实用医学杂志 ›› 2021, Vol. 37 ›› Issue (14): 1872-1878.doi: 10.3969/j.issn.1006⁃5725.2021.14.020

• 医学检查与临床诊断 • 上一篇    下一篇

基于CT影像组学诺模图预测头颈部恶性肿瘤淋巴结转移

李羚1, 胡大涛2, 夏春华2, 李红霞3    

  1. 安徽医科大学第三附属医院合肥市第一人民医院1 耳鼻咽喉科,2 放射科,3 肿瘤科(合肥 230061)

  • 出版日期:2021-07-25 发布日期:2021-07-25
  • 基金资助:
    2020 年度省级引进境外人才项目(编号:S2020340028)

Prediction of lymph node metastasis of head and neck malignant tumor based on CT radiomics nomogram

LI Ling*,HU Datao,XIA Chunhua,LI Hongxia.   

  1. *Department of OtolaryngologyHefei First People′s Hospitalthe Third Affiliated Hospital of Anhui Medical UniversityHefei 230061China
  • Online:2021-07-25 Published:2021-07-25

摘要:

目的 探讨基于 CT 的影像组学诺模图预测头颈部恶性肿瘤淋巴结转移的诊断价值。 方法 回顾分析 2017 1 月至 2020 10 月就诊于合肥市第一人民医院,有明确淋巴结病理且治疗 前有质量较好的 CT 平扫及增强的头颈部恶性肿瘤患者 75 例,收集 201 枚淋巴结,将 2017 1 月至 2019 11 月的 113 枚淋巴结归为训练组,2019 12 月至 2020 10 月的 88 枚淋巴结归为测试组。运用 Mazda 软件提取影像组学特征,使用 LASSO 法降维,建立影像组学标签。经过多因素逻辑回归筛选, 将平扫 CT 值、增强动脉期 CT 值分别与影像组学标签建立影像组学诺模图预测模型,与影像组学标 签组和 CT 征象诊断模型进行对比,使用受试者工作特征(receiver operator characteristic,ROC)曲线、曲线 下面积(area under cuive,AUC)和校准曲线校准度对诊断效能进行对比评价。结果 在每个 CT 扫描 的颈部淋巴结区域内提取 298 个三维特征,通过降维,筛选出 6 ~ 7 个最具相关性的影像组学特征建立 影像组学标签。影像组学标签+平扫 CT 值诺模图预测模型在训练组和测试组 AUC 为(0.955,0.885), 影像组学标签+增强 CT 诺模图值预测模型在训练组和测试组 AUC 为(0.982,0.920),均大于只有影像组学标签模型为[AUC(0.940,0.905),P<0.05],平扫 CT 值+平扫影像组学标签、增强 CT 值+影像组 学标签、影像组学标签和 CT 征象诊断模型的准确率分别是 0.920、0.930、0.801 0.716;敏感性分别是 0.925、0.925、0.877 0.792;特异性分别为 0.916、0.926、0.716 0.632;且具有满意的校准度。结论 基于 CT 的影像组学诺模图对头颈部恶性肿瘤淋巴结转移的诊断效能良好,具有简便、安全、可靠的 临床应用价值。

关键词: 头颈部淋巴结,  , 计算机体层成像,  , 影像组学,  , 诺模图

Abstract:

Objective To explore the diagnostic value of CT ⁃ based radiomics nomogram in predicting lymph node metastasis of head and neck cancer. Methods A retrospective analysis was performed on the patients who visited the first people′s Hospital of Hefei from January 2017 to October 2020. They had clear lymph node pathology and had good quality CT scan before treatment. A total of 201 lymph nodes were collected from 75 patients with head and neck malignant tumors. One hundred and thirteen lymph nodes from January 2017 to November 2019 were enrolled into the training group,and 88 lymph nodes from December 2019 to October 2020 were enrolled into the test group. Mazda software was used to extract the radiomics features,and lasso method was used to reduce the dimension and establish the radiomics tags. After the multivariate logistic regression screening,the plain scan CT value,enhanced arterial phase CT value and radiomics tags were seperately used to establish radiomics nomogram diagnostic models. Receiver operator characteristic(ROC)curve,area under curve(AUC)and calibration curves were used to evaluate the diagnostic efficacy. Results In each CT scan,298 3D features were extracted from the cervical lymph node region. Through dimensionality reduction,6 ~ 7 most relevant radiomics features were selected to establish the radiomics tags. The AUC of nomograph prediction model of radiomics tags plus plain CT value in training set and test set was(0.955,0.885),and that of radiomics tags plus enhanced CT value in training set and test set was(0.982,0.920),which were higher than that in the only radiomics tags model(0.940,0.905)(P < 0.05). The accuracy of plain CT value plus plain radiomics tags,enhanced CT value plus radiomics tags,radiomics tags and CT sign diagnosis model were 0.920,0.930,0.801 and 0.716,respectively,and the sensitivity was 0.925,0.925,0.877 and 0.792,respectively,and the specificity was 0.916,0.926,0.716 and 0.632,respectively. At the same time,it had satisfactory calibration. Conclusion CT⁃based radiomics nomogram has a good diagnostic effect on lymph node metastasis of head and neck cancer,which is simple,safe and reliable.

Key words:

head and neck lymph nodes, computed tomography, radiomics, nomogram