南方医科大学学报

南方医科大学学报 ›› 2024, Vol. 44 ›› Issue (1): 194-200.doi: 10.12122/j.issn.1673-4254.2024.01.23

• • 上一篇    

基于磁共振图像机器学习放射组学模型预测脑胶质瘤的强化

何慧珊，郭二嘉，蒙文仪，王 彧，王 雯，何文乐，吴元魁，阳 维   

  1. 南方医科大学南方医院（第一临床医学院），广东 广州 510515；广东三九脑科医院影像中心，广东 广州 510515；南方医科大学生物医学工程学院，广东 广州 510515
  • 发布日期:2024-01-19

Predicting cerebral glioma enhancement pattern using a machine learning-based magnetic resonance imaging radiomics model

HE Huishan, GUO Erjia, MENG Wenyi, WANG Yu, WANG Wen, HE Wenle, WU Yuankui, YANG Wei   

  1. Nanfang Hospital/First School of Clinical Medicine, Southern Medical University, Guangzhou 510515, China; Medical Imaging Center, Guangdong 999 Brain Hospital, Guangzhou 510515, China; School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China
  • Published:2024-01-19

RichHTML

26

PDF

316

摘要： 目的 开发一个基于T2液体衰减反转恢复序列（T2-FLAIR）图像准确预测胶质瘤MRI强化模式的机器学习放射组学模型，为优化胶质瘤患者的MRI检查流程提供潜在的理论依据。方法 回顾性收集385例手术病理确诊的脑胶质瘤的术前MRIT2-FLAIR图像，根据强化模式分成强化和无强化两类，在训练组（201例）基于高斯过程、线性回归、线性回归最小绝对收缩和选择算子、支持向量机、线性判别分析和朴素贝叶斯这6种分类器分别建立预测胶质瘤强化模式的组学模型，并在内部验证组（85例）和外部验证组（99例）进行测试。应用受试者操作特征曲线评估其预测性能。结果 以高斯过程作为分类器的由15个放射组学特征组成的预测模型在训练组和内部验证组均具有最高的预测性能，其曲线下面积分别是0.88（95% CI：0.81，0.94）和0.80（95% CI：0.71，0.88），在外部验证组的曲线下面积、敏感性、特异性、阳性预测值、阴性预测值分别是0.81（95% CI：0.71，0.90）、0.98、0.61、0.76、0.96。结论 基于T2-FLAIR的机器学习放射组学模型可准确预测胶质瘤的MRI强化模式。

关键词: 脑肿瘤；磁共振成像；机器学习；放射组学

Abstract: Objective To establish a machine learning radiomics model that can accurately predict MRI enhancement patterns of glioma based on T2 fluid attenuated inversion recovery (T2-FLAIR) images for optimizing the workflow of magnetic resonance imaging (MRI) examinations of glioma patients. Methods We retrospectively collected preoperative MR T2-FLAIR images from 385 patients with pathologically confirmed glioma, who were divided into enhancing and non-enhancing groups according to the enhancement pattern. Predictive radiomics models were established using Gaussian Process, Linear Regression, Linear Regression-Least absolute shrinkage and selection operator, Support Vector Machine, Linear Discriminant Analysis or Naive Bayes as the classifiers in the training cohort (n=201)and tested both in the internal (n=85) and external validation cohorts (n=99). The receiver-operating characteristic curve was used to assess the predictive performance of the models. Results The predictive model constructed based on 15 radiomics features using Gaussian Process as the classifier had the best predictive performance in both the training cohort and the internal validation cohort, with areas under the curve (AUC) of 0.88 (95% CI: 0.81-0.94) and 0.80 (95% CI: 0.71-0.88), respectively. In the external validation cohort, the model showed an AUC of 0.81 (95% CI: 0.71-0.90) with sensitivity, specificity, positive predictive value and negative predictive value of 0.98, 0.61, 0.76 and 0.96, respectively. Conclusion The T2-FLAIR-based machine learning radiomics model can accurately predict the enhancement pattern of gliomas on MRI.

Key words: brain tumor; magnetic resonance imaging; machine learning; radiomics