MR high-resolution vessel wall imaging radiomics combined with attention mechanism for predicting stroke recurrence in patients with symptomatic intracranial atherosclerosis stenosis
10.13929/j.issn.1003-3289.2025.02.010
- VernacularTitle: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
- Author:
Yu GAO
1
;
Zi'ang LI
;
Zhengqi WEI
;
Lin HAN
;
Jie WANG
;
Ruifang YAN
;
Hongling ZHAO
;
Hongkai CUI
Author Information
1. 新乡医学院第一附属医院影像中心,河南新乡 453100
- Publication Type:Journal Article
- Keywords:
stroke;
magnetic resonance imaging;
radiomics;
deep learning
- From:
Chinese Journal of Medical Imaging Technology
2025;41(2):229-233
- CountryChina
- Language:Chinese
-
Abstract:
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.
