基于胸部CT的影像组学对周围型肺癌和肺炎性肌纤维母细胞瘤的鉴别价值

Radiomics Based on Chest CT was Used to Differentiate Peripheral Lung Cancer and Pneumonia Myofibroblastic Tumor

目的探讨基于动态增强计算机断层扫描(CT)的机器学习模型对周围型肺癌与肺炎性肌纤维母细胞瘤的鉴别价值。方法回顾性收集了53例周围型肺癌患者和59例肺炎性肌纤维母细胞患者的动态增强CT图像及临床信息;受试者按7∶3比例随机分为训练集(79例),验证集(33例)。在达尔文智能科研平台通过图像处理、病灶分割、特征提取和特征选择得到组学特征,用于建立机器学习模型-逻辑回归(LR)模型;通过受试者工作特征曲线下面积和校准曲线对上述模型的预测性能进行了评估。模型的临床实际应用价值由决策曲线来评价。结果基于影像组学特征建立的逻辑回归模型的鉴别性能较好,其在训练集中的曲线下面积为0.949[0.894~1.000],准确度为0.972,在验证集中,AUC为0.872[0.734~1.000],准确度为0.889。结论研究中建立的机器学习模型可以准确鉴别周围型肺癌与肺炎性肌纤维母细胞。

Objective To investigate the value of machine learning model based on dynamic enhanced computed tomography(CT)between peripheral lung cancer and pneumonia myofibroblastic tumor.Methods Dynamic enhanced CT images and clinical information of 53 patients with peripheral lung cancer and 59 patients with pneumonia myofibroblaststic tumor were retrospectively collected;all subjects were randomized at 7∶3 into training set(79)and validation set(33).In the Darwin intelligent research platform,radiomics features were obtained through image processing,lesion segmentation,feature extraction and feature selection,which are used to establish machine learning model-Logistic regression(LR);the prediction performance of the Logistic regression was evaluated by the area under the receiver operating curve and calibration curve.And decision curve analysis were used to evaluate the clinical use of the model.Results The classification performance of the Logistic regression based on radiomics features was good,and the area under the curve in the training set was 0.949[0.894~1.000]and the accuracy was 0.972.In the validation set,the area under the curve was 0.872[0.734~1.000]and the accuracy was 0.889.Conclusion The machine learning model developed in this study can be used to accurately identify peripheral lung cancer from pneumonia myofibroblaststic tumor.