A diffusion weighted imaging radiomics and clinical characteristics-based prediction model for prognosis of mechanical thrombectomy in acute anterior circulation large vessel occlusion stroke
10.3969/j.issn.1672-5921.2025.09.001
- VernacularTitle:基于扩散加权成像影像组学特征和临床因素构建急性前循环大血管闭塞性卒中机械取栓预后预测模型的价值分析
- Author:
Dong YANG
1
;
Weihe YAO
1
;
Wusheng ZHU
1
;
Xinfeng LIU
1
Author Information
1. 210002 南京大学医学院附属金陵医院(东部战区总医院)神经内科
- Publication Type:Journal Article
- Keywords:
Ischemic stroke;
Mechanical thrombectomy;
Radiomics;
Prediction model
- From:
Chinese Journal of Cerebrovascular Diseases
2025;22(9):587-600
- CountryChina
- Language:Chinese
-
Abstract:
Objective Build a predictive model integrating radiomics features with clinical characteristics for the prognosis prediction of acute anterior circulation large vessel occlusion(LVO)stroke patients after mechanical thrombectomy(MT),and explore its predictive value.Methods Patients with acute ischemic stroke who underwent endovascular treatment for LVO of the anterior circulation were enrolled consecutively from the endovascular treatment registry database for acute anterior circulation ischemic stroke(ACTUAL)and the Nanjing stroke registry system from January 2014 to January 2025 retrospectively.Baseline,clinical and imaging data were collected from enrolled patients,including gender,age,medical history(atrial fibrillation,hypertension,diabetes),smoke history,admission blood pressure,blood glucose,National Institutes of Health stroke scale(NIHSS)score,Alberta stroke program early CT score(ASPECTS),occluded blood vessels(internal carotid artery,middle cerebral artery),trial of Org 10172 in acute stroke treatment(TOAST)classification(atherosclerotic,cardiogenic embolism,others),collateral status(American Society of Interventional and Therapeutic Neuroradiology/Society of Interventional Radiology[ASITN/SIR]classification),the onset-to-door time,the time from onset to puncture,the operation time,the time from onset to recanalization,recanalization status(modified thrombolysis in cerebral infarction[mTICI]score),symptomatic intracerebral hemorrhage(sICH)within 72 hours after MT and functional outcome at 90 days post-MT(modified Rankin scale[mRS]score).Divide all patients into a training set and a validation set in a ratio of 7∶3.The training set is used to build the predictive model,and the validation set is used to verify the predictive model.In the training set,patients were divided into a good prognosis group(mRS score 0-2)and a poor prognosis group(mRS score 3-6),the variables with P＜0.05 from the univariate Logistic regression analysis were enrolled into the multivariate Logistic regression analysis to screen the clinical risk factors affecting prognosis.The preoperative head MR axial diffusion weighted imaging sequence images of patients in the training set were selected.The Pyradiomics toolkit of the Python 3.6 platform was used to implement radiomics feature extraction.After conducting consistency analysis on the extracted features,standardization processing was performed.In the training set,feature dimension reduction is carried out on the radiomics feature values obtained after extraction and processing.The least absolute shrinkage and selection operator(LASSO)model was used to screen the features.The support vector machine(SVM),k-nearest neighbor,lightweight gradient boosting algorithm,random forest method and extreme gradient boosting algorithm are used to respectively construct models based on the screened radiomics features,use grid search with cross validation(GridSearchCV)to gain specific parameters in each model.The receiver operating characteristic(ROC)curve was used to analyze and compare the area under the curve(AUC)of each radiomics model,screen the most suitable radiomics model,and verify it in the validation set.The predicted probability value of prognosis calculated by this model is taken as the radiomics score.In the training set,the radiomics scores and the screened clinical risk factors were taken as independent variables,and a multivariate Logistic regression analysis was conducted.A nomogram was used to construct a comprehensive prediction model of radiomics plus clinical factors for predicting the prognosis of MT in acute stroke patients of LVO.The AUC of the clinical factor prediction model,the radiomics prediction model,and the radiomics plus clinical factor comprehensive prediction model were compared in the training set and the validation set,respectively.Results A total of 107 acute anterior LVO patients who underwent MT were included,comprising 72 males and 35 females,aged 27 to 87 years,with a median age of 64(56,71)years.There were 74 cases in the training set,among which 48 cases had a good prognosis and 26 cases had a poor prognosis.There were 33 cases in the validation set,among which 24 cases had a good prognosis and 9 cases had a poor prognosis.The NIHSS score of patients in the training set was lower than that of patients in the validation set(12[8,19]points vs.15[11,21]points,P=0.03),while there were no statistically significant differences in the remaining baseline,clinical and imaging data compared with the validation set(all P＞0.05).(1)Included the variables with P＜0.05 from the univariate Logistic regression analysis into the multivariate Logistic regression analysis.The results showed that age(OR,1.066,95％CI 1.003-1.133,P=0.039)and admission NIHSS score(OR,1.126,95％CI 1.028-1.233,P=0.011)were independent risk factors for poor prognosis of MT in patients with acute anterior circulation LVO stroke.(2)A total of 725 radiomics features were extracted.The results of intra-observer consistency analysis showed that the median intraclass correlation coefficient(ICC)of radiomics features was 0.75(0.56,0.87),and there were 424 features with ICC＞0.7 and 127 features with ICC＞0.9.The results of the inter-observer consistency analysis showed that the median ICC of radiomics features was 0.73(0.53,0.86).After dimensionality reduction using the LASSO,12 most relevant features were selected and incorporated into the radiomics-based prognostic model.The AUCs of the radiomics prediction models constructed by applying SVM,k-nearest neighbor,lightweight gradient boosting algorithm,random forest method and extreme gradient boosting algorithm were 0.803,0.890,0.969,1.000 and 1.000,respectively.The AUCs in the validation set were 0.769,0.743,0.817,0.792 and 0.799,respectively.SVM was selected as the final algorithm for the construction of the radiomics model.The radiomics data were input into SVM to obtain the radiomics score of each patient.(3)A comprehensive predictive nomogram model combining radiomics and clinical factors was constructed based on radiomics score,age,and the NIHSS score at admission.In the validation group,the integrated model demonstrated a significantly higher AUC-ROC(0.918,95％CI 0.831-0.969)compared to the radiomics model(AUC 0.803,95％CI0.694-0.886,P=0.026)and the clinical-feature model(AUC 0.784,95％CI0.674-0.872,P=0.009).In the validation set,there were no statistically significant difference among the integrated model(AUC 0.935,95％CI 0.792-0.991),radiomics model(AUC 0.769,95％CI 0.589-0.897,P=0.111)and the clinical-feature model(AUC 0.894,95％CI 0.737-0.974,P=0.602).The integrated model exhibited good calibration in both the training set and the validation set(Hosmer-Lemeshow test,P values were respectively 0.350,0.580).Conclusion The integrated radiomics-clinical model can provide effective prediction of MT on outcomes in acute anterior circulation LVO stroke patients,and it may offer an objective basis for clinical decision-making.
