基于磁共振影像组学列线图预测中晚期鼻咽癌放化疗疗效

Predicting the efficacy of chemoradiotherapy for advanced nasopharyngeal carcinoma based on MRI radiomic nomogram

目的探讨基于多模态磁共振的影像组学特征结合临床信息构建的列线图在中晚期鼻咽癌临床放化疗疗效中的预测价值。方法回顾性分析160例经病理证实为鼻咽癌的初诊患者的影像及临床资料。按照7:3的比例将患者分为训练组(n=112)与验证组(n=48)。在训练组提取T2加权脂肪抑制序列、T1加权增强序列和弥散加权成像序列图像的影像组学特征,经过最小绝对收缩和选择算子数据降维,筛选出最有效的特征构建影像组学预测模型。纳入相关的临床信息,利用Logistic逻辑回归,筛选最有价值的临床信息并构建临床信息模型;联合临床信息模型与组学特征模型构建联合模型,并构建列线图。通过ROC曲线及曲线下面积来评估各模型的诊断效能,通过决策曲线分析和校正曲线评估列线图的临床应用价值。结果联合2项临床信息和9项影像组学特征构建的临床-影像组学模型在训练组和验证组的曲线下面积分别为0.852(95%CI:0.765~0.940)、0.736(95%CI:0.574~0.898),显示出良好的预测效能。结论基于多模态磁共振的影像组学列线图在预测中晚期鼻咽期临床放化疗疗效中具有可行性,具有较好的临床应用价值。

Objective To investigate the predictive value of nomogram based on the radiomics features of multimodal magnetic resonance combined with clinical information in the efficacy of chemoradiotherapy in advanced nasopharyngeal carcinoma.Methods The imaging and clinical data of 160 patients with pathologically confirmed nasopharyngeal carcinoma at first diagnosis were retrospectively analyzed.Patients were divided into training group(n=112)and validation group(n=48)according to a ratio of 7:3.In the training group,the radiomics features of T2-weighted fat suppression sequence,T1-weighted enhancement sequence and diffusion-weighted imaging sequence images were extracted,and the most effective features were filtered to construct the radiomics prediction model after dimensionality reduction of the LASSO data.The relevant clinical information were incorporated,and the most valuable clinical information were screened a clinical information model constructed using logistics regression.The clinical information model was combined with the radiomics feature model to construct a combined model,and the nomogram was constructed.The diagnostic efficacy of each model was assessed by the ROC curve and area under the curve.The clinical application value of the nomogram was assessed by decision curve analysis and calibration curve.Results The clinic-radiomics model constructed by combining two clinical information and nine radiomics features showed good predictive efficacy with AUCs of 0.852(95%CI:0.765-0.940)and 0.736(95%CI:0.574-0.898)in the training and validation groups,respectively.Conclusion Multimodal MRI-based radiomics nomograms are feasible in predicting the efficacy of chemoradiotherapy in advanced nasopharyngeal carcinoma and have good clinical application.