2024
Subtype-Aware Dynamic Unsupervised Domain Adaptation
Liu X, Xing F, You J, Lu J, Kuo C, Fakhri G, Woo J. Subtype-Aware Dynamic Unsupervised Domain Adaptation. IEEE Transactions On Neural Networks And Learning Systems 2024, 35: 2820-2834. PMID: 35895653, DOI: 10.1109/tnnls.2022.3192315.Peer-Reviewed Original ResearchTarget domainSource domain to target domainUnsupervised domain adaptationWithin-class compactnessHeart disease dataPseudo-labelsDomain adaptationClass centersLatent spaceCluster centroidsConditional alignmentLabel shiftTransfer knowledgeQueueing frameworkLocal proximityAlternative processing schemesSubtype labelsExperimental resultsProcessing schemeSubtype structureDomainNetVisDADisease dataDomainLabelingDiffusion-based Bayesian posterior distribution prediction of kinetic parameters in dynamic PET
Djebra Y, Liu X, Marin T, Tiss A, Dhaynaut M, Guehl N, Johnson K, Fakhri G, Ma C, Ouyang J. Diffusion-based Bayesian posterior distribution prediction of kinetic parameters in dynamic PET. 2024, 00: 1-1. DOI: 10.1109/nss/mic/rtsd57108.2024.10657955.Peer-Reviewed Original ResearchConditional variational autoencoderEfficient deep learning-based approachMarkov chain Monte CarloDenoising diffusion probabilistic modelDeep learning-based approachDiffusion probabilistic modelLearning-based approachApproximate posterior distributionPosterior distributionVariational autoencoderHeavy computationTau protein aggregationBayesian inferenceProbabilistic modelData-drivenStudy molecular processesBayesian posterior distributionProtein aggregationMetropolis-Hastings Markov chain Monte CarloMolecular processesAlzheimer's diseaseNeurodegenerative diseasesKinetic parametersEstimate posterior distributionsAutoencoderPoint-supervised Brain Tumor Segmentation with Box-prompted Medical Segment Anything Model
Liu X, Woo J, Ma C, Ouyang J, Fakhri G. Point-supervised Brain Tumor Segmentation with Box-prompted Medical Segment Anything Model. 2011 IEEE Nuclear Science Symposium Conference Record 2024, 00: 1-1. PMID: 39445308, PMCID: PMC11497479, DOI: 10.1109/nss/mic/rtsd57108.2024.10656071.Peer-Reviewed Original ResearchAblation Study of Diffusion Model with Transformer Backbone for Low-count PET Denoising
Huang Y, Liu X, Miyazaki T, Omachi S, Fakhri G, Ouyang J. Ablation Study of Diffusion Model with Transformer Backbone for Low-count PET Denoising. 2011 IEEE Nuclear Science Symposium Conference Record 2024, 00: 1-2. PMID: 39445309, PMCID: PMC11497477, DOI: 10.1109/nss/mic/rtsd57108.2024.10655179.Peer-Reviewed Original ResearchIR tasksImage restorationImage super-resolution taskField of image restorationSuper-resolution taskLatent feature spaceConventional UNetDenoising iterationDenoising taskTransformer backboneDenoising autoencoderTexture restorationVision transformerFeature spaceAblation studiesLearning schemeBackbone networkImage generationDenoisingUNetIR modelPSNRSpatial informationAutoencoderTaskSubject-aware PET Denoising with Contrastive Adversarial Domain Generalization
Liu X, Marin T, Eslahi S, Tiss A, Chemli Y, Johson K, Fakhri G, Ouyang J. Subject-aware PET Denoising with Contrastive Adversarial Domain Generalization. 2011 IEEE Nuclear Science Symposium Conference Record 2024, 00: 1-1. PMID: 39445307, PMCID: PMC11497478, DOI: 10.1109/nss/mic/rtsd57108.2024.10656150.Peer-Reviewed Original ResearchDomain generalizationDenoising performanceDenoising moduleDeep learningSubject-independent mannerSubject-invariant featuresSuperior denoising performanceAdversarial learning frameworkSubject-related informationConventional UNetBottleneck featuresTrustworthy systemsLearning frameworkDL modelsDL model performanceDenoisingNoise realizationsNegative samplesList-mode dataImage volumesModel performancePerformancePerformance of positron emission tomographyUNetFraction of eventsInvestigating muscle coordination patterns with Granger causality analysis in protrusive motion from tagged and diffusion MRI
Park H, Xing F, Stone M, Kang H, Liu X, Zhuo J, Fels S, Reese T, Wedeen V, Fakhri G, Prince J, Woo J. Investigating muscle coordination patterns with Granger causality analysis in protrusive motion from tagged and diffusion MRI. JASA Express Letters 2024, 4: 095201. PMID: 39240196, PMCID: PMC11384280, DOI: 10.1121/10.0028500.Peer-Reviewed Original ResearchDiffusion Model-Based Posterior Distribution Prediction for Kinetic Parameter Estimation in Dynamic PET
Djebra Y, Liu X, Marin T, Tiss A, Dhaynaut M, Guehl N, Johnson K, Fakhri G, Ma C, Ouyang J. Diffusion Model-Based Posterior Distribution Prediction for Kinetic Parameter Estimation in Dynamic PET. 2011 IEEE International Symposium On Biomedical Imaging: From Nano To Macro 2024, 00: 1-5. PMID: 39530051, PMCID: PMC11554386, DOI: 10.1109/isbi56570.2024.10635805.Peer-Reviewed Original ResearchPosterior distributions of kinetic parametersDenoising diffusion probabilistic modelHyperphosphorylated tauP-tauDiffusion probabilistic modelAlzheimer's diseaseNeurodegenerative diseasesKinetic parametersPosterior distributionInference efficiencyComputational needsEstimate kinetic parametersProbabilistic modelComputation timeExploring Backdoor Attacks in Off-the-Shelf Unsupervised Domain Adaptation for Securing Cardiac MRI-Based Diagnosis
Liu X, Xing F, Gaggin H, Kuo C, El Fakhri G, Woo J. Exploring Backdoor Attacks in Off-the-Shelf Unsupervised Domain Adaptation for Securing Cardiac MRI-Based Diagnosis. 2011 IEEE International Symposium On Biomedical Imaging: From Nano To Macro 2024, 00: 1-5. PMID: 39421190, PMCID: PMC11483644, DOI: 10.1109/isbi56570.2024.10635403.Peer-Reviewed Original ResearchUnsupervised domain adaptationTarget domain modelBackdoor attacksDomain adaptationTraining dataLabeled source domain dataSusceptible to backdoor attacksAccurate pseudo labelsDomain modelSource domain dataPatient data privacyTarget training dataOff-the-shelfPseudo-labelsData privacySource domainMulti-vendorRandom initializationTraining phaseDomain dataDiagnosis modelTarget modelMulti-diseaseAttacksAuxiliary modelCross noise level PET denoising with continuous adversarial domain generalization
Liu X, Eslahi S, Marin T, Tiss A, Chemli Y, Huang Y, Johnson K, Fakhri G, Ouyang J. Cross noise level PET denoising with continuous adversarial domain generalization. Physics In Medicine And Biology 2024, 69: 085001. PMID: 38484401, PMCID: PMC11195012, DOI: 10.1088/1361-6560/ad341a.Peer-Reviewed Original ResearchDomain generalization techniqueDomain generalizationDenoising performanceSuperior denoising performanceLatent feature representationGeneral techniqueDistribution shiftsAdversarial trainingDenoised imageFeature representationDomain labelsDistribution divergenceNoise levelDeep learningImage spaceDenoisingPerformance degradationCore ideaNoise realizationsCD methodNoiseImage volumesPerformanceImagesPSNRThe role of 18F-FDG PET in minimizing variability in gross tumor volume delineation of soft tissue sarcomas
Najem E, Marin T, Zhuo Y, Lahoud R, Tian F, Beddok A, Rozenblum L, Xing F, Moteabbed M, Lim R, Liu X, Woo J, Lostetter S, Lamane A, Chen Y, Ma C, El Fakhri G. The role of 18F-FDG PET in minimizing variability in gross tumor volume delineation of soft tissue sarcomas. Radiotherapy And Oncology 2024, 194: 110186. PMID: 38412906, PMCID: PMC11042980, DOI: 10.1016/j.radonc.2024.110186.Peer-Reviewed Original ResearchGross tumor volume delineationGross tumor volumeDice similarity coefficientF-FDG PET imagingSoft tissue sarcomasInter-reader variabilityGTV delineationRadiation therapy treatment planningF-FDGF-FDG PETTherapy treatment planningPerformance level estimationTumor volume delineationTissue sarcomasPET imagingVolume delineationSimultaneous truthHausdorff distanceDice similarity coefficient scoreAccurate gross tumor volumeImaging modality groupsWilcoxon signed-rank testStatistically significant decreaseSigned-rank testTumor volumeSpeech motion anomaly detection via cross-modal translation of 4D motion fields from tagged MRI
Liu X, Xing F, Zhuo J, Stone M, Prince J, El Fakhri G, Woo J. Speech motion anomaly detection via cross-modal translation of 4D motion fields from tagged MRI. Proceedings Of SPIE--the International Society For Optical Engineering 2024, 12926: 129262w-129262w-5. PMID: 39238547, PMCID: PMC11377028, DOI: 10.1117/12.3006874.Peer-Reviewed Original ResearchCross-modal translationHealthy individualsTongue cancer patientsMotion fieldOut-of-distributionOne-class SVMPatient dataAnomaly detectionAnomaly detectorCancer patientsTagged MRISpeech-related disordersGeneralization capabilityReconstruction qualitySpeech qualityArticulatory-acoustic relationsPatientsSpeech waveformTraining setInnovative treatmentsMRITest setMotion patternsArticulatory featuresTraining translators
2023
Attentive continuous generative self-training for unsupervised domain adaptive medical image translation
Liu X, Prince J, Xing F, Zhuo J, Reese T, Stone M, El Fakhri G, Woo J. Attentive continuous generative self-training for unsupervised domain adaptive medical image translation. Medical Image Analysis 2023, 88: 102851. PMID: 37329854, PMCID: PMC10527936, DOI: 10.1016/j.media.2023.102851.Peer-Reviewed Original ResearchConceptsUnsupervised domain adaptationImage translationProblem of domain shiftSelf-trainingImage modality translationLabeled source domainTarget domain dataSelf-attention schemeAlternating optimization schemeHeterogeneous target domainContinuous value predictionPseudo-labelsDomain adaptationUDA methodsDomain shiftSoftmax probabilitiesSource domainTarget domainVariational BayesBackground regionsTranslation tasksTraining processDomain dataGeneration taskOptimization schemeIncremental Learning for Heterogeneous Structure Segmentation in Brain Tumor MRI
Liu X, Shih H, Xing F, Santarnecchi E, El Fakhri G, Woo J. Incremental Learning for Heterogeneous Structure Segmentation in Brain Tumor MRI. Lecture Notes In Computer Science 2023, 14221: 46-56. PMID: 38665992, PMCID: PMC11045038, DOI: 10.1007/978-3-031-43895-0_5.Peer-Reviewed Original ResearchDeep learningDL modelsBrain tumor segmentation taskAbsence of training dataIncremental learning settingSegmenting various anatomical structuresBig medical dataInitial model trainingTumor segmentation taskBatch renormalizationCatastrophic forgettingIncremental learningSegmentation taskSource domainTraining dataModel trainingLearning structureSegmentation modelNetwork optimizationDiverse datasetsMedical dataEvolving environmentLearning settingsDistribution shiftsIncremental structureSpeech Audio Synthesis from Tagged MRI and Non-negative Matrix Factorization via Plastic Transformer
Liu X, Xing F, Stone M, Zhuo J, Fels S, Prince J, El Fakhri G, Woo J. Speech Audio Synthesis from Tagged MRI and Non-negative Matrix Factorization via Plastic Transformer. Lecture Notes In Computer Science 2023, 14226: 435-445. PMID: 38651032, PMCID: PMC11034915, DOI: 10.1007/978-3-031-43990-2_41.Peer-Reviewed Original ResearchWeight mapAudio waveformEnd-to-end deep learning frameworkMatrix factorization-based approachesFactorization-based approachDeep learning frameworkNon-negative matrix factorizationEnd-to-endAdversarial trainingProcess of speech productionTwo-dimensional spectrogramConventional convolutionLearning frameworkMotion featuresTraining samplesAudio synthesisDimension expansionMatrix inputMatrix factorizationTagged MRISpeech productionTransformation modelExperimental resultsSpectrogramPlastic transformationSuccessive Subspace Learning for Cardiac Disease Classification with Two-Phase Deformation Fields from Cine MRI
Liu X, Xing F, Gaggin H, Kuo C, El Fakhri G, Woo J. Successive Subspace Learning for Cardiac Disease Classification with Two-Phase Deformation Fields from Cine MRI. 2011 IEEE International Symposium On Biomedical Imaging: From Nano To Macro 2023, 00: 1-5. PMID: 38031559, PMCID: PMC10686280, DOI: 10.1109/isbi53787.2023.10230746.Peer-Reviewed Original ResearchTraining samplesCardiovascular disease classificationCNN-based approachDeep learning modelsCardiac disease classificationSubspace learningSSL modelClassification performanceDeep learningCardiac cine magnetic resonance imagingSubspace approximationSupervised regressionLearning modelsAccurate characterization resultsDisease classificationClassificationCardiac atlasLearningDeformation fieldEnd-systolic phaseFrameworkFeedforward designPerformanceTrainingSSLNoise-Robust Sleep Staging via Adversarial Training With an Auxiliary Model
Yoo C, Liu X, Xing F, Fakhri G, Woo J, Kang J. Noise-Robust Sleep Staging via Adversarial Training With an Auxiliary Model. IEEE Transactions On Biomedical Engineering 2023, 70: 1252-1263. PMID: 36227815, DOI: 10.1109/tbme.2022.3214269.Peer-Reviewed Original ResearchConceptsProblem of domain shiftTarget networkConsumer-level devicesPre-trained modelsSleep staging performanceSevere noise levelsEfficient training approachAdversarial trainingAdversarial noiseDomain shiftAdversarial transformationsAuxiliary modelDL modelsInput perturbationsTraining modelArbitrary noiseTesting stageEnhanced robustnessNoise signalsTraining approachTest environmentExperimental resultsNoiseSleep stagesRobustness
2022
Memory consistent unsupervised off-the-shelf model adaptation for source-relaxed medical image segmentation
Liu X, Xing F, El Fakhri G, Woo J. Memory consistent unsupervised off-the-shelf model adaptation for source-relaxed medical image segmentation. Medical Image Analysis 2022, 83: 102641. PMID: 36265264, PMCID: PMC10016738, DOI: 10.1016/j.media.2022.102641.Peer-Reviewed Original ResearchConceptsUnsupervised domain adaptationUnsupervised domain adaptation methodsSource domain dataBN statisticsTarget domainLabeled source domain dataDomain dataLabeled source domainSelf-training strategyPatient data privacyHeterogeneous target domainBrain tumor segmentationPseudo-labelsDomain adaptationUnsupervised adaptationData privacySegmentation taskSource domainImage segmentationVital protocolAdaptation frameworkDecay strategyBoost performanceModel adaptationTumor segmentation
2021
Adversarial Unsupervised Domain Adaptation with Conditional and Label Shift: Infer, Align and Iterate
Liu X, Guo Z, Li S, Xing F, You J, Kuo C, Fakhri G, Woo J. Adversarial Unsupervised Domain Adaptation with Conditional and Label Shift: Infer, Align and Iterate. 2021, 00: 10347-10356. DOI: 10.1109/iccv48922.2021.01020.Peer-Reviewed Original ResearchUnsupervised domain adaptationDomain adaptationLabel shiftUnsupervised domain adaptation methodsAdversarial unsupervised domain adaptationAlternating optimization schemeUDA methodsTarget domainTraining stageOptimization schemeTesting stageExperimental resultsDistribution w.AdversaryP(x|yP(y|xDomainSchemeClassificationMethodInferenceAdaptationDomain Generalization under Conditional and Label Shifts via Variational Bayesian Inference
Liu X, Hu B, Jin L, Han X, Xing F, Ouyang J, Lu J, El Fakhri G, Woo J. Domain Generalization under Conditional and Label Shifts via Variational Bayesian Inference. 2021, 881-887. DOI: 10.24963/ijcai.2021/122.Peer-Reviewed Original ResearchDomain generalizationDomain-invariant feature learningVariational Bayesian inference frameworkLabel shiftCross-domain accuracyLabeled source domainVariational Bayesian inferenceFeature learningBayesian inference frameworkLatent spaceSource domainTarget domainDistribution matchingTransfer knowledgeInference frameworkSuperior performanceP(x|yBayesian inferenceLabelingDomainFrameworkGeneralizationBenchmarksPosterior alignmentLearning
2020
Severity-Aware Semantic Segmentation with Reinforced Wasserstein Training
Liu X, Ji W, You J, Fakhri G, Woo J. Severity-Aware Semantic Segmentation with Reinforced Wasserstein Training. 2020, 00: 12563-12572. DOI: 10.1109/cvpr42600.2020.01258.Peer-Reviewed Original ResearchSemantic segmentationCARLA simulatorCross-entropyGround distance matrixWasserstein training frameworkAlternating optimization schemeCityscapes datasetDNN architecturesCE lossTraining frameworkSemantic classesGround metricInter-class correlationAutonomous vehiclesSuperior performanceOptimization schemeDNNCARLADistance matrixSurgery systemCamVidDeepLabSimulationCityscapesPixel