| 摘 要: 本文将影像组学的方法和机器学习算法结合起来，对脑部胶质瘤进行分级预测。利用BraTS2019公开数据集，从多模态MRI图像中分别提取肿瘤的448 维影像组学特征：肿瘤形态学特征、一阶灰度特征、纹理特征等；然后通过最小绝对收缩和选择算子(Lasso)算法筛选出15 个最佳的影像组学特征；最后根据筛选出的最佳特征集，利用随机森林分类算法构建脑部胶质瘤的分级预测模型。基于机器学习建立的模型在训练组患者中预测胶质瘤级别的准确率达到95.6％，ROC曲线下面积(AUC)达到0.99；在验证组患者中预测胶质瘤级别的准确率达到89.3％，AUC达到0.96。可见，基于机器学习算法，利用影像组学的方法可以对脑部肿瘤的高低级别进行准确的预测和分类。 |
| 关键词: 肿瘤分级；影像组学；机器学习；随机森林 |
|
中图分类号: TP39
文献标识码: A
|
|
| Research on Grading Model for Brain Glioma based on Radiomics and Machine Learning |
|
WANG Junxiu
|
(Taiyuan Institute of Technology, Taiyuan 030008, China )
wangjx@tit.edu.cn
|
| Abstract: This paper proposes to combine radiomics and machine learning algorithm to classify and predict the brain glioma. Based on BraTS2019 public dataset, 448-dimensional radiomics features of tumors are extracted from multimodal MRI (Magnetic Resonance Imaging) images, including tumor morphological features, first-order grayscale features, and texture features, etc. Then 15 best radiomics features are screened through the least absolute shrinkage and selection operator (Lasso) algorithm. Finally, according to the best screened feature set, the random forest classification algorithm is used to construct the brain glioma grading prediction Model. The accuracy of machine learning-based model is 95.6% and the area under the ROC (AUC) is 0.99 in the training group, and 89.3% and 0.96 in the validation group, respectively. Application of machine learning algorithm and radiomics realizes accurate prediction and classification of brain glioma level. |
| Keywords: brain glioma grading; radiomics; machine learning; random forest |
