基于PET图像的深度学习模型预测阿尔茨海默病不同阶段的研究  

Prediction of Different Stages of Alzheimer's Disease by Deep Learning Model Based on PET Image

作　　者：严彩英 冷一林 刘晨鹭[1] 张帆 郑健[3] 陈双庆[1] YAN Caiying;LENG Yilin;LIU Chenlu(Department of Radiology,the Affiliated Suzhou Hospital of Nanjing Medical University,Suzhou,Jiangsu Province 215001,P.R.China)

机构地区：[1]南京医科大学附属苏州医院放射科,215001 [2]上海大学通信与信息工程学院,200444 [3]中国科学院苏州生物医学工程技术研究所,215163

出　　处：《临床放射学杂志》2022年第9期1615-1620,共6页Journal of Clinical Radiology

基　　金：苏州市卫生计生委科技项目(编号:LCZX201909);江苏省卫健委重点项目(编号:ZDB2020011)资助。

摘　　要：目的探讨基于^(18)F-FDG PET图像构建的深度学习模型对不同阶段阿尔茨海默病(AD)的诊断价值。方法根据病程进展分为正常认知(CN)组、主观认知下降(SCD)组、轻度认知障碍(MCI)组和AD组。回顾性搜集ADNI数据库中519例(SCD组94例,AD组125例,MCI及CN组各150例)符合纳入标准病例的^(18)F-FDG PET图像,在此基础上加入临床资料(年龄、性别、受教育程度、MMSE评分、Aβ1-42值及载脂蛋白Eε4表型)。基于两种数据集建立AD分类深度学习双线性池化模型,经五折交叉验证后,采用分类准确性、AUC值等评价两种模型分类性能。结果基于^(18)F-FDG PET图像单模态分类模型,在区分SCD-AD组分类性能最佳,准确性达91.97%,在CN-SCD组分类效果相对较低,准确性为74.86%;融合临床资料后,仅CN-SCD组分类准确性下降了1.53%,其余各组的分类性能均得到提高,其中准确性升幅依次降序排列为CN-MCI组(升至90.33%,升幅6.83%)、SCD-AD组(升至94.34%,升幅2.37%)、SCD-MCI组(升至82.78%,升幅2.30%)、MCI-AD组(升至80.45%,升幅2.26%)、CN-AD组(升至92.13%,升幅1.88%)。结论联合^(18)F-FDG PET图像及临床资料构建的深度学习融合模型可为AD临床早期诊断及干预提供帮助。Objective To explore the diagnostic value of deep learning model based on ^(18)F-FDG PET image in different stages of Alzheimer’s disease(AD).Methods According to the course of disease,they were divided into cognitive normal(CN)group,subjective cognitive decline(SCD)group,mild cognitive impairment(MCI)group and AD group.^(18)F-FDG PET images of 519 cases(94 cases in SCD group,125 cases in AD group,150 cases in MCI and CN group)in ADNI database were collected retrospectively.On this basis,clinical data,including age,gender,years of education,MMSE score,Aβ142 and Apolipoprotein Eε4 were added.Based on the two data sets,the AD classification deep learning bilinear pooling model was established.After 5 fold cross validation experiments,the classification accuracy and AUC value were used to evaluate the classification performance of the two models.Results Based on the single mode classification model of^(18)F-FDG PET images,the classification performance was the best in distinguishing SCD and AD group,with an accuracy of 91.97%,and the classification effect was relatively low in distinguishing in CN and SCD group,with an accuracy of 74.86%.After integrating the clinical data,only the classification performance of CN/SCD group decreased by 1.53%,and the classification performance of other groups was improved.The increase order of accuracy was CN MCI group(increased to 90.33%,increased by 6.83%),SCD AD group(increased to 94.34%,increased by 2.37%),SCD MCI group(increased to 82.78%,increased by 2.30%),MCI AD group(increased to 80.45%,increased by 2.26%)and CN AD group(increased to 92.13%,increased by 1.88%).Conclusion The deep learning fusion model combined^(18)F-FDG PET images with clinical data can provide help for the early clinical diagnosis and intervention of AD.

关 键 词：^(18)F-FDG PET AD不同阶段 深度学习 

分 类 号：R749.16[医药卫生—神经病学与精神病学]