基于多视角的脑胶质瘤分级模型研究

GLIOMA GRADING BASED ON MULTI-VIEW MODEL

针对当前研究中的脑胶质瘤分级模型难以充分利用磁共振影像序列间的互补信息的问题,提出一种基于多视角的脑胶质瘤分级模型(MBMED)。以最大熵判别模型(MED)为基础分类器;利用AdaBoost对多视角脑胶质瘤数据集进行模型训练。训练时进行等权重初始化,通过优化误差率对样本和多个视角的权重进行迭代更新,输出基础分类器的组合,实现对脑胶质瘤的精准分级预测。在公开数据集BraTS2017和自建数据集GliomaHPPH2018上进行十折交叉验证实验,平均曲线下面积(AUC)分别为0.9485和0.9612。实验结果证明了该模型在脑胶质瘤分级中的有效性和准确性。

Present study model of glioma grading is difficult to make full use of complementary information between magnetic resonance image sequences.This paper proposes,a multi-view AdaBoost maximum entropy discrimination model(MBMED)to achieve accurate grading of glioma.In MBMED model,the maximum entropy discriminant model(MED)was used as the base learner.The multi-view AdaBoost algorithm was used and the model was trained by using the data set of multi-view glioma magnetic resonance.In the training process,the sample weights were initialized,and the weights of the samples and multiple views were iteratively updated according to the error rate.Finally,all the base learners were combined and output to predict the grading of glioma.The model was subjected to ten-fold cross-validation on the public data set BraTS2017 and the self-built data set GliomaHPPH2018.The area under the average curve(AUC)of the BraTS2017 data set and the GliomaHPPH2018 data set were 0.9485 and 0.9612 respectively.The experimental results demonstrate the validity and accuracy of the model in glioma grading.