Yao Xiang, Liao Lishang, Han Yuxiao, Wei Ting, Wu Hai, Wang Yiying, Li Yanfei, Zhang Xinyuan, Ren Ke
Department of Radiology, The Xiang'an Hospital of Xiamen University, Xiamen.
Department of Neurosurgery, Hospital (TCM) Affiliated to Southwest Medical University, Luzhou.
J Craniofac Surg. 2019 Nov-Dec;30(8):e768-e771. doi: 10.1097/SCS.0000000000005765.
To evaluate the edema area around basal ganglia hemorrhage by the application of computerized tomography (CT)-based radiomics as a prognostic factor and improve the diagnosis efficacy, a total of 120 patients with basal ganglia hemorrhage were analyzed retrospectively. The texture analysis software Mazda 3.3 was used to preprocess the CT images and manually sketch the region of interest to extract the texture features. The extracted texture features were selected by Fisher coefficient, POE+ACC and mutual information. The texture discriminant analysis uses the B11 module in the Mazda 3.3 software. The data were randomly divided into a training dataset (67%) and test dataset (33%). To further study the texture features, the training dataset can be divided into groups according to the median of GCS score, NIHSS score, and maximum diameter of hematoma. Random forest model, support vector machine model, and neural network model were built. AUC of the receiver operating characteristics curve was used to assess the performance of models with test dataset. Among all texture post-processing methods, the lowest error rate was 2.22% for the POE+ACC/nonlinear discriminant. For the maximum diameter of hematoma, GCS score, and NIHSS score group, the lowest error rate were 26.66%, 23.33%, and 30.00%, respectively. The values of AUCs were 0.87, 0.81, and 0.76, for random forest model, support vector machine model, and neural network model in the test dataset, respectively. Radiomic method with proper model may have a potential role in predicting the edema area around basal ganglia hemorrhage. It can be used as a secondary group in the diagnosis of edema area around basal ganglia hemorrhage.
为了通过基于计算机断层扫描（CT）的放射组学评估基底节区脑出血周围的水肿面积作为预后因素并提高诊断效能，回顾性分析了120例基底节区脑出血患者。使用纹理分析软件Mazda 3.3对CT图像进行预处理，并手动勾勒感兴趣区域以提取纹理特征。通过Fisher系数、POE+ACC和互信息选择提取的纹理特征。纹理判别分析使用Mazda 3.3软件中的B11模块。数据被随机分为训练数据集（67%）和测试数据集（33%）。为了进一步研究纹理特征，训练数据集可根据格拉斯哥昏迷量表（GCS）评分、美国国立卫生研究院卒中量表（NIHSS）评分和血肿最大直径的中位数进行分组。构建了随机森林模型、支持向量机模型和神经网络模型。使用受试者工作特征曲线的AUC评估测试数据集模型的性能。在所有纹理后处理方法中，POE+ACC/非线性判别法的最低错误率为2.22%。对于血肿最大直径、GCS评分和NIHSS评分组，最低错误率分别为26.66%、23.33%和30.00%。测试数据集中随机森林模型、支持向量机模型和神经网络模型的AUC值分别为0.87、0.81和0.76。采用合适模型的放射组学方法在预测基底节区脑出血周围水肿面积方面可能具有潜在作用。它可作为基底节区脑出血周围水肿面积诊断的次要指标。
