基于脂肪抑制增强T1WIDelta影像组学预测鼻咽癌患者放射治疗后口干分级

Delta radiomics based on fat suppression contrast enhanced T1WI for predicting xerostomia grade of nasopharyngeal carcinoma patients after radiotherapy

目的 基于脂肪抑制(FS)增强T1WI构建预测放射治疗(放疗)后鼻咽癌(NPC)患者口干级别的Delta影像组学模型,并评估其效能。方法 回顾性分析227例于诱导化学治疗(IC)后接受放疗或同步放化疗(CCRT)NPC患者IC前2周内及IC结束后鼻咽部MRI。于放疗结束后1周根据放射治疗肿瘤协作组标准评估患者口干分级(0~4级),将0、1级口干归为轻度口干(n=81),≥2级口干归为重度口干(n=146)。基于IC前、后FS增强T1WI提取双侧腮腺影像组学特征,计算Delta影像组学特征值。采用最小绝对收缩和选择算子和随机森林算法降维,筛选最优Delta影像组学特征,构建Delta影像组学模型。按8∶2比例将患者分为训练集(n=182)及测试集(n=45),绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评估Delta影像组学模型预测训练集及测试集NPC患者放疗后口干分级的效能;绘制校准曲线,评估模型性能。结果 分别基于IC前、后FS增强T1WI提取2 286个影像组学特征,最终筛选出8个最优Delta影像组学特征并以之构建模型,其预测训练集和测试集NPC患者放疗后口干分级的AUC分别为0.762、0.733;校准曲线显示,模型对于训练集及测试集的预测性能均良好。结论 基于FS增强T1WI Delta影像组学模型能及时有效地预测NPC患者放疗后口干分级。

Objective To establish a Delta radiomics model for predicting xerostomia grade in nasopharyngeal carcinoma(NPC)patients after radiotherapy based on fat suppression(FS)contrast enhanced T1WI,and to explore its diagnostic efficacy.Methods Nasopharyngeal MRI data within 2 weeks before and after induction chemotherapy(IC)of 227 NPC patients who underwent IC and radiotherapy or concurrent chemoradiotherapy(CCRT)were retrospectively analyzed.One week after radiotherapy,the patients'xerostomia grades(grade 0—4)were assessed according to the Radiation Therapy Oncology Group criteria,and grade 0 or 1 xerostomia was classified as mild xerostomia(n=81),while grade≥2 xerostomia was classified as severe xerostomia(n=146).Radiomics features of bilateral parotid gland were extracted based on FS contrast enhanced T1WI before and after IC,and values of Delta radiomics features were calculated.The least absolute shrinkage and selection operator and random forest algorithm were used to reduce dimension,and the optimal Delta radiomics features were screened to construct Delta radiomics model.The patients were divided into training set(n=182)and testing set(n=45)at the ratio of 8∶2.Receiver operating characteristic(ROC)curve was drawn,and the area under the curve(AUC)was calculated to evaluate the efficacy of Delta radiomics model for predicting xerostomia grade of NPC patients after radiotherapy in training and testing sets.Calibration curve was plotted to assess the performance of the model.Results Totally 2286 radiomics features were extracted based on FS contrast enhanced T1WI before and after IC,respectively,and 8 optimal Delta radiomics features were finally screened to construct the model.AUC of Delta radiomics model for predicting xerostomia grade of NPC patients in training and testing sets after radiotherapy was 0.762 and 0.733,respectively.Calibration curve showed that the model had good prediction performance in both training and testing sets.Conclusion Delta radiomics based on FS contrast enhanced T1WI could be used to timely and effectively predict xerostomia grade of NPC patients after radiotherapy.