韩琳 王杰 闫瑞芳[1] 赵红领[2] 崔红凯[1,3] GAO Yu;LI Zi'ang;WEI Zhengqi;HAN Lin;WANG Jie;YAN Ruifang;ZHAO Hongling;CUI Hongkai(Imaging Center,the First Affiliated Hospital of Xinxiang Medical University,Xinxiang 453100,China;Department of Neurointerventional,Xinxiang Central Hospital,Xinxiang 453000,China;Department of Neurointerventional,the First Affiliated Hospital of Xinxiang Medical University,Xinxiang 453100,China)

机构地区：[1]新乡医学院第一附属医院影像中心,河南新乡453100 [2]新乡市中心医院神经介入科,河南新乡453000 [3]新乡医学院第一附属医院神经介入科,河南新乡453100

出　　处：《中国医学影像技术》2025年第2期229-233,共5页Chinese Journal of Medical Imaging Technology

基　　金：河南省医学科技攻关联合共建项目(LHGJ20220629)。

摘　　要：目的观察MR高分辨率血管壁成像(HR-VWI)影像组学联合注意力机制集成模型预测症状性颅内动脉粥样硬化性狭窄(sICAS)患者卒中复发的价值。方法将363例接受HR-VWI检查的sICAS患者根据数据来源分为训练集(n=254)与验证集(n=109)。以基于HR-VWI T1及增强MRI影像组学模型为特征提取器,于责任斑块中捕获图像信息,结合Transformer注意力机制构建Trans模型;评估常规影像组学模型及Trans模型预测sICAS患者卒中复发的性能及其临床实用性。结果Trans模型预测训练集和验证集sICAS患者卒中复发的曲线下面积分别为0.992及0.988,优于T1模型、T1增强模型及双序列模型(P均<0.05)。校准曲线及决策曲线分析显示,Trans模型具有良好预测概率及临床实用性。结论HR-VWI影像组学联合注意机制集成模型用于评估sICAS患者卒中复发具有一定价值。Objective To observe the value of the integrated model of MR high-resolution vascular wall imaging(HR-VWI)and attention mechanism for predicting stroke recurrence in symptomatic intracranial atherosclerotic stenosis(sICAS)patients.Methods A total of 363 patients with sICAS who underwent HR-VWI were enrolled and stratified into training set(n=254)and validation set(n=109)according to their origins.Employing a radiomics model that utilized HR-VWI T1 and contrast-enhanced sequences for feature extraction,image data were captured from relevant plaques.Subsequently,a Trans model was developed by integrating the Transformer attention mechanism.The predictive performance and clinical utility of conventional radiomics models and Trans models for forecasting stroke recurrence among patients with sICAS were evaluated.Results In training set and validation set,the area under the curve of Trans model for predicting stroke recurrence in sICAS patients was 0.992 and 0.988,respectively,both superior to that of T1 model,T1 enhanced model and dual sequence model(all P<0.05).The calibration curve and decision curve analysis showed that Trans model had good predictive probability and clinical practicality.Conclusion The obtained integrated model of HR-VWI radiomics combined with attention mechanism had certain value for predicting stroke recurrence in patients with sICAS.

关 键 词：卒中 磁共振成像 影像组学 深度学习 

分 类 号：R743.1[医药卫生—神经病学与精神病学] R445.2[医药卫生—临床医学]