文献速递神经MR

机器学习识别胶质瘤患者IDH突变的诊断准确性及潜在协变量:一项Meta分析

Diagnostic accuracy and potential covariates for machine learning to identify IDH mutations in glioma patients: evidence from a meta-analysis.
PubMed 文献 · Eur Radiol · 2020-03-19

该系统评价与Meta分析纳入9篇原始文献共996例胶质瘤患者,评估机器学习预测IDH突变的诊断准确性。验证集AUC为89%(95%CI 86-92%),汇总敏感度和特异度分别为87%和90%;训练集AUC为93%(95%CI 91-95%)。临床信息、MRI序列和胶质瘤分级是影响诊断特异度的主要因素。

要点

  • 机器学习预测胶质瘤IDH突变表现优异,训练集AUC 93%、验证集AUC 89%。
  • 联合临床与影像特征较单用影像特征提高敏感度(90% vs. 83%)和特异度(90% vs. 82%)。
  • 使用常规MRI序列的机器学习模型特异度高于联合常规与先进MRI序列(89% vs. 85%)。

诊断标准与关键数值

  • 训练集机器学习预测IDH突变的AUC为93%(95%CI 91-95%),验证集AUC为89%(95%CI 86-92%)。
  • 训练集汇总敏感度87%(95%CI 82-91%)、特异度88%(95%CI 83-92%);验证集汇总敏感度87%(95%CI 76-93%)、特异度90%(95%CI 72-97%)。
  • 训练集亚组分析中,联合临床与影像特征较单用影像特征的敏感度90% vs. 83%、特异度90% vs. 82%。
  • 使用常规MRI序列的机器学习模型预测IDH突变的特异度高于联合常规与先进MRI序列(89% vs. 85%)。
  • 机器学习对高级别胶质瘤的诊断表现优于低级别胶质瘤。

临床应用与阅片要点

  • 机器学习模型可用于预测胶质瘤IDH突变,联合临床与影像特征可能提高诊断敏感度和特异度。
  • 使用常规MRI序列的模型在预测IDH突变时特异度高于联合常规与先进MRI序列的模型。
  • 机器学习对高级别胶质瘤的预测表现优于低级别胶质瘤。

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

OBJECTIVES: To assess the diagnostic accuracy of machine learning (ML) in predicting isocitrate dehydrogenase (IDH) mutations in patients with glioma and to identify potential covariates that could influence the diagnostic performance of ML.

METHODS: A systematic search of PubMed, Web of Science, and the Cochrane library up to 1 August 2019 was conducted to collect all the articles investigating the diagnostic performance of ML for prediction of IDH mutation in glioma. The search strategy combined synonyms for 'machine learning', 'glioma', and 'IDH'. Pooled sensitivity, specificity, and their 95% confidence intervals (CIs) were calculated, and the area under the receiver operating characteristic curve (AUC) was obtained.

RESULTS: Nine original articles assessing a total of 996 patients with glioma were included. Among these studies, five divided the participants into training and validation sets, while the remaining four studies only had a training set. The AUC of ML for predicting IDH mutation in the training and validation sets was 93% (95% CI 91-95%) and 89% (95% CI 86-92%), respectively. The pooled sensitivity and specificity were, respectively, 87% (95% CI 82-91%) and 88% (95% CI 83-92%) in the training set and 87% (95% CI 76-93%) and 90% (95% CI 72-97%) in the validation set. In subgroup analyses in the training set, the combined use of clinical and imaging features with ML yielded higher sensitivity (90% vs. 83%) and specificity (90% vs. 82%) than the use of imaging features alone. In addition, ML performed better for high-grade gliomas than for low-grade gliomas, and ML that used conventional MRI sequences demonstrated higher specificity for predicting IDH mutation than ML using conventional and advanced MRI sequences.

CONCLUSIONS: ML demonstrated an excellent diagnostic performance in predicting IDH mutation of glioma. Clinical information, MRI sequences, and glioma grade were the main factors influencing diagnostic specificity.

KEY POINTS: • Machine learning demonstrated an excellent diagnostic performance for prediction of IDH mutation in glioma (the pooled sensitivity and specificity were 88% and 87%, respectively). • Machine learning that used conventional MRI sequences demonstrated higher specificity in predicting IDH mutation than that based on conventional and advanced MRI sequences (89% vs. 85%). • Integration of clinical and imaging features in machine learning yielded a higher sensitivity (90% vs. 83%) and specificity (90% vs. 82%) than that achieved by using imaging features alone.

来源:Eur Radiol,查看原文。版权归原作者所有。

引用

  1. Zhao J, Huang Y, Song Y, et al. Diagnostic accuracy and potential covariates for machine learning to identify IDH mutations in glioma patients: evidence from a meta-analysis. Eur Radiol. 2020. PMID: 32193643. doi:10.1007/s00330-020-06717-9

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。