比较基于CT图像的纹理分析与临床评分在预测急性缺血性脑卒中出血性转化中的价值
CT images texture analysis versus clinical scores in predicting hemorrhagic transformation of acute ischemic stroke
宋心雨 1孙正 1李跃华1
作者信息
- 1. 200233 上海 上海交通大学医学院附属第六人民医院放射介入科
- 折叠
摘要
目的 本研究试图利用CT图像的纹理特征预测缺血性脑卒中出血性转化(hemorrhagic transformation,HT)的发生,并与传统的临床预测评分进行比较.方法 本研究共纳入73例急性前循环缺血性脑卒中患者,所有患者都进行再灌注治疗.根据随访ADC图像弥散受限区勾画出梗死区感兴趣区域(region of interesting,ROI),并匹配到计算机断层血管造影(computed tomographic angiography,CTA)和平扫CT(non-contrast CT,NCCT)相应的缺血区域.在所有患者中随机抽取5个HT和5个非HT组成测试组,余下组成训练组.分别在CTA和NCCT训练组中选取最有预测价值的6个纹理特征,然后将这些特征带入不同分类器进行5倍交叉验证训练,最后根据训练后的分类器对测试组进行预测评估.此外,对所有训练组患者进行4项临床评分(HAT、SEDAN、HIAT2、THRIVE-c).结果 训练后的分类器模型在CTA和NCCT中都表现出明显的预测价值.在CTA预测模型中,线性支持向量机(linear SVM)分类器预测效能最好,其在5倍交叉验证中的平均预测准确度为0.816,AUC值为0.890;在测试组中预测准确度为0.800,敏感性为0.600,特异性为1.000.逻辑回归(logistic regression,LR)为NCCT中表现最好的分类器,但NCCT模型HT的预测性能稍逊于CTA模型,其在训练组中预测准确度为0.697,AUC值为0.763.NCCT测试组的预测准确度为0.700,敏感性为0.600,特异性为0.800.相比于纹理分析模型,4种临床评分预测表现较差,AUC值均小于0.700.结论 基于CT图像脑缺血区的纹理分析(CTA和NCCT)具有预测AIS患者再灌注治疗后HT的能力,预测效能优于传统的临床评分方法.
Abstract
Objective To assess the value of CT image texture features in predicting the occurrence of hemorrhagic transformation(HT)in ischemic stroke,and to compare it with the traditional clinical prediction scores.Methods A total of 73 patients with acute anterior circulation ischemic stroke were enrolled in this study.All patients received reperfusion treatment.The region of interesting(ROI)of the infarction area was outlined according to the diffusion restricted area displayed on the follow-up ADC images,which were matched to the corresponding ischemic region on computed tomographic angiography(CTA)and on plain CT scan(non-contrast CT,NCCT).Five patients with HT and 5 patients with non-HT were randomly selected and used as the test set,and the remaining patients were assigned to the train set.The 6 texture features that had the most predictive value were separately selected from the CTA sets and NCCT train set,then the training of classifiers was earried out by using the 5-fold cross-validation method.Finally,the test set was evaluated according to the trained classifier.Besides,the determination of four clinical scores(HAT,SEDAN,HIAT2,THRIVE-c)was performed for all patients in the train set.Results The trained classifiers model performed well in not only CTA but also NCCT.In the CTA prediction model,linear SVM was chosen as the final classifier with 0.816 validation accuracy and 0.890 AUC value;and with 0.800 test accuracy,0.600 sensitivity,and 1.000 specificity in external test set Logistic regression(LR)was the best-performing classifier in NCCT.The predicted performance of HT was slightly worse than that of CTA,which had 0.697 validation accuracy and 0.763 AUC value.The test set of NCCT achieved 0.700 accuracy with 0.600 sensitivity and 0.800 specificity.Compared to the texture analysis models,all the four clinical scores showed a modest prediction efficiency in HT and AUC values,which were no more than 0.700.Conclusion Texture analysis of cerebral ischemic area based on CT images(CTA and NCCT)has the ability to predict HT after reperfusion treatment in AIS patients,and it is superior to traditional clinical scoring methods.(J Intervent Radiol,2024,33:230-235)
关键词
缺血性脑卒中/出血性转化/纹理分析/临床评分/预测Key words
ischemic stroke/hemorrhagic transformation/texture analysis/clinical score/prediction引用本文复制引用
基金项目
国家自然科学基金(8225024)
上海市临床重点专科建设项目(shslczdzk03203)
出版年
2024
NETL
NSTL
万方数据