摘要
目的3.0 T MRI影像数据具有评估局部进展期直肠癌(locally advanced rectal cancer,LARC)术前新辅助放化疗(neoadjuvant chemoradiotherapy,nCRT)疗效临床价值,但是构建模型多种机器学习间的比较并没有被探究过。我们将比较4种常用的机器学习方法在评估直肠癌新辅助治疗疗效临床价值的效能。材料与方法回顾性分析2021年9月至2023年1月于哈尔滨医科大学附属第二医院就诊经病理检查证实的LARC并行nCRT的病例共160例,按8∶2比例分为训练集及验证集。分别建立支持向量机(support vector machines,SVM)、朴素贝叶斯(naive Bayes,NB)、神经网络(neural network,NN)、卷积神经网络(convolutional neural networks,CNN)四种分类器模型,采用DeLong检验比较受试者工作特征(receiver operating characteristic,ROC)曲线的差异,对4种分类器诊断效能进行评估比较。结果2组患者年龄、性别差异无统计学意义(P>0.05)。通过最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法得到9个与治疗效果分组相关的特征,9个特征在病理完全缓解(pathological complete response,pCR)/病理非完全缓解(non-pathological complete response,non-pCR)中有差异,但差异不具有统计学意义(P>0.05)。SVM模型在训练集的ROC曲线下面积为0.9150,评估nCRT效果最为显著。结论基于磁共振高分辨T2WI的纹理特征,通过SVM、NB、NN、CNN分类器模型可以对直肠癌nCRT疗效进行评估,其中SVM分类器模型诊断效能最佳。基于高分辨T2WI的影像组学可以评估LARC患者nCRT治疗效果。
Objective:3.0 T MRI data has clinical value in evaluating the efficacy of neoadjuvant therapy for locally advanced rectal cancer(LARC),but the comparison between multiple machine learning models has not been explored.We will compare the efficacy of four commonly used machine learning methods in evaluating the clinical value of neoadjuvant chemoradiotherapy(nCRT)for LARC.Materials and Methods:A total of 160 LARC patients who were diagnosed and confirmed by pathological examination at the Second Affiliated Hospital of Harbin Medical University from September 2021 to January 2023,underwent nCRT.They were divided into a training set and a validation set in an 8∶2 ratio.Establish four classifier models:support vector machine(SVM),naive Bayes(NB),convolutional neural networks(CNN)and neural network(NN),and use DeLong test to compare the differences in receiver operating characteristic(ROC)curves.Evaluate and compare the diagnostic performance of four classifiers.Results:There was no statistically significant difference in age and gender between the two groups of patients(P>0.05).Nine features related to treatment efficacy grouping were obtained through least absolute shrinkage and selection operator(LASSO),and there were differences between pathological complete response(pCR)non-pathological complete response(non-pCR)groups,but the differences were not statistically significant(P>0.05).The area under the ROC curve of SVM in the training set is 0.9150,which indicates the most significant evaluation of the efficacy of nCRT and chemotherapy.Conclusions:Based on the texture features of high-resolution T2WI MRI,SVM,NB,NN,and CNN classifier models can be used to evaluate the effectiveness of colorectal cancer nCRT treatment.SVM classifier models have the best diagnostic performance,and imaging omics based on high-resolution T2WI can evaluate the effectiveness of nCRT treatment in LARC patients.
作者
胡鸿博
赵升
姜昊
姜慧杰
蔺雪
张莹
HU Hongbo;ZHAO Sheng;JIANG Hao;JIANG Huijie;LIN Xue;ZHANG Ying(Department of Radiology,the Second Affiliated Hospital of Harbin Medical University,Harbin 150086,China;Department of MedicalGenetics,Medicine Basic sciences,Harbin Medical University,Harbin 150086,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2023年第11期77-83,共7页
Chinese Journal of Magnetic Resonance Imaging
基金
国家自然科学基金项目(编号:62171167)。
关键词
直肠癌
磁共振成像
影像组学
机器学习
支持向量机
rectal cancer
magnetic resonance imaging
imaging omics
machine learning
support vector machine
