PDF(Pubmed)
Abstract：
BACKGROUND: Xerostomia is one of the most common side effects in nasopharyngeal carcinoma (NPC) patients after chemoradiotherapy. To establish a Delta radiomics model for predicting xerostomia secondary to chemoradiotherapy for NPC based on magnetic resonance T1-weighted imaging (T1WI) sequence and evaluate its diagnostic efficacy.
METHODS: Clinical data and Magnetic resonance imaging (MRI) data before treatment and after induction chemotherapy (IC) of 255 NPC patients with stage III-IV were collected retrospectively. Within one week after CCRT, the patients were divided into mild (92 cases) and severe (163 cases) according to the grade of xerostomia. Parotid glands in T1WI sequence images before and after IC were delineated as regions of interest for radiomics feature extraction, and Delta radiomics feature values were calculated. Univariate logistic analysis, correlation, and Gradient Boosting Decision Tree (GBDT) methods were applied to reduce the dimension, select the best radiomics features, and establish pretreatment, post-IC, and Delta radiomics xerostomia grading predictive models. The receiver operating characteristic (ROC) curve and decision curve were drawn to evaluate the predictive efficacy of different models.
RESULTS: Finally, 15, 10, and 12 optimal features were selected from pretreatment, post-IC, and Delta radiomics features, respectively, and a xerostomia prediction model was constructed with AUC values of 0.738, 0.751, and 0.843 in the training set, respectively. Only age was statistically significant in the clinical data of both groups (P < 0.05).
CONCLUSIONS: Delta radiomics can predict the degree of xerostomia after chemoradiotherapy for NPC patients and it has certain guiding significance for clinical early intervention measures.
METHODS: Clinical data and Magnetic resonance imaging (MRI) data before treatment and after induction chemotherapy (IC) of 255 NPC patients with stage III-IV were collected retrospectively. Within one week after CCRT, the patients were divided into mild (92 cases) and severe (163 cases) according to the grade of xerostomia. Parotid glands in T1WI sequence images before and after IC were delineated as regions of interest for radiomics feature extraction, and Delta radiomics feature values were calculated. Univariate logistic analysis, correlation, and Gradient Boosting Decision Tree (GBDT) methods were applied to reduce the dimension, select the best radiomics features, and establish pretreatment, post-IC, and Delta radiomics xerostomia grading predictive models. The receiver operating characteristic (ROC) curve and decision curve were drawn to evaluate the predictive efficacy of different models.
RESULTS: Finally, 15, 10, and 12 optimal features were selected from pretreatment, post-IC, and Delta radiomics features, respectively, and a xerostomia prediction model was constructed with AUC values of 0.738, 0.751, and 0.843 in the training set, respectively. Only age was statistically significant in the clinical data of both groups (P < 0.05).
CONCLUSIONS: Delta radiomics can predict the degree of xerostomia after chemoradiotherapy for NPC patients and it has certain guiding significance for clinical early intervention measures.
摘要：
背景：口干症是鼻咽癌（NPC）患者放化疗后最常见的副作用之一。建立基于磁共振T1加权成像(T1WI)序列预测鼻咽癌放化疗后继发口干症的Delta影像组学模型，并评价其诊断效能。
方法：回顾性收集255例III-IV期NPC患者治疗前和诱导化疗（IC）后的临床资料和磁共振成像（MRI）数据。CCRT后一周内，根据口干症的程度分为轻度（92例）和重度（163例）。将IC前后T1WI序列图像中的腮腺划分为影像组学特征提取的感兴趣区域，并计算了Delta影像组学特征值。单变量逻辑分析，相关性，采用梯度提升决策树(GBDT)方法降维，选择最佳的影像组学功能，并建立预处理，后IC，和Delta放射组学口干症分级预测模型。绘制受试者工作特征（ROC）曲线和决策曲线，以评估不同模型的预测效果。
结果：最后，从预处理中选择15、10和12个最佳特征，后IC，和Delta影像组学特征，分别,并构建了口干症预测模型，训练集中的AUC值分别为0.738、0.751和0.843，分别。两组患者临床资料仅年龄差异有统计学意义(P<0.05)。
结论：Delta影像组学能够预测鼻咽癌患者放化疗后口干的程度，对临床早期干预措施具有一定的指导意义。
方法：回顾性收集255例III-IV期NPC患者治疗前和诱导化疗（IC）后的临床资料和磁共振成像（MRI）数据。CCRT后一周内，根据口干症的程度分为轻度（92例）和重度（163例）。将IC前后T1WI序列图像中的腮腺划分为影像组学特征提取的感兴趣区域，并计算了Delta影像组学特征值。单变量逻辑分析，相关性，采用梯度提升决策树(GBDT)方法降维，选择最佳的影像组学功能，并建立预处理，后IC，和Delta放射组学口干症分级预测模型。绘制受试者工作特征（ROC）曲线和决策曲线，以评估不同模型的预测效果。
结果：最后，从预处理中选择15、10和12个最佳特征，后IC，和Delta影像组学特征，分别,并构建了口干症预测模型，训练集中的AUC值分别为0.738、0.751和0.843，分别。两组患者临床资料仅年龄差异有统计学意义(P<0.05)。
结论：Delta影像组学能够预测鼻咽癌患者放化疗后口干的程度，对临床早期干预措施具有一定的指导意义。
