| 杨鑫淼,梁长华,吴青霞,潘犇,危涵羽,甄思雨,杨紫晴,王慧慧.增强CT影像组学联合临床及血液学指标诊断食管鳞状细胞癌淋巴结转移[J].中国医学影像技术,2024,40(11):1682~1687 |
| 增强CT影像组学联合临床及血液学指标诊断食管鳞状细胞癌淋巴结转移 |
| Contrast-enhanced CT radiomics combined with clinical and hematology indicators for diagnosing lymph node metastasis of esophageal squamous cell carcinoma |
| 投稿时间：2024-06-11 修订日期：2024-07-15 |
| DOI：10.13929/j.issn.1003-3289.2024.11.011 |
| 中文关键词: 食管肿瘤 淋巴转移 体层摄影术,X线计算机 影像组学 血液学 |
| 英文关键词:esophageal neoplasms lymphatic metastasis tomography, X-ray computed radiomics hematology |
| 基金项目:河南省重点研发与推广专项(科技攻关)项目(232102310262)、新乡市食管癌影像诊断与人工智能研究重点实验室。 |
|
| 摘要点击次数: 461 |
| 全文下载次数: 102 |
| 中文摘要: |
| 目的 观察增强CT影像组学联合临床及血液学指标诊断食管鳞状细胞癌(ESCC)淋巴结(LN)转移(LNM)的价值。方法 回顾性纳入218例ESCC,将pN1及pN2期视为LNM (n=90)、pN0期视为无LNM (n=128)。按8:2比例划分训练集(n=174)与测试集(n=44)。于训练集筛选可独立评估LNM的临床及LN影像学特征并建立临床-影像学模型,筛选可能与ESCC LNM相关的血液学指标并构建血液学模型;基于静脉期增强CT图像提取LN ROI及ESCC感兴趣容积(VOI)影像组学特征,筛选可能与LNM相关者并构建影像组学模型;以上述特征构建联合模型。采用受试者工作特征曲线及曲线下面积(AUC)评价各模型诊断LNM的效能,以决策曲线分析(DCA)评价临床净收益。结果 体质量指数(BMI)和目标LN内部坏死均为ESCC LNM的独立评估因素(P均＜0.05),相应临床-影像学模型诊断训练集和测试集LNM的AUC分别为0.747及0.687;以7个血液学指标构建的血液学模型在训练集和测试集的AUC分别为0.623及0.583;以10个LN影像组学特征及15个ESCC原发灶影像组学特征构建的影像组学模型在训练集和测试集的AUC分别为0.769及0.745;联合模型在训练集和测试集的AUC分别为0.822和0.739,在训练集优于其他模型(P均＜0.05),而在测试集与其他模型差异均无统计学意义(P均＞0.05)。DCA结果显示,0.55~0.80阈概率区间内,联合模型在测试集的净收益高于其他模型。结论 基于静脉期增强CT影像组学联合临床及血液学指标可较为有效地诊断ESCC LNM并可能提升临床净收益。 |
| 英文摘要: |
| Objective To observe the value of contrast-enhanced CT radiomics combined with clinical and hematology indicators for predicting lymph node (LN) metastasis (LNM) of esophageal squamous cell carcinoma (ESCC). Methods Totally 218 ESCC patients were retrospectively enrolled. Stage pN1 and pN2 were clustering as LNM (n=90), while stage pN0 were taken as non-LNM (n=128). The patients were divided into training set (n=174) and test set (n=44) at the ratio of 8 : 2. In training set, clinical and LN imaging features which could be used to independently judge LNM were screened and a clinical-imaging model was constructed. The hematological indicators that might be associated with ESCC LNM were screened, and a hematological model was constructed. Radiomics features in LN ROI and ESCC volume of interest (VOI) were extracted based on venous-phase contrast-enhanced CT images, and those might be associated with LNM were screened, and a radiomics model was constructed. Finally a combined model was constructed based on all the above features. The efficacy of each model for diagnosing LNM was evaluated with the area under the curve (AUC) of receiver operating characteristic curves, and the clinical net benefit was evaluated using decision curve analysis (DCA). Results Body mass index (BMI) and internal necrosis of target LN were both independent judging factors for ESCC LNM (both P＜0.05), and AUC of clinical-imaging model for diagnosing LNM in training and test sets was 0.747 and 0.687, respectively. Seven hematological indicators were included in hematological model, and AUC in training and test sets was 0.623 and 0.583, respectively. Ten LN radiomics features and 15 ESCC radiomics features were included in radiomics model, and AUC in training and test sets was 0.769 and 0.745, respectively. AUC of the combined model for diagnosing LNM in training and test sets was 0.822 and 0.739, respectively, better than other models in training set (all P＜0.05), but no significantly different in test set (all P＞0.05). DCA showed that combined model had higher net gain than the other models in 0.55—0.80 threshold probability interval. Conclusion Combined model based on venous-phase contrast-enhanced CT radiomics and clinical and hematology indicators could relatively effectively evaluate ESCC LNM, which might bring some promotions in clinical benefit. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
