基于磁共振影像组学和语义特征对高级别胶质瘤和转移瘤的鉴别研究

Differentiation of high-grade glioma and metastatic tumor based on MRI radiomics and semantic features

目的本研究旨在结合传统MRI序列及增强检查,提取多模态高通量影像组学特征并联合语义特征,使用不同的机器学习分类器构建不同的模型并绘制列线图来鉴别高级别胶质瘤(high-grade glioma,HGG)和单发性脑转移瘤(solitary brain metastasis,SBM)。材料与方法本研究对101名患者的多参数MR图像进行了回顾性分析,由两位资深医师标定肿瘤感兴趣区,然后对每个序列分别提取影像组学特征后进行组合,共提取428组影像组学特征。为消除人为标定差异,进行组内相关系数一致性检验,并运用最大相关最小冗余算法选取最具相关性的特征,然后进一步通过最小绝对收缩和选择算子算法筛除冗余特征。本研究采用支持向量机、逻辑回归、随机森林及K近邻四种算法建立分类模型。结合放射科医生评估的七项语义特征,通过卡方检验和多因素分析去除差异无统计学意义的语义特征。然后结合组学特征建立综合模型并绘制列线图。最终,评价各模型的诊断能力,以确定最优分类器。结果HGG及SBM患者建立的影像组学模型中LR的受试者工作特征曲线下面积(area under the curve,AUC)值最高,训练集与测试集分别为0.90和0.90。语义特征建立的模型中随机森林模型性能最好,训练集和测试集AUC分别为0.82和0.87。语义特征联合影像组学评分后采用逻辑回归建立的模型性能最好,训练集和测试集AUC分别为0.91和0.92。结论本研究使用影像组学机器学习分类器并联合其他图像语义特征绘制列线图对HGG及SBM进行鉴别,这是一种非侵入性方法,具有较好的准确性,为临床决策和实践提供了较大的帮助。

Objective:To combine traditional MRI sequences and enhancement scans,extract multimodal high-throughput radiomics features along with semantic features,and use different learning classifiers to construct various models and draw Normogragh for the differentiation of high-grade glioma(HGG)and solitary brain metastasis(SBM).Materials and Methods:This study retrospectively analyzed multiparametric MRI images of 101 patients.Tumor region of interest(ROI)were delineated by two experienced physicians,and 107 sets of radiomic features for each sequence were extracted using the Pyradiomics software package.To eliminate variability in manual delineation,an intraclass correlation coefficient(ICC)consistency test was carried out.The features with the highest relevance were selected using the maximum relevance minimum redundancy algorithm,and then redundant features were further eliminated using the least absolute shrinkage and selection operator method.Classification models were established using four algorithms:support vector machine,logistic regression,random forest,and K-nearest neighbors.Combining seven semantic features evaluated by radiologists,chi-square test and multivariate analysis were used to remove semantically irrelevant features.Then,a comprehensive model incorporating both radiomics and semantic features was formed and illustrated using nomogram.Finally,the diagnostic capability of each model was evaluated to determine the optimal classifier.Results:Among the radiomics models for HGG and SBM patients,the model with the highest area under the curve(AUC)value was logistic regression,with AUC values of 0.90 for both the training set and test set.In models constructed using semantic features,the random forest model exhibited the best performance,with AUC values of 0.82 and 0.87 for the training and test sets,respectively.After combining semantic features with radiomics scores,the model constructed using logistic regression demonstrated optimal performance,with AUC values of 0.91 and 0.92 for the training and test sets,respectively.Conclusions:The non-invasive approach proposed in this study that utilizes radiomics machine learning classifiers and combines image semantic features to draw nomogram for differentiating between HGG and SBM,demonstrates good accuracy and provides significant assistance for clinical decision-making and practice.