Objective: Non-invasive prediction of IDH1 mutations by establishing a quantitative radiographic model based on DTI-based whole-tumor texture analysis. Methods: Preoperative MRI images of patients with surgically confirmed astrocytoma were collected in the First Affiliated Hospital of Soochow University from February 2016 to June 2019, including T(1)WI, T(2)WI, DTI, and T(1)-contrast enhancement images.A total of 38 patients were included, consisting of 12 mutants and 26 wilds, 20 males and 18 females, the average age was (49±15) years old.The ROIs were drawn on each level of the T(2)WI image using MaZda software and copied to the ADC and FA maps to extract texture feature parameters. The LASSO regression was used to determine the best radiomics features, radiological scores were calculated, and binary Logistic regression was used to construct a prediction model, then the ROC curve was used to analyze the diagnostic efficiency and the calibration curve was used to evaluate model prediction performance. Results: The four most valuable radiomics features were determined by LASSO regression, and then the radiomics scores and Logistic regression models of each patient were established. The radiomics scores of the wild and mutant groups were 2.3±0.3 and 1.8±0.4. There were significant differences between the groups (P<0.05). The ROC curve analysis showed an AUC of 0.837 with sensitivity and specificity of 91.7% and 61.5%, respectively. The Logistic regression model had good predictive performance with AUC of 0.907, sensitivity and specificity of 91.7% and 84.6%. Conclusions: DTI-based whole tumor radiomics model is benefit for predicting astrocytoma IDH1 mutations.
目的： 通过基于扩散张量成像（DTI）的全肿瘤纹理分析建立放射组学模型，从而实现无创性预测异柠檬酸脱氢酶1（IDH1）突变。 方法： 收集2016年2月至2019年6月苏州大学附属第一医院收治并经手术确诊星形细胞瘤患者的术前MRI图像，包括T(1)WI、T(2)WI、DTI和T(1)对比增强。最终共纳入38例患者，男20例、女18例，其中12例突变型，26例野生型，年龄（49±15）岁。使用MaZda软件在T(2)WI图像的每个层面上绘制感兴趣区（ROI）并将其复制到表观扩散系数（ADC）和各向异性分数（FA）图，提取肿瘤全体素的纹理特征参数。使用最小绝对值收敛和选择算子（LASSO）回归确定最佳放射性组学特征，计算放射组学分数，使用二元Logistic回归构建组学预测模型，绘制受试者工作特征（ROC）曲线分析诊断效能以及校正曲线评估模型预测性能。 结果： 经LASSO回归最终确定了4个最有价值的放射组学特征，分别为ADC(偏度3D)、ADC(Perc.01%3D)、ADC(Perc.50%3D)和FA(Perc.99%3D)，随后基于放射组学特征建立每例患者的放射组学标签和Logistic回归模型。野生组和突变组的放射组学评分分别为2.3±0.3和1.8±0.4，组间差异有统计学意义（P<0.05），ROC曲线分析显示AUC为0.837，灵敏度和特异度分别为91.7%和61.5%。Logistic回归模型具有良好的预测性能，AUC为0.907，灵敏度和特异度分别为91.7%和84.6%。 结论： 基于DTI的全肿瘤放射组学模型有助于预测胶质瘤IDH1突变。.
Keywords: Astrocytoma; Diffusion tensor imaging; IDH mutation; Radiomics.