Objective: To investigate the spectral CT-based radiomics in predicting preoperatively the lymph node metastasis (LNM) of advanced gastric cancer. Methods: From January 2014 to October 2018, the spectral CT imaging and clinical data of 196 gastric adenocarcinoma patients confirmed by pathology in the First Affiliated Hospital of Zhengzhou University were retrospectively enrolled (training set and test set were randomly divided according to the ratio of 1∶1). These 196 patients include143 males and 53 females, aged from 28 to 81 years, with an average age of (59±11) years, and were divided into nodular metastasis group and non-metastasis group according to clinicopathological data. The spectral parameters were measured and calculated, and the CT-reported lymph node (LN) status from CT images were obtained. 273 radiomics features were extracted from the dual-phases CT images in different energy level (40, 65 and 100 keV) to build the radiomics signature respectively. Univariate analysis was used to compare the differences of spectral parameters and radiomics features between two groups, and then the significant indicators were put into multivariable logistic regression analysis to construct combined prediction model and radiomics nomogram. In addition, the performance of prediction model in training and test set were measured using the receiver operating characteristics (ROC) curves and were compared using DeLong test. Results: Both in training set and in test set, the iodine concentration (IC) of tumor in venous phase (VP) in nodular metastasis group were higher than that in non-metastasis group [training set: 22.98 (100 mg/L)>20.31 (100 mg/L), P=0.086; test set: 25.14 (100 mg/L)>21.07 (100 mg/L), P=0.009]. The CT-reported LN status showed significant differences between the two group (P<0.001, P=0.001). The radiomics signatures 40 keV-arterial phase, 65 keV-venous phase, IC-VP of tumor and CT-reported LN status were independent indicators for prediction of preoperative LNM of advanced gastric cancer in combined prediction model (P<0.05). The radiomics nomogram predicated LNM with an area under curve (AUC) and 95% confidence interval (CI) of 0.822 (0.739-0.906) in training set and 0.819(0.732-0.906) in test set, and there were no significant differences in AUC between two sets (P>0.05). Conclusions: The spectral CT-based radiomics can be used to quantitatively predict the LNM of advanced gastric cancer preoperatively.
目的: 探讨基于能谱CT的影像组学模型术前预测进展期胃癌淋巴结转移的价值。 方法: 回顾性收集2014年1月至2018年10月郑州大学第一附属医院共196例经手术病理确诊的胃腺癌患者的临床及能谱CT影像资料,其中男143例,女53例,年龄28~81(59±11)岁,按照1∶1的比例随机分为训练集和验证集,并根据病理资料细分为淋巴结转移组和非转移组。测量并计算胃腺癌能谱参数值并进行淋巴结CT评估。基于不同能级(40、65和100 keV)的双期图像提取273个影像组学特征并建立组学标签。对能谱参数和组学特征进行单因素分析,将差异有统计学意义的变量纳入多因素Logistic回归分析中构建联合预测模型并绘制诺模图。受试者工作特征(ROC)曲线分析用于评价预测模型诺模图的诊断性能,不同数据集之间的曲线下面积(AUC)比较采用Delong检验。 结果: 训练集和验证集中,淋巴结转移组的肿瘤静脉期碘基值均大于非转移组[训练集:22.98(100 mg/L)>20.31(100 mg/L),P=0.086;验证集:25.14(100 mg/L)>21.07(100 mg/L),P=0.009];淋巴结CT评估在两组间的差异均有统计学意义(P<0.001,P=0.001)。联合预测模型中组学标签40 keV-动脉期、65 keV-静脉期、肿瘤静脉期碘基值和淋巴结CT评估作为进展期胃癌淋巴结转移的独立预测指标(P<0.05)。ROC曲线分析中,诺模图预测淋巴结转移在训练集和验证集中对应的AUC及95%CI分别为0.822(0.739~0.906)和0.819(0.732~0.906),且两者差异无统计学意义(P>0.05)。 结论: 基于能谱CT的影像组学对术前定量预测进展期胃癌淋巴结转移状态具有较好的应用价值。.
Keywords: Lymph nodes; Radiomics; Spectral imaging; Stomach neoplasms.