Prediction of ischemic penumbra in patients with acute cerebral infarction based on multiparameter MRI radiomics and deep learning
10.12007/j.issn.0258-4646.2025.05.013
- VernacularTitle:基于多参数MRI影像组学和深度学习的急性脑梗死患者缺血半暗带预测研究
- Author:
Yuanhui CHEN
1
;
Yumeng LEI
1
;
Wenwen SHI
1
Author Information
1. 南昌市第一医院影像科,南昌 330008
- Publication Type:Journal Article
- Keywords:
acute cerebral infarction;
ischemic penumbra;
multiparameter magnetic resonance imaging radiomics;
deep learning;
fusion model
- From:
Journal of China Medical University
2025;54(5):455-460
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the clinical feasibility of predicting ischemic penumbra(IP)in patients with acute cerebral infarction based on multiparameter magnetic resonance imaging(MRI)radiomics and deep learning.Methods A total of 105 patients with acute cerebral infarction treated at our hospital between January 2020 and January 2024 were divided into non-IP(n=36)and IP(n=69)groups according to the results of MRI diffusion-weighted imaging(DWI).The clinical data of the two groups were collected,and multipa-rameter MRI radiomics features were screened.Clinical,radiomics,and deep learning models were constructed,and their distinguishing powers were evaluated.A fused model was constructed,and the receiver operating characteristic(ROC),calibration,and clinical decision curves were used to evaluate the predictive efficacy of the four models.Results Increases in the admission National Institutes of Health Stroke Scale(NIHSS)score,anisotropy score(FA),apparent diffusion coefficient(ADC),average diffusion coefficient(DCavg),and N-acetyl aspartate(NAA)were protective factors for IP.An increase in the lactic acid(Lac)level was a risk factor for IP(P<0.05).Of the six considered deep learning models,the support vector machine model performed the best,with an accuracy of 0.952(100/105),sensi-tivity of 0.957(66/69),and specificity of 0.944(34/36).These three models can distinguish the patients with IP.The four models exhibited high differentiation,accuracy,and effectiveness;the prediction efficiency of the fused model was the highest.Conclusion The fused model based on clinical features,multiparameter MRI radiomics,and deep learning accurately predicts IP in patients with acute cerebral infarction and can be used to provide personalized prediction results.
