基于融合学习的低剂量CT图像重建算法研究

基于融合学习的低剂量CT图像重建算法研究基于融合学习的低剂量CT图像重建算法研究

摘要：

由于低剂量CT对于肺癌筛查具有广阔应用前景，近年来为了减少低剂量脑CT造成的辐射剂量，提高重建质量，发展了许多低剂量CT图像重建算法。然而，这些方法仍面临诸多挑战，如结构体积效应、深度学习模型泛化能力难以掌控等。为解决这些问题，本文提出了一种基于融合学习的低剂量CT图像重建算法。首先，利用瑞利-拉斯福德反演算法进行初步重建，获得初步重建图像；然后，将初步重建图像连接入一个深度融合模型中，从而结合了低剂量和高剂量图像信息进行重建；最后，通过对比实验和主观评估，证明了该算法的有效性和可行性。

关键词：低剂量CT；图像重建；融合学习；深度网络

Abstract:

Duetothebroadapplicationprospectsoflow-doseCTinlungcancerscreening,inrecentyears,manylow-doseCTimagereconstructionalgorithmshavebeendevelopedtoreducetheradiationdosecausedbylow-dosebrainCTandimprovethereconstructionquality.However,thesemethodsstillfacemanychallenges,suchasthestructurevolumeeffectandthedifficultyincontrollingthegeneralizationabilityofdeeplearningmodels.Tosolvetheseproblems,thispaperproposesalow-doseCTimagereconstructionalgorithmbasedonfusionlearning.First,thepreliminaryreconstructioniscarriedoutbyusingtheRayleigh-Lassondeinversionalgorithmtoobtainthepreliminaryreconstructionimage.Then,thepreliminaryreconstructionimageisconnectedtoadeepfusionmodel,whichcombineslow-doseandhigh-doseimageinformationforreconstruction.Finally,throughcomparativeexperimentsandsubjectiveevaluation,theeffectivenessandfeasibilityoftheproposedalgorithmareproved.

Keywords:low-doseCT;imagereconstruction;fusionlearning;deepnetworInrecentyears,withthedevelopmentofmedicaltechnology,theadvancementoflow-doseCThasbecomeanimportantresearchtopic.ThetraditionalCTscanmethodrequireshighradiationdose,whichcausessevereharmtopatients.ByreducingtheradiationdoseduringCTscans,low-doseCThasgraduallyattractedattentionasitcaneffectivelyreducetheharmtopatientswhileobtainingaccuratemedicalinformation.

However,conductingimagereconstructionbasedonlow-doseCTpresentschallengesasthelowradiationdoseresultsinnoiseandimageblurinthereconstructedimages.Toaddressthischallenge,adeeplearning-basedframeworkisproposedinthisstudy.

TheproposedmethodcombinestheadvantagesofRayleigh-Lassondeinversionalgorithmanddeepfusionlearning.TheRayleigh-Lassondeinversionalgorithmisusedtoobtainthepreliminaryreconstructionimage,whichprovidestheinitialimagefordeepfusionlearning.Then,thepreliminaryreconstructionimageandthehigh-doseimagearecombinedthroughdeepfusionlearningtoreconstructhigh-qualityimages.

Toevaluatetheeffectivenessandfeasibilityoftheproposedalgorithm,experimentswereconductedonclinicaldata.Theexperimentalresultsdemonstratedthattheproposedalgorithmsignificantlyoutperformedotherstate-of-the-artreconstructionalgorithmsintermsofimagequalityandnoisesuppression.

Inconclusion,theproposedmethodisaneffectiveandfeasibleapproachforlow-doseCTimagereconstruction.ItprovidesapromisingsolutionforreducingtheradiationdoseduringCTscansandproducinghigh-qualitymedicalimageswithreducednoiseandimageblurFutureworkcanfocusonexploringtheeffectivenessoftheproposedalgorithmonlargerdatasetsandclinicaltrials,aswellasoptimizingthehyperparameterstomaximizeitsbenefits.Additionally,itwouldbeinterestingtoinvestigatetheperformanceofthealgorithmonvarioustypesofCTscans,includingthoseofdifferentbodypartsandorgans,andtocompareitwithotherreconstructionalgorithmsinthosescenarios.

Moreover,theproposedalgorithmcanalsobeappliedtoothermedicalimagingmodalities,suchasmagneticresonanceimaging(MRI)andpositronemissiontomography(PET),inordertoreduceradiationexposureandimproveimagequality.Futureresearchcanexplorethepotentialofthealgorithmintheseareas.

Intermsofimplementation,theproposedmethodcanbeintegratedintomodernCTmachinestoenablereal-timelow-doseCTimagingforclinicaluse.ThiswouldgreatlyimprovethesafetyofCTscansandprovidephysicianswithhigh-qualityimagesfordiagnosisandtreatmentplanning.

Lastly,theproposedalgorithmhassignificantpotentialforimprovingradiationtherapyplanningandmonitoring.Byprovidinghigh-qualityCTimageswithreducednoiseandimageblur,thealgorithmcanenhancetheaccuracyoftreatmentplanningandallowformoreprecisetargetingoftumors.Furthermore,byreducingtheradiationdose,thealgorithmcanhelpminimizethesideeffectsofradiationtherapyonhealthytissues.

Insummary,theproposedlow-doseCTimagereconstructionalgorithmisapromisingsolutionforreducingradiationexposureduringCTscanswhilemaintaininghigh-qualitymedicalimages.Furtherresearchcanexploreitsefficacyonlargerdatasetsandinclinicalsettings,aswellasitspotentialapplicationsinothermedicalimagingmodalitiesandradiationtherapyplanningOnepotentialapplicationofthelow-doseCTimagereconstructionalgorithmisinlungcancerscreeningprograms.Thecurrentstandardscreeningmethodislow-doseCT,butconcernsaboutradiationexposurelimititswidespreadimplementation.Withtheuseoftheproposedalgorithm,theradiationdosecouldbefurtherreducedwithoutcompromisingtheaccuracyoftheimages,makinglungcancerscreeningasaferandmoreaccessibleoptionforat-riskindividuals.

Anotherareawherethisalgorithmcouldbeusefulisinradiationtherapyplanning.Medicalimagesarecriticalfortreatmentplanning,buttheradiationdosereceivedduringCTscanscancausesideeffectsandcomplicationsthatreducetheefficacyofthetreatment.Byusingthelow-doseCTimagereconstructionalgorithm,radiationexposurecanbeminimized,allowingformorepreciseandeffectiveradiationtherapy.

Moreover,thealgorithmcouldbeappliedinothermedicalimagingmodalitieslikePETandSPECTscans,whereradiationisalsoaconcern.Byreducingtheradiationdoseinthesescans,morepatientscanbenefitfromtheseimagingtechniques,leadingtomoreaccuratediagnosesandbettertreatmentoutcomes.

Inclinicalsettings,theproposedalgorithmmayalsohelptoreducecostsassociatedwithCTscans,suchasequipmentmaintenance,staffsalaries,andpatientbilling.Byreducingtheradiationdose,fewerimagesmaybeneeded,loweringthenumberofscansperpatient,andultimatelydecreasinghealthcarecost.

Inconclusion,thelow-doseCTimagereconstructionalgorithmisapromisingdevelopmentinmedicalimagingtechnologythathasthepotentialtoimprovepatientsafety,reduceradiationexposure,andenhanceoverallclini