摘要
目的:探讨基于CT平扫图像的影像组学模型在预测自发性脑出血患者早期血肿扩大时的价值,并与常规影像征象预测效能进行比较。方法:回顾性分析2015年6月至2019年2月苏北人民医院209例自发性脑出血患者,发病6h以内均行首次头颅CT检查及24h内头颅CT复查,根据复查CT结果变化分为血肿扩大组(71例)和未扩大组(138例)。采用Darwin智能科研平台提取并筛选影像组学特征,分别结合支持向量机(SVM)、逻辑回归(LR)分类器构建预测模型,同时构建由常规影像征象(初始血肿体积、血肿形状、漩涡征、混合征和岛征)组成的多变量二元逻辑回归分析模型,比较2种影像组学模型和常规影像征象模型的预测价值(训练集167例,测试集42例)。使用受试者操作特征(ROC)曲线评估预测性能。结果:共提取1 223个特征参数,通过最小绝对收缩与选择算子(LASSO)回归分析筛选出16个特征参数。构建的SVM模型中,训练集和测试集的ROC曲线下面积(AUC)分别为0.933、0.918;LR模型中,训练集和测试集的AUC分别为0.939、0.900;常规影像征象模型中,训练集和测试集的AUC分别为0.744、0.852。结论:构建的影像组学模型在预测自发性脑出血患者早期血肿扩大方面具有较高性能,并优于常规影像征象的预测效能。
Purpose: To explore the value of radiomics based on non-contrast CT in predicting early hematoma expansion in patients with spontaneous intracerebral hemorrhage and to compare the predictive performance with conventional radiological predictor. Methods: Retrospective analysis was performed on 209 patients with spontaneous intracerebral hemorrhage in Northern Jiangsu People’s Hospital from June 2015 to February 2019, all of whom received the first cranial CT examination within 6 h of onset and CT reexamination within 24 h after admission. All patients were divided into hematoma expansion group(n=71) and non-expansion group(n=138)according to CT reexamination results. Radiomics features were extracted and selected using Darwin Intelligent Research Platform, then the predictive models were constructed by combining the selected features with support vector machine(SVM) and logistic regression(LR) classifier respectively. And a multivariate binary LR analysis model composed of conventional radiological predictors(initial hematoma volume, hematoma shape, swirl sign,blend sign and island sign) was also established, then the predictive value of two radiomics models and conventional radiological predictors model was compared(the training set and the test set were 167 and 42 cases, respectively).Predictive performance was assessed with the receiver operating characteristic(ROC) curve analysis.Results: A total of 1 223 features were extracted, and 16 features were selected by the least absolute shrinkage and selection operator(LASSO) regression. In the SVM model, the area under ROC curve(AUC) of the training set and the test set were0.933 and 0.918, respectively. In the LR model, the AUC of the training set and the test set were 0.939 and 0.900,respectively. In the conventional radiological predictors model, the AUC of the training set and the test set were0.744 and 0.852, respectively.Conclusion: The constructed radiomics models showed high performance in predicting early hematoma expansion in patients with spontaneous intracerebral hemorrhage and outperformed the conventional radiological predictors.
作者
李青润
韩雷
陈红日
常璐璠
叶靖
张洪英
Li Qingrun;Han Lei;Chen Hongri;Chang Lufan;Ye Jing;Zhang Hongying(Department of Radiology.Northern Jiangsu People's Hospital,Yangzhou University;Imaging Medicine and Nuclear Medicine Class of 2018,Graduate School of Dalian Medical University;Beijing Medical Al)
出处
《中国医学计算机成像杂志》
CSCD
北大核心
2021年第2期91-96,共6页
Chinese Computed Medical Imaging
基金
国家自然科学基金面上项目(81471642)。
关键词
自发性脑出血
血肿扩大
影像组学
计算机体层成像
Spontaneous intracerebral hemorrhage
Hematoma expansion
Radiomics
Computed tomography