•计算机工程与应用•MRI图像降噪技术综述蒲晓蓉1*，陈佳俊2，高励3，赵越1，罗纪翔1，刘军池1，任亚洲1(1.电子科技大学计算机科学与工程学院成都611731；2.电子科技大学格拉斯哥学院成都611731；3.成都市第三人民医院神经内科成都610031)【摘要】核磁共振成像技术已广泛用于脑部、脊髓和心脏等相关疾病的临床诊断。然而，由采样时间、环境、设备质量等多种因素导致的成像噪声制约着诊断精度的进一步提高。综合研究了MRI降噪技术的发展脉络，系统梳理了基于滤波、变换、统计等传统MRI图像降噪方法，并重点分析了当前基于深度学习的MRI图像降噪系列新方法，展望了MRI图像降噪的未来发展趋势。最后，总结了现有医学图像质量评估方法，并指出针对依赖大量数据和人工标注医学图像样本、而可解释性较差的现有深度学习方法，需要探索性研究面向临床实际任务的医学图像质量评估新方法。关键词深度学习；核磁共振；医学图像质量评估；MRI图像降噪中图分类号TN415文献标志码Adoi:10.12178/1001-0548.2022248SurveyonMagneticResonanceImageDenoisingPUXiaorong1*,CHENJiajun2,GAOLi3,ZHAOYue1,LUOJixiang1,LIUJunchi1,andRENYazhou1(1.SchoolofComputerScienceandEngineering,UniversityofElectronicScienceandTechnologyofChinaChengdu611731;2.GlasgowCollege,UniversityofElectronicScienceandTechnologyofChinaChengdu611731;3.DepartmentofNeurology,TheThirdPeople'sHospitalofChengduChengdu610031)AbstractMagneticResonanceImaging(MRI)hasbeenextensivelyemployedasanauxiliarymeanstodiagnosepathologicaldeteriorationofbrain,spinalcordandheartrelateddiseasesclinically.Nevertheless,imagingnoiseinducedbybothinternalandexternalimpactsrestrictfurtherimprovementondiagnosticaccuracy.Thispapercarriesoutareviewontechnologicalinnovationsrangingfromearlierconventionalapproachesbasedonfiltertechniqueostate-of-the-artalternativesutilizingthedeeplearningnetwork.Finally,someinductivesummariesofthemedicalimagequalityassessmentshavebeenintroduced.Italsopointsoutthatexistingdeeplearningmethods,whichrelyonalargeamountofdataandmanualannotationofmedicalimagesamples,arepoorlyinterpretable.Itisvitalthatclinical-orientedevaluationmechanismshouldbeexploredforclinicaldemands.Keywordsdeeplearning;magneticr...
