弥散加权成像影像组学联合中性粒细胞/淋巴细胞构建急性脑卒中无效再通预测模型

Prediction model construction of futile recanalization in acute stroke based on DWI radiomics and neutrophil⁃lymphocyte ratio

目的 基于DWI影像组学特征联合中性粒细胞/淋巴细胞的机器学习构建急性脑卒中机械取栓治疗后无效再通预测模型并与基于DWI影像组学特征的预测模型进行比较。方法 对100例在南京市第一医院行机械取栓治疗后完全再通(改良脑梗死溶栓分级2b~3级)的急诊脑卒中患者的临床及影像资料进行回顾性分析。无效再通定义为患者机械取栓治疗后完全再通但3个月时预后不良(mRS 3~6分)。应用A.K.软件进行提取DWI急性脑卒中病灶区影像组学特征并应用最低绝对收缩和选择算子(LASSO)筛选与无效再通相关的最佳影像组学特征。通过支持向量机(SVM)分类器建立预测急性脑卒中无效再通的预测模型(模型1:基于DWI影像组学特征;模型2:DWI影像组学特征联合中性粒细胞/淋巴细胞),采用Delong检验比较两个模型的效能。结果 共提取396个DWI影像组学特征,经降维后共筛选出6个与无效再通预测最相关的特征。ROC分析显示模型1预测急性脑卒中无效再通的AUC为0.885,敏感度和特异度分别为0.825、0.833。模型2预测急性脑卒中无效再通的AUC为0.967,敏感度和特异度分别为0.900、0.917。两个模型的效能比较差异有统计学意义(z=-2.094,P=0.036)。结论 基于DWI影像组学特征联合中性粒细胞/淋巴细胞的机器学习构建急性脑卒中机械取栓术后无效再通的预测模型明显优于基于DWI影像组学特征的预测模型,能够较为准确的预测急性脑卒中机械取栓术后无效再通。

Objective To construct the prediction model of futile recanalization after mechanical thrombectomy in acute stroke based on DWI radiomics and neutrophil⁃lymphocyte ratio(NLR)and compare with the prediction model based on DWI radiomics.Methods Clinical and imaging data of 100 acute stroke patients who underwent mechanical thrombectomy and complete recanalization(mTICI grade 2b~3)in Nanjing First Hospital were analyzed retrospectively.Futile recanalization was defined as complete recanalization after mechanical thrombectomy but poor outcome at 3 months(mRS 3~6).A.K.software was used to extract the DWI radiomics of the acute stroke lesions,and the least absolute shrinkage and selection operator(LASSO)regression model was used to screen the best radiomics related to futile recanalization.The prediction model of futile recanalization in acute stroke was constructed by support vector machine(SVM)(model 1:based on DWI radiomics;model 2:based on DWI radiomics and NLR).The performance of two models were compared using Delong test.Results A total of 396 DWI radiomics were extracted.After dimensionality reduction,a total of 6 features related to futile recanalization were selected.ROC showed that the AUC of model 1 for predicting futile recanalization of acute stroke was 0.885,with a sensitivity of 0.825 and a specificity of 0.833,respectively.The AUC of model 2 for predicting futile recanalization of acute stroke was 0.967,with a sensitivity of 0.900 and a specificity of 0.917,respectively.The efficacy of two models showed significant differences(z=-2.094,P=0.036).Conclusion The prediction model of futile recanalization after mechanical thrombectomy in acute stroke based on DWI radiomics and NLR is significantly better than the prediction model based on DWI radiomics,and can accurately predict the futile recanalization after mechanical thrombectomy in acute stroke.