Desktop version

Home arrow Health

  • Increase font
  • Decrease font


>>

Auto-Segmentation for Radiation Oncology: State of the Art


IntroductionEvolution of Auto-SegmentationEvaluation of Auto-SegmentationBenchmark DatasetClinical Implementation ConcernsReferencesI Multi-Atlas for Auto-SegmentationIntroduction to Multi-Atlas Auto-SegmentationIntroductionDatabase ConstructionAtlas SelectionQuery Image RegistrationLabel FusionLabel Post-ProcessingSummary of This Part of the BookReferencesEvaluation of Atlas Selection: How Close Are We to Optimal Selection?Motivation for Atlas SelectionMethods of Atlas SelectionEvaluation of Image-Based Atlas SelectionImplementationBrute-Force SearchAtlas Selection Performance AssessmentDiscussion and Implications for Atlas SelectionLimitationsImpact of Atlas Selection on Clinical PracticeSummary and Recommendations for Future ResearchReferencesDeformable Registration Choices for Multi-Atlas SegmentationIntroductionDeformable RegistrationB-Spline RegistrationDemons AlgorithmPlastimatch MABS Implementation DetailsEvaluation MetricsExperimental stepsResultsSummaryReferencesEvaluation of a Multi-Atlas Segmentation SystemIntroductionMethodsPatient DataOnline Atlas Selection for Multi-Atlas SegmentationFirst Phase of Atlas SelectionDeformable Image RegistrationSecond Phase of Atlas SelectionContour FusionEvaluation MetricsEsophagus Segmentation for Head and Neck Cancer PatientsValidation Using Public Benchmark DatasetResultsAtlas Ranking and SelectionEsophagus Segmentation for Head and Neck Cancer PatientsValidation with Public Benchmark DatasetDiscussionConclusionsReferencesII Deep Learning for Auto-SegmentationIntroduction to Deep Learning-Based Auto-Contouring for RadiotherapyIntroductionHistorical ContextArtificial Neural NetworksConvolution Neural NetworksComputational PowerWhat Makes Deep Learning-Based Contouring So Different to Atlas-Based or Model-Based ApproachesUnderlying AssumptionsUse of DataDegrees of FreedomSummary of This Part of the BookReferencesDeep Learning Architecture Design for Multi-Organ SegmentationIntroductionDeep Learning Architecture in Medical Image Multi-Organ SegmentationAuto-Encoder MethodsAuto-Encoder and Its VariantsOverview of WorksDiscussionCNN MethodsNetwork DesignsOverview of WorksDiscussionFCN MethodsNetwork DesignsOverview of WorksDiscussionGAN MethodsNetwork DesignsOverview of WorksDiscussionR-CNN MethodsNetwork DesignsOverview of WorksDiscussionHybrid MethodsNetwork DesignsOverview of WorksDiscussionBenchmarkConclusionAcknowledgmentsReferencesComparison of 2D and 3D U-Nets for Organ SegmentationIntroductionStructures of 2D and 3D U-NetsD U-NetD U-NetExperimental ResultsDatasetsEvaluation MetricsImplementation DetailsResultsDiscussionsSummaryReferencesOrgan-Specific Segmentation Versus Multi-Class Segmentation Using U-NetIntroductionMaterials and MethodsDatasetsNetwork StructurePre-Processing and DownsamplingQuantitative Evaluation MetricsImplementation and Comparison ExperimentsResultsDiscussionConclusionsAcknowledgmentsReferencesEffect of Loss Functions in Deep Learning-Based SegmentationIntroductionAdmissibility of a Loss FunctionPresenting the ProblemCommon Loss FunctionsMean Squared ErrorCross EntropyBinary Cross EntropyCategorical Cross EntropyDice LossHausdorff Distance LossDealing with Class ImbalanceWeighted Cross EntropyGeneralized Dice LossNo-Background Dice LossFocal LossSensitivity Specificity LossTversky LossCompound Loss FunctionsDice + Cross EntropyDice + Focal LossNon-Linear CombinationsDealing with Imperfect DataEvaluating a Loss FunctionReferencesData Augmentation for Training Deep Neural NetworksOverviewIntroduction and Literature ReviewGeometric TransformationsIntensity TransformationArtificial Data GenerationApplications of Data AugmentationDatasets and Image PreprocessingTraining, Validation, and Testing for Organ SegmentationEvaluation CriteriaResultsDiscussionSummaryAcknowledgmentsReferencesIdentifying Possible Scenarios Where a Deep Learning Auto-Segmentation Model Could FailBackgroundSite-Specific ModelsLimitations of Training DataDeep Learning ArchitectureTwo-Stage U-Net ModelImage PreprocessingStage 1: Localization through Coarse SegmentationsStage 2: OAR Segmentation through Fine-Detail SegmentationTest-Time Cluster Cropping TechniqueQuantitative and Qualitative Review of Auto-SegmentationsPerformance on Challenge Test SetDifferent Anatomical SitesHead and Neck ScansAbdominal ScansBreast Cancer Simulation ScansDifferent Clinical PresentationsAtelectasis and Pleural EffusionPresence of Motion Management DevicesUse of Contrast and the Presence of Implanted DevicesAdapting to the UnseenDiscussion and ConclusionsAcknowledgmentsReferencesIII Clinical Implementation ConcernsClinical Commissioning GuidelinesIntroductionStages in Clinical CommissioningNeed for Robust and Clinically Useful MetricsNeed for Curated Datasets for Clinical CommissioningAuto-Segmentation Clinical Validation Studies – Current State-of-the-ArtData Curation Guidelines for Radiation OncologyEvaluation Metrics Guidelines for Clinical CommissioningCommissioning and Safe Use in the ClinicTraining and Validation PhaseTesting and Verification PhaseSummaryReferencesData Curation Challenges for Artificial IntelligenceIntroductionThe Complexity of Medical ImagingThe Challenge of Generalizability and Data HeterogeneityData SelectionThe Need for Large Quantities of DataBarriers to Sharing Patient Data and Distributed Deep LearningData Quality IssuesData AnnotationsData Curation via CompetitionsBias and Curation of Fair DataOverview of Data Curation ProcessConclusionReferencesOn the Evaluation of Auto-Contouring in RadiotherapyIntroductionQuantitative EvaluationStrengths and LimitationsImplementationClassification AccuracyImplementationAdvantages and LimitationsDice Similarity CoefficientImplementationAdvantages and LimitationsDistance MeasuresHausdorff Distance% Hausdorff DistanceAverage DistanceImplementationAdvantages and LimitationsGeometric PropertiesCentroid Location ComparisonVolume ComparisonImplementationAdvantages and LimitationsMeasures of Estimated EditingImplementationAdvantages and LimitationsHandling Inter-Observer Variation in Quantitative AssessmentSummary of Quantitative EvaluationExample ImplementationSubjective EvaluationThe Acceptance of ContoursThe Source of ContouringThe Preference for ContouringChallenges of Subjective AssessmentSummary of Subjective EvaluationAssessment Based on Intended Clinical UseEvaluation of Time SavingChallenges for Study DesignImpact of Auto-Contouring on PlanningChallenges for Study DesignSummary of Clinical Impact EvaluationDiscussion and RecommendationsReferences
 
>>

Related topics