首页|期刊导航|医疗卫生装备|基于多脑区影像组学指纹特征与机器学习的阿尔茨海默病分类方法研究

基于多脑区影像组学指纹特征与机器学习的阿尔茨海默病分类方法研究OA

Alzheimer's disease classification method based on multi-region radiomics fingerprint features and machine learning

中文摘要英文摘要

目的:为了实现对阿尔茨海默病(Alzheimer's disease,AD)的早期评估,提出一种基于多脑区影像组学指纹特征与机器学习的AD分类方法.方法:首先,回顾性选取阿尔茨海默病神经影像学倡议(Alzheimer's Disease Neuroimaging Initiative,ADNI)公开数据库中300例受试者的临床与影像资料,其中,轻度认知障碍(mild cognitive impairment,MCI)患者、正常老化(cognitively normal,CN)患者和AD患者各100例;其次,对所有原始T1WI结构像进行预处理;再次,选取11个与AD病程高度相关的核心脑区并提取多维影像组学特征;然后,采用最小绝对收缩与选择算子(least absolute shrinkage and selection operator,LASSO)回归剔除冗余变量后,筛选出核心影像组学指纹特征;最后,基于支持向量机(support vector machine,SVM)构建AD/CN、AD/MCI和MCI/CN 3个二分类模型.采用ROC曲线评估各模型的预测效能.结果:构建的SVM分类模型在3个分类任务中均展现出较高的诊断效能,其中,AD/CN模型的区分度最佳,训练集AUC为0.991,测试集AUC为0.925;MCI/CN模型与AD/MCI模型的分类效能略低,在测试集中的AUC分别为0.875和0.855(对应训练集AUC分别为0.940与0.937).结论:提出的方法能够较为准确地识别AD连续谱系的各个发展阶段,可为认知障碍的早期筛查提供参考.

Objective To propose an Alzheimer's disease(AD)classification method based on multi-region radiomics fingerprint features and machine learning so as to realize AD early assessment.Methods Firstly,300 subjects from the Alzheimer's Disease Neuroimaging Initiative(ADNI)public database had their clinical and imaging data selected retrospectively,including 100 patients with mild cognitive impairment(MCI),100 cases of cognitively normal(CN)and 100 AD patients;secondly,all the original T1WI structural images were preprocessed;thirdly,11 core brain regions highly correlated with the course of AD were selected and multi-dimensional radiomics features were extracted;fourthly,the core radiomics fingerprint features were screened out after eliminating redundant variables by using the least absolute shrinkage and selection operator(LASSO);finally,three binary classification models,AD/CN,AD/MCI and MCI/CN,were constructed based on the support vector machine(SVM).The ROC curves were used to evaluate the prediction efficacy of each model.Results The constructed SVM classification models demonstrated high diagnostic efficacy in all the three classification tasks,in which the AD/CN model had the best discriminative performance with an AUC of 0.991 in the training set and 0.925 in the test set;the MCI/CN model and the AD/MCI model had a slightly lower classification efficacy,with AUCs of 0.875 and 0.855 respectively in the test set and 0.940 and 0.937 respectively in the training set.Conclusion The method proposed identifies the development stages of AD continuum,and provides references for early screening of cognitive impairment.[Chinese Medical Equipment Journal,2026,47(9):10-18]

张宇辰;朱海雪

南京大学医学院附属泰康仙林鼓楼医院放射科,南京 210046南京大学医学院附属泰康仙林鼓楼医院放射科,南京 210046

医药卫生

阿尔茨海默病轻度认知障碍脑区指纹特征结构磁共振成像影像组学支持向量机

Alzheimer's diseasemild cognitive impairmentbrain regional fingerprint featurestructural magnetic reso-nance imagingradiomicssupport vector machine

《医疗卫生装备》 2026 (9)

10-18,9

10.19745/j.1003-8868.2026137

评论