基于LASSO回归的脑部胶质肿瘤MRI预测方法  

MRI prediction method for brain glioma based on LASSO regression

作　　者：徐坤财 张宁 廖益龙 XU Kuncai;ZHANG Ning;LIAO Yilong(School of Intelligent Engineering,Guiyang Institute of Information Science and Technology,Guiyang 550025,China)

机构地区：[1]贵阳信息科技学院智能工程学院,贵阳550025

出　　处：《智能计算机与应用》2024年第11期113-116,共4页Intelligent Computer and Applications

基　　金：贵州省青年科技人才成长项目(黔教技[2024]279号);贵州省科技计划项目(黔科合支撑[2024]一般178号);贵州省教育科学规划课题(2024C018)。

摘　　要：针对临床中不同级别的脑部胶质肿瘤预测准确性低的问题,本文提出基于LASSO回归的脑部胶质肿瘤MRI预测方法。首先,基于Pyradiomics提取BraTS2019数据集中的T1ce增强图像脑部胶质肿瘤的一阶统计特征、形态特征和纹理特征。然后,采用Mann-Whitney U检验对不同级别的脑部胶质肿瘤进行特征选择,并基于LASSO采用10折交叉验证构建临床预测模型。最后,通过对比SVM模型和ELM模型的诊断效能,结果显示基于LASSO回归的预测模型具有更优的诊断性能。Addressing the issue of low accuracy in predicting brain gliomas of different levels in clinical practice,this paper proposes a brain glioma MRI prediction method based on LASSO regression.Firstly,based on Pyradiomics,the first-order statistical features,morphological features,and texture features of brain gliomas in T1ce enhanced images from the BraTS2019 dataset are extracted.Then,Mann Whitney U test is used to select features for different levels of brain gliomas,and a clinical prediction model is constructed using 10 fold cross validation based on LASSO.Finally,by comparing the diagnostic performance of SVM model and ELM model,the results show that the prediction model based on LASSO regression has better diagnostic performance.

关 键 词：脑部胶质肿瘤 影像特征 统计特征 交叉验证 LASSO 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]