利用放射组学模型预测肺癌和食管癌患者发生放射性肺炎的研究

目的 分析探讨肺癌和食管癌易患放射性肺炎(RP)患者的放射组学共同特征,建立能够同时预测两种肿瘤放疗后发生RP的预测模型。方法 回顾性分析行根治性放疗的Ⅲ期肺癌和Ⅲ期食管癌各 100例,依据随访影像学资料及临床信息进行RP分级,并收集其定位CT图像,将全肺作为感兴趣区域进行放射组学特征的提取,分析与RP相关的放射组学特征及临床、剂量学特征,利用机器学习进行模型构建。结果 提取出放射组学特征1691个,肺癌和食管癌患者经过方差分析、最小绝对值收敛和选择算子降维后与RP相关的放射组学特征分别为8个和6个,其中相同的参数为5个。使用随机森林构建预测模型,将肺癌和食管癌分别交替作为训练集和验证集,食管癌和肺癌作为独立验证集曲线下面积分别为0.662和0.645。结论 构建肺癌和食管癌患者发生RP的共同预测模型是可行的,但还需进一步扩充样本量,并且纳入临床和剂量学参数增加其准确度、稳定性和泛化能力。

Study of application of radiomics model in predicting radiation pneumontis in patients with lung cancer and esophageal cancer

Objective To analyze and explore the common radiomics features of radiation pneumonitis (RP) in patients with lung cancer and esophageal cancer, and then establish a prediction model that can predict the occurrence of RP in two types of cancer after radiotherapy. Methods Clinical data of 100 patients with stage Ⅲ lung cancer and 100 patients with stage Ⅲ esophageal cancer who received radical radiotherapy were retrospectively analyzed. The RP was graded by imaging data and clinical information during follow-up, and the planning CT images were collected. The whole lung was used as the volume of interest to extract radiomics features. The radiomics features,clinical and dosimetric parameters related to RP were analyzed, and the model was constructed by machine learning. Results A total of 1691 radiomics features were extracted from CT images. After ANOVA and LASSO dimensionality reduction in lung cancer and esophageal cancer patients, 8 and 6 radiomics features associated with RP were identified, and 5 of them were the same. Using the random forest to construct the prediction model,lung cancer and esophageal cancer were alternately used as the training and validation sets. The AUC values of esophageal cancer and lung cancer as the independent validation set were 0.662 and 0.645. Conclusions It is feasible to construct a common prediction model of RP in patients with lung cancer and esophageal cancer. Nevertheless, it is necessary to further expand the sample size and include clinical and dosimetric parameters to increase its accuracy, stability and generalization ability.