Machine learning models based on CT angiography morphology combined with blood inflammation indicators for predicting rupture of intracranial aneurysms
10.13929/j.issn.1003-3289.2025.01.007
- VernacularTitle:基于CT血管成像形态学联合血液炎症指标机器学习模型预测颅内动脉瘤破裂
- Author:
Jia CHEN
1
;
Yu GAO
1
;
Hailiang WANG
1
Author Information
1. 嘉兴市中医医院放射科,浙江嘉兴 314000
- Publication Type:Journal Article
- Keywords:
intracranial aneurysm;
rupture;
inflammation;
machine learning;
prospective studies
- From:
Chinese Journal of Medical Imaging Technology
2025;41(1):29-34
- CountryChina
- Language:Chinese
-
Abstract:
Objective To observe the value of machine learning(ML)models based on CT angiography(CTA)morphology combined with blood inflammation indicators for predicting rupture of intracranial aneurysms(IA).Methods Totally 286 IA patients were enrolled,including 143 cases in ruptured group and 143 case in unruptured group.Among them 200 cases(100 in ruptured subgroup and 100 in unruptured subgroup)were taken as training set,while 86 cases(43 in ruptured subgroup and 43 in unruptured subgroup)were taken as validation set.CTA morphological parameters and blood inflammation indicators were compared between subgroups in training set,and the impact factors of IA ruptured among univariate variables were screened using stepwise logistic regression analysis.Logistic regression(LR),classification and regression tree(CART)and backpropagation neural network(BPNN)models were constructed based on the above impact factors,respectively.Receiver operating characteristic curves were drawn,the area under the curves(AUC)were calculated to evaluate the efficacy of these models for predicting IA rupture.Results Neutrophil,neutrophil-lymphocyte ratio,interleukin-10,tumor necrosis factor-α,transforming growth factor-β,as well as tumor width,height and size ratio(SR)were all impact factors of IA rupture(all P＜0.05),and collinearity diagnosis suggested that no covariate relationship existed among these factors.LR,CART and BPNN models had good efficacy for predicting IA rupture in both training and validation sets with AUC of 0.878-0.993,and BPNN model had the best predictive efficacy(AUC of 0.993,0.976).Conclusion ML models based on CTA morphology combined with blood inflammation indicators could be used to predict IA rupture effectively,among which BPNN model had the best efficacy.
