2023年第37卷第4期测试技术学报Vol.37No.42023(总第160期)JOURNALOFTESTANDMEASUREMENTTECHNOLOGY(SumNo.160)文章编号:1671-7449(2023)04-0348-08①基于迁移学习的多模态AD病程分类研究乔悦,李瑞红(中北大学软件学院,山西太原030051)摘要:近年来,患阿尔茨海默病(Alzheimer’sDisease,AD)的人数逐年增加。临床研究显示,轻度认知障碍(MildCognitiveImpairment,MCI)转化为AD的概率很大,因此,提高磁共振成像(MagneticResonanceImaging,MRI)和正电子发射断层扫描(PositronEmissionTomography,PET)等神经影像图对AD、MCI的分类准确率十分必要。为了解决数据量少、标注困难的问题,首先使用CycleGAN网络对缺少的PET图进行生成;然后采用基于区域能量融合准则的小波变换算法对MRI图和PET图进行图像融合,能够极大程度的保留图像中的数据信息;最后利用迁移学习技术对轻量级网络MobileNet进行训练与微调。实验结果显示,在数据量较少的情况下,所提方法在泛化能力较好的同时,也获得了较高的准确率。关键词:神经影像图;图像生成;图像融合;迁移学习;AD病程分类中图分类号:TP391.4文献标识码:Adoi:10.3969/j.issn.1671-7449.2023.04.012ResearchonMultimodalAlzheimerDiseaseCourseClassificationBasedonTransferLearningQIAOYue,LIRuihong(SchoolofSoftware,NorthUniversityofChina,Taiyuan030051,China)Abstract:Inrecentyears,thenumberofpeoplesufferingfromAlzheimer’sDisease(AD)hasin-creasedyearbyyear.Clinicalresearchshowsthattheprobabilityofmildcognitiveimpairment(MCI)transformingintoADisveryhigh,soitisnecessarytoimprovetheclassificationaccuracyofADandMCIbymagneticresonanceimaging(MRI),positronemissiontomography(PET)andotherneuroim-agingimages.Inordertosolvetheproblemoffewerdataanddifficultyinlabeling,thispaperfirstusestheCycleGANnetworktogeneratethemissingPETmap;Secondly,thewavelettransformalgorithmbasedontheregionalenergyfusioncriterionisusedtofuseMRIimagesandPETimages,whichcanre-tainthedatainformationintheimagestoagreatextent;Finally,usethetransferlearningtechnologytotrainandfine-tunethelightweightnetworkMobileNet.Theexperimentalresultsshowthat,inthecaseoffewerdata,themethodusedinthispaperhasbettergeneralizationabilityandhigheraccuracy.Keywords:neuroimagingimage;imagegeneration...
