| 黄蓉,卢晓旭,孙学明,吴慧.18F-FDG PET/CT联合临床建立多模态影像组学模型预测局部晚期头颈部鳞癌对于新辅助免疫联合化学治疗的反应[J].中国医学影像技术,2024,40(10):1493~1498 |
| 18F-FDG PET/CT联合临床建立多模态影像组学模型预测局部晚期头颈部鳞癌对于新辅助免疫联合化学治疗的反应 |
| Multimodal models established combined 18F-FDG PET/CT radiomics with clinical data for evaluating response of locally advanced head and neck squamous cell carcinoma to neoadjuvant immuno-chemotherapy |
| 投稿时间：2024-04-04 修订日期：2024-05-31 |
| DOI：10.13929/j.issn.1003-3289.2024.10.008 |
| 中文关键词: 癌,鳞状细胞 头部 颈 新辅助治疗 正电子发射断层显像 体层摄影术,X线计算机 氟脱氧葡萄糖F18 影像组学 |
| 英文关键词:carcinoma,squamous cell head neck neoadjuvant therapy positron-emission tomography tomography,X-ray computed fluorodeoxyglucose F18 radiomics |
| 基金项目: |
|
| 摘要点击次数: 482 |
| 全文下载次数: 172 |
| 中文摘要: |
| 目的 观察以18F-FDG PET/CT影像组学联合临床资料建立的多模态模型预测局部晚期头颈部鳞癌(LA-HNSCC)对新辅助免疫联合化学治疗(化疗)反应的价值。方法 回顾性纳入213例LA-HNSCC,按8:2比例将其随机分为训练集(n=170)与测试集(n=43)。基于训练集18F-FDG PET/CT提取并遴选最优影像组学特征,序贯以单因素及多因素logistic回归分析遴选独立临床预测因素;分别以自适应提升(AdaBoost)、决策树、朴素贝叶斯、随机森林(RF)、支持向量机(SVM)及极限梯度提升(XGBoost)算法分别构建影像组学模型、临床模型及二者联合的多模态模型,绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评估各模型预测LA-HNSCC对新辅助免疫联合化疗反应的效能,以决策曲线分析(DCA)评估各模型临床净收益。结果 共获得110个与新辅助免疫联合化疗反应相关的LA-HNSCC 18F-FDG PET/CT影像组学特征;CD4/CD8比值为其临床独立预测因素。基于AdaBoost及XGBoost算法所获模型的预测效能较高且较为稳定;其中,多模态模型的效能(AUC=0.943、0.930)优于影像组学模型(AUC=0.939、0.925)及临床模型(AUC=0.903、0.910)(P均＜0.05)。各多模态模型在测试集中带来的临床净收益优于或类于影像学模型与临床模型。结论 联合18F-FDG PET/CT影像组学与CD4/CD8比值建立的多模态模型预测LA-HNSCC对新辅助免疫联合化疗反应的效能优于单一模型。 |
| 英文摘要: |
| Objective To observe the value of multimodal models established combined 18F-FDG PET/CT radiomics with clinical data for evaluating response of locally advanced head and neck squamous cell carcinoma (LA-HNSCC) to neoadjuvant immuno-chemotherapy. Methods Totally 213 LA-HNSCC patients were retrospectively enrolled and randomized into training set (n=170) and test set (n=43) at the ratio of 8:2. Radiomics features of tumors on 18F-FDG PET/CT were extracted and selected from training set, and the independent clinical predictors were screened with sequential univariate and multivariate logistic regressions. Radiomics models, clinical models and combined multimodal models were constructed using different algorithms, including adaptive boosting (AdaBoost), decision tree, naive Bayes, random forest (RF), support vector machine (SVM) and extreme gradient boosting (XGBoost), respectively. The receiver operating characteristic (ROC) curves were drawn, and the area under the curves (AUC) were calculated to assess the efficacy of each model for predicting the response of LA-HNSCC to neoadjuvant immuno-chemotherapy, and the decision curve analysis (DCA) was performed to explore the net benefit of each model. Results Totally 110 radiomics features were selected, and CD4/CD8 ratio was the independent clinical predictor of the response of LA-HNSCC to neoadjuvant immuno-chemotherapy. Models based on AdaBoost and XGBoost algorithms had high and stable efficacy for predicting tumor response to neoadjuvant immuno-chemotherapy, among which the multimodal models had better performance (AUC=0.943, 0.930) than radiomics models (AUC=0.939, 0.925) and clinical models (AUC=0.903, 0.910) in test set (all P＜0.05). Conclusion Multimodal models established combined 18F-FDG PET/CT radiomics with CD4/CD8 ratio were more effective for predicting response of LA-HNSCC to neoadjuvant immuno-chemotherapy than any single model. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
