Machine learning models based on brain functional network features combining clinical indicators for predicting postoperative outcomes of patients with drug-resistant mesial temporal lobe epilepsy
10.13929/j.issn.1003-3289.2025.09.007
- VernacularTitle:脑功能网络特征联合临床指标机器学习模型预测药物难治性内侧颞叶癫痫患者术后转归
- Author:
Lidan LIN
1
;
Xiaoyang WANG
;
Zhifeng HUANG
;
Jianzhou CHEN
;
Sifan QIU
;
Yaling CHEN
;
Shangwen XU
Author Information
1. 福建医科大学福总临床医学院,福建 福州 350025;中国人民解放军联勤保障部队第九○○医院放射诊断科,福建 福州 350025
- Publication Type:Journal Article
- Keywords:
epilepsy,temporal lobe;
magnetic resonance imaging;
surgical procedures,operative;
treatment outcome;
forecasting
- From:
Chinese Journal of Medical Imaging Technology
2025;41(9):1488-1493
- CountryChina
- Language:Chinese
-
Abstract:
Objective To observe the value of machine learning(ML)models based on brain functional network features combining clinical indicators for predicting postoperative outcomes of patients with drug-resistant mesial temporal lobe epilepsy(DR-mTLE).Methods Totally 84 patients with unilateral DR-mTLE who underwent surgery were retrospectively enrolled and classified into seizure free(SF)group(n=55)and non-seizure free(NSF)group(n=29)according to one-year postoperative follow-up.Clinical data were analyzed to screen independent predictors of postoperative outcomes.Based on brain preoperative resting-state functional MRI,brain functional networks were constructed using graph theory analysis,and 587 features were extracted.Five-fold cross validation was used to divide the data into training set and test set,then the optimal brain functional network features related to postoperative outcomes of DR-mTLE patients were selected.Combining with clinically relevant independent predictors,ML models were constructed using classifiers including Gaussian process(GP),logistic regression(LR),support vector machine(SVM)and quadratic discriminant analysis(QDA),respectively,and the prediction efficacy,calibration and clinical value of each ML model were evaluated.Results Both course of disease and lesion location were clinically relevant independent predictors of postoperative outcome of DR-mTLE patients(OR=0.928,5.710,P=0.010,0.016).Four optimal brain function network features were selected,including betweenness centrality of the third zone of cerebellar vermis,degree centrality of right globus pallidus,nodal efficiency of temporal left inferior temporal gyrus and nodal clustering coefficient of left inferior parietal lobule.The average area under the curve(AUC)of GP,LR,SVM and QDA models in test set was 0.868,0.864,0.875 and 0.870,respectively.Calibration curves and decision curve analysis indicated that each ML model had good calibration and high clinical net benefit.Conclusion ML models based on brain functional network features combining with clinical indicators could be used to effectively predict postoperative outcomes in DR-mTLE patients.
