CT影像组学鉴别脾脏高代谢与正常代谢的价值

The Value of CT Radiomics Models in Differentiating Hypermetabolism and Normal Metabolism of Spleen

摘要: 目的：评估CT影像组学对于鉴别脾脏高代谢与正常代谢的价值。方法：回顾性分析正常代谢组274例及高代谢组93例的CT平扫资料，按8:2比例将所有样本分为训练集和测试集，提取脾脏纹理特征共1247个，对其进行特征选择，获得两组之间存在显著差异的特征参数共336，采用5倍交叉验证，并以逻辑回归(LR)、随机梯度下降(SGD)及线性判别分析(LDA) 3种机器学习方法建立预测模型，通过训练集及测试集数据进行评估及验证，绘制相应ROC曲线，使用曲线下面积(AUC)评估模型的鉴别诊断效能，以决策曲线分析预测模型的临床应用价值。结果：分别建立预测模型，分别获得3种机器学习方法的ROC曲线及AUC值，其中以SGD方法建立的预测模型在测试集中的AUC为0.8287，优于LR、LDA模型。决策分析曲线显示SGD模型较优于LR、LDA模型。结论：基于CT影像组学可以有效鉴别脾脏高代谢与正常代谢。

Abstract: Objective: To evaluate the value of CT radiomics models in differentiating hypermetabolism from normal metabolism in spleen. Methods: The CT data of 274 cases of normal metabolism group and 93 cases of high metabolism group were retrospectively analyzed. All the samples were divided into training set and test set according to the ratio of 8:2, the texture features of the spleen were ex-tracted, and the feature parameters with significant differences between the two groups were ob-tained. The 5-fold cross-validation method was used, and the prediction model was established by Logistic Regression (LR), stochastic gradient descent (SGD) and linear discriminant analysis (LDA) machine learning methods. The training set data were evaluated, and then the test set data were validated. The corresponding ROC curve was drawn, and the area under the curve (AUC) was used to evaluate the differential diagnostic performance of the model. Results: A total of 1247 texture fea-tures were extracted from spleen, 336 of which were significantly different between groups. To es-tablish a prediction model, evaluate and verify it, the ROC curves of the three machine learning methods were obtained respectively. The AUC of the prediction model established by SGD method in the training set and test set was 0.8782 and 0.8287, respectively, with the highest accuracy. Con-clusion: CT radiomics models can effectively distinguish hypermetabolism from normal metabolism in spleen.