Vision Transformer fusion network assisted multiparametric MRI for diagnosing intracranial large vessel stenosis/occlusion
10.13929/j.issn.1003-3289.2025.07.008
- VernacularTitle:Vision Transfomer融合网络辅助多参数MRI诊断颅内大血管狭窄/闭塞
- Author:
Yangyang ZHOU
1
;
Huimin LI
1
;
Hongjin HE
1
;
Yuan LI
1
Author Information
1. 内蒙古人民医院核医学科,内蒙古呼和浩特 010010
- Publication Type:Journal Article
- Keywords:
stroke;
stenosis;
occlusion;
magnetic resonance imaging;
deep learning
- From:
Chinese Journal of Medical Imaging Technology
2025;41(7):1068-1072
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the value of Vision Transformer(ViT)fusion network assisted multiparametric MRI for diagnosing intracranial large vessel stenosis/occlusion(LVSO).Methods Totally 221 cases of ischemic stroke(IS)were retrospectively enrolled as test set,while other 28 cases of IS were taken as validation set,and data of diffusion weighted imaging(DWI),3D pseudo-continuous arterial spin labeling(3D-pCASL)and CT angiography(CTA)were collected.Taken CTA as reference standard,the value of ViT fusion network based on DWI,3D-pCASL and their combination for diagnosing intracranial LVSO were observed.Results In test set,the area under the curve(AUC)of ViT fusion network based on DWI,3D-pCASL and their combination for diagnosing intracranial LVSO was 0.879,0.908 and 0.925,respectively,no significant difference was found between each two(all P＞0.05),but AUC of DWI+3D-pCASL was slightly higher than that of the other two.In validation set,the AUC of ViT fusion network based on DWI+3D-pCASL for diagnosing intracranial LVSO was 0.850.Conclusion The performance of ViT fusion network based on DWI combined with 3D-pCASL for diagnosing intracranial LVSO was better.
