基于靶扫描和常规扫描CT图像的影像组学模型预测肺磨玻璃结节的2年生长

 目的 比较基于靶扫描CT（targeted CT，T-CT）和常规扫描CT（conventional CT，C-CT）图像建立的CT影像组学特征集预测肺磨玻璃结节（ground-glass nodules，GGN）2年生长的价值，并建立影像组学列线图以帮助管理GGN。方法 回顾性收集2018年10月—2019年1月间414个随访肺GGN的T-CT、C-CT图像和临床资料，并按7∶3分为训练组（n=290）和验证组（n=124）。分别采用最小绝对收缩与选择算法逻辑回归、多因素逻辑回归筛选GGN 2年生长相关的影像组学特征及临床特征，构建影像组学特征集、临床特征集，结合成影像组学列线图。采用Delong检验比较基于T-CT和C-CT图像建立的CT影像组学特征集，并分别用C-CT和T-CT数据进行交叉预测。分别采用受试者工作特征曲线下面积（AUC）和临床决策曲线评估各模型效能和临床实用性。结果 T-CT和C-CT图像分别筛选出7个和6个特征用于构建影像组学特征集，两者AUC差异无统计学意义。筛选出年龄、性别和毛刺征3个临床特征构建临床特征集，结合C-CT影像组学特征集构建影像组学列线图。影像组学列线图在训练组和验证组中的AUC分别为0.948和0.933。临床特征的纳入未能显著提高模型预测效能（训练组和验证组P值分别为0.168和0.160），影像组学列线图较影像组学特征集获得更高临床净收益。结论 T-CT和C-CT影像组学特征集均能有效预测GGN的2年生长，且差异无统计学意义。影像组学列线图较影像组学特征集获得更高临床净收益，有助于管理GGN。

Radiomics models based on targeted and conventional CT images in predicting the two-year growth of pulmonary ground-glass nodules

Objectives To compare the value of radiomics signatures based on targeted CT (T-CT) and conventional CT (C-CT) images in predicting the two-year growth of pulmonary ground-glass nodules (GGNs), and to establish a radiomics nomogram to help manage GGNs. Methods T-CT and C-CT images and clinical data of 414 follow-up pulmonary GGNs from Oct 2018 to Jan 2019 were retrospectively reviewed and divided in 7: 3 ratio into the training group (n=290) and validation group (n=124). The least absolute shrinkage and selection operator and multivariate Logistic regression were used respectively to select the radiomics features and clinical features associated with GGNs' two-year growth to form the radiomics signatures and clinical signature, which were combined to construct the radiomics nomogram. The prediction performances of T-CT and C-CT radiomics signatures were compared by Delong test and were validated crossly using C-CT and T-CT images, respectively. The area under the curve (AUC) of receiver operating characteristic and clinical decisive curve were used to evaluate the prediction performance and clinical usefulness of the models. Results Based on T-CT and C-CT images, 7 and 6 radiomics features were selected to form the radiomics signatures, respectively. There was no significant AUC difference between the two radiomics signatures. The age, gender and spiculation were selected to form the clinical signature, which was combined with C-CT radiomics signature to construct a radiomics nomogram. The radiomics nomogram yielded AUCs of 0.948 and 0.933 respectively in the training and validation group. The radiomics nomogram combined with clinical features did not significantly improve the prediction performance (P=0.168, 0.160 in the training and validation groups, respectively), but did achieve a higher clinical net benefit than radiomics signature. Conclusions Both T-CT and C-CT radiomics signatures could effectively predict the two-year growth of GGNs. The radiomics nomogram achieved a higher clinical net benefit than radiomics signature and was helpful for the management of GGNs.