文献速递胸部CT

基于影像组学列线图构建磨玻璃结节手术切除的预测模型

Establishment of a Predictive Model for Surgical Resection of Ground-Glass Nodules.
PubMed 文献 · J Am Coll Radiol · 2018-11-19

本研究收集239例磨玻璃结节患者的CT图像,其中160例为训练集、79例为验证集,采用最小绝对收缩和选择算子算法筛选影像组学特征并构建影像组学列线图预测浸润性肺腺癌以指导手术切除;联合影像组学与临床信息的列线图在训练集AUC 0.831、验证集AUC 0.816,预测能力最佳。

要点

  • 联合影像组学与临床信息的列线图预测浸润性肺腺癌,训练集AUC 0.831、验证集AUC 0.816。
  • 仅用临床信息的模型预测效能相对较低,AUC 0.711。
  • 决策曲线分析显示影像组学列线图具有一定临床价值。

核心结论

  • 整合影像组学特征与临床信息构建的影像组学列线图预测模型有助于临床医生把控手术节点、减少过度治疗的发生。

研究设计与样本

  • 回顾性收集239例磨玻璃结节患者的CT图像,160例纳入训练集构建预测模型,79例纳入验证集验证模型;采用最小绝对收缩和选择算子算法筛选影像组学特征并构建影像组学标签,基于影像组学列线图构建浸润性肺腺癌手术切除的预测模型。

诊断标准与关键数值

  • 7个影像组学特征与浸润性肺腺癌的存在显著相关(P < .01),为独立预测因子。
  • 影像组学特征构建的预测模型在训练集AUC 0.792(95%CI:0.720-0.864),验证集AUC 0.773(95%CI:0.668-0.877)。
  • 仅用临床信息构建的预测模型效能相对较低,AUC 0.711(95%CI:0.634-0.787)。
  • 影像组学特征联合临床信息构建的影像组学列线图预测能力与校准最佳,训练集AUC 0.831(95%CI:0.765-0.897),验证集AUC 0.816(95%CI:0.724-0.909)。
  • 决策曲线分析显示影像组学列线图具有一定临床价值。

临床应用与阅片要点

  • 作者认为,基于影像组学列线图整合影像组学特征与临床信息构建的浸润性肺腺癌手术切除预测模型,可帮助临床医生把控手术节点、减少过度治疗的发生。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

PURPOSE: To establish a predictive model for surgical resection of invasive pulmonary adenocarcinoma (IPA) presenting as ground-glass nodules (GGNs) based on a radiomics nomogram.

METHODS: The CT images of 239 patients with GGNs were collected, of which 160 cases were included in the training set to construct the predictive model and 79 cases were included in the validation set to verify the established predictive model. The least absolute shrinkage and selection operator algorithm was used to select the radiomic features and construct the radiomics tagging. The predictive model for the surgical resection of IPA was constructed using the radiomics nomogram.

RESULTS: The presence of IPA showed significant correlations with seven radiomics features (P < .01), which were the independent predictors. The predictive model constructed using the radiomics features performed well on the training set (area under the curve [AUC] 0.792, 95% confidence interval [CI]: 0.720-0.864) and the validation set (AUC 0.773, 95% CI: 0.668-0.877). The predictive model constructed using the clinical information alone was relatively less effective (AUC 0.711, 95% CI: 0.634-0.787). The predictive model constructed by integrating the radiomics features into the clinical information using the radiomics nomogram showed the best predictive ability and calibration in the training set (AUC 0.831, 95% CI: 0.765-0.897) and the validation set (AUC 0.816, 95% CI: 0.724-0.909). Decision curve analysis showed that radiomics nomogram has a certain clinical value.

CONCLUSION: The predictive model for surgical resection of IPA constructed by integrating the radiomics features and the clinical information based on the radiomics nomogram can help clinicians control the operative node and reduce the occurrence of overtreatment.

来源:J Am Coll Radiol,查看原文。版权归原作者所有。

引用

  1. Liu CL, Zhang F, Cai Q, et al. Establishment of a Predictive Model for Surgical Resection of Ground-Glass Nodules. J Am Coll Radiol. 2018. PMID: 30466899. doi:10.1016/j.jacr.2018.09.043

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。