Vol.40No.2第40卷第2期2023年06月ChineseJMagnReson,2023,40(2):220-238波谱学杂志ChineseJournalofMagneticResonanceJun.2023doi:10.11938/cjmr20223013基于深度学习的阿尔兹海默症影像学分类研究进展钱程一,王远军*上海理工大学医学影像技术研究所,上海200093摘要:随着全球老龄化的加剧与深度学习的发展,基于深度学习的阿尔兹海默症(AD)影像学分类成为当前的一个研究热点.本文首先阐述了AD影像学分类任务中常用的深度学习模型、评估标准及公开数据集;接着讨论了不同图像模态在AD影像学分类中的应用;然后着重探讨了应用于AD影像学分类的深度学习模型改进方法;进一步引入了对模型可解释性研究的探讨;最后总结并比较了文中提及的分类模型,归纳了与AD影像分类相关的大脑区域,并对该领域未来的研究方向进行了展望.关键词:阿尔茨海默症(AD);深度学习;医学影像;分类;可解释性中图分类号:TP391文献标识码:AResearchProgressonImagingClassificationofAlzheimer'sDiseaseBasedonDeepLearningQIANChengyi,WANGYuanjun*InstituteofMedicalImagingTechnology,UniversityofShanghaiforScienceandTechnology,Shanghai200093,ChinaAbstract:Asglobalagingworsensanddeeplearningadvances,theimagingclassificationofAlzheimer'sdisease(AD)basedondeeplearninghasbecomeahottopicofresearch.Thispaperreviewedthecommondeeplearningmodels,evaluationcriteriaandpublicdatasetsinADimagingclassificationtasks,discussedtheapplicationofdifferentimagemodalitiesinADimagingclassification.ThecontentwasfocusedontheimprovementofdeeplearningmodelsappliedtoADimagingclassification.Thestudiesofmodelinterpretabilitywerealsointroduced.Finally,thepapersummarizedandcomparedtheclassificationmodelsmentioned,identifiedthebrainregionsrelatedtoADimageclassification,andoutlinedthefutureresearchdirectionsinthisfield.Keywords:Alzheimer'sdisease(AD),deeplearning,medicalimaging,classification,interpretability收稿日期:2022-08-12;在线发表日期:2022-11-07基金项目:上海市自然科学基金资助项目(18ZR1426900).通信作者(Correspondingauthor):*Tel:13761603606,E-mail:yjusst@126.com.第2期引言阿尔兹海默症(Alzheimerdisease,AD)是痴呆症最常见的类型,占所有痴呆症的50~70%,其症状为认知、功能和行为的退化,...