2022
ACT: Semi-supervised Domain-Adaptive Medical Image Segmentation with Asymmetric Co-training
Liu X, Xing F, Shusharina N, Lim R, Jay Kuo C, El Fakhri G, Woo J. ACT: Semi-supervised Domain-Adaptive Medical Image Segmentation with Asymmetric Co-training. Lecture Notes In Computer Science 2022, 13435: 66-76. PMID: 36780245, PMCID: PMC9911133, DOI: 10.1007/978-3-031-16443-9_7.Peer-Reviewed Original ResearchSemi-supervised domain adaptationUnsupervised domain adaptationSemi-supervised learningMedical image segmentationDomain adaptationDomain shiftLabel supervisionTarget domainImage segmentationDomain dataLeverage different knowledgePseudo-label noiseSignificant domain shiftSupervised joint trainingLabeled source domainUnlabeled target dataUnlabeled target domainLabeled target samplesTarget domain dataSource domain dataState-of-the-artMRI segmentation taskSubstantial performance gainsPseudo-labelsLabel noiseUnsupervised Black-Box Model Domain Adaptation for Brain Tumor Segmentation
Liu X, Yoo C, Xing F, Kuo C, Fakhri G, Kang J, Woo J. Unsupervised Black-Box Model Domain Adaptation for Brain Tumor Segmentation. Frontiers In Neuroscience 2022, 16: 837646. PMID: 35720708, PMCID: PMC9201342, DOI: 10.3389/fnins.2022.837646.Peer-Reviewed Original ResearchUnsupervised domain adaptationDomain adaptationSource domainTarget domainLabeled source domain to unlabeled target domainTransfer of domain knowledgeTarget-specific representationsUnlabeled target domainTarget domain dataKnowledge distillation schemeDeep learning backbonesEntropy minimizationTrained model parametersDifficulty of labelingDomain knowledgeSensitive informationPrivacy concernsPerformance gainsNetwork parametersSegmentation modelDomain dataSource dataCross-center collaborationDistillation schemePotential leaksUnsupervised domain adaptation for segmentation with black-box source model
Liu X, Yoo C, Xing F, Kuo C, El Fakhri G, Kang J, Woo J. Unsupervised domain adaptation for segmentation with black-box source model. Proceedings Of SPIE--the International Society For Optical Engineering 2022, 12032: 1203210-1203210-6. PMID: 35983176, PMCID: PMC9385170, DOI: 10.1117/12.2607895.Peer-Reviewed Original ResearchUnsupervised domain adaptationSource domainDomain adaptationTarget-specific representationsLabeled source domainUnlabeled target domainTarget domain dataWell-labeled dataKnowledge distillation schemeTrained model parametersModel adaptation approachOriginal source dataDifficulty of labelingTarget domainSegmentation modelDomain dataTransfer knowledgeEntropy minimizationAdaptive approachSource dataConventional solutionsPractical solutionDistillation schemePrivacyLarge-scale
2021
Generative Self-training for Cross-Domain Unsupervised Tagged-to-Cine MRI Synthesis
Liu X, Xing F, Stone M, Zhuo J, Reese T, Prince J, El Fakhri G, Woo J. Generative Self-training for Cross-Domain Unsupervised Tagged-to-Cine MRI Synthesis. Lecture Notes In Computer Science 2021, 12903: 138-148. PMID: 34734217, PMCID: PMC8562649, DOI: 10.1007/978-3-030-87199-4_13.Peer-Reviewed Original ResearchUnsupervised domain adaptationTarget domainUDA methodsImage synthesisProblem of domain shiftUnsupervised domain adaptation frameworkSelf-trainingTraining deep learning modelsVariational Bayes learningUnlabeled target domainAlternating optimization schemePseudo-label selectionDeep learning modelsContinuous value predictionPseudo-labelsDomain adaptationDomain shiftCross-domainSynthesis qualityBayes learningDiscrete histogramsPrediction confidenceLearning modelsGeneration taskOptimization scheme