基于CT平扫影像组学模型预测食管癌淋巴结转移

Radiomics model based on plain CT for predicting lymph node metastasis of esophageal cancer

目的评估术前基于胸部CT平扫影像组学模型预测食管癌患者淋巴结转移的价值。方法回顾性分析368例经术前内镜活检及术后病理确诊的食管癌患者,其中100例淋巴结转移、268例无淋巴结转移,按比例3∶1将其分为训练组(包括201例无淋巴结转移和75例淋巴结转移)和验证组(67例无淋巴结转移和25例淋巴结转移)。自胸部CT中提取食管癌病灶的影像组学特征,并以最小绝对收缩和选择算子(LASSO)回归进行降维,筛选与食管癌淋巴结转移关联度高的特征;采用支持向量机构建预测淋巴结转移的影像组学模型,并以受试者工作特征(ROC)曲线分析模型的诊断效能。结果共提取1046个组学特征参数,经LASSO降维筛选出11个特征参数用于建立预测淋巴结转移模型。影像组学模型预测训练组淋巴结转移的曲线下面积(AUC)为0.84,敏感度为84.00%,特异度为75.12%,准确率为77.54%;于验证组的AUC为0.82,敏感度为80.00%,特异度为77.61%,准确率为78.26%。结论术前基于胸部CT平扫影像组学模型预测食管癌患者淋巴结转移具有较高价值。

Objective To explore the value of radiomics model based on chest plain CT in preoperative predicting lymph node metastasis of esophageal cancer.Methods Data of 368 patients with esophageal cancer diagnosed with preoperative endoscopic biopsy and postoperative pathology were retrospectively analyzed including 100 cases with lymph node metastasis and 268 cases without lymph node metastasis.The patients were classified as lymph node metastasis and non-lymph node metastasis and were divided into training group(201 cases without lymph node metastasis and 75 cases with lymph node metastasis)and validation group(67 cases without lymph node metastasis and 25 cases with lymph node metastasis)at the ratio of 3∶1.Then radiomics features of esophageal cancer on chest CT were extracted,the least absolute shrinkage and selection operator(LASSO)regression was used for dimensionality reduction,and the features with high correlation with lymph node metastasis of esophageal cancer were selected.Support vector machine classifier was used to construct a radiomics model for predicting lymph node metastasis.Then the diagnostic efficacy of the radiomics model was estimated with the receiver operating characteristic(ROC)curve.Results A total of 1046 radiomics characteristic parameters were extracted.After dimension reduction with LASSO,11 characteristic parameters were selected to establish model of predicting lymph node metastasis.In training group,the area under the curve(AUC)of the model for predicting lymph node metastasis was 0.84,the diagnostic sensitivity was 84.00%,the specificity was 75.12%and the accuracy was 77.54%.In validation group,AUC of the radiomics model was 0.82,with sensitivity of 80.00%,specificity of 77.61%.Conclusion Radiomics model based on plain CT had good diagnostic value for pre-operative predicting lymph node metastasis of esophageal cancer.