Computational Biology, Part 23 Segmentation and Feature Calculation for Automated Interpretation of Subcellular PatternsInitial GoalPreprocessingSlide 4Feature levels and granularityCell SegmentationSingle cell segmentation approachesVoronoi diagramVoronoi Segmentation ProcessSlide 10Slide 11Slide 12Slide 13Slide 14Watershed SegmentationSlide 16Seeded Watershed SegmentationSlide 18Feature ExtractionMorphological Features2D Features Morphological FeaturesSlide 22Slide 23Illustration – SkeletonEdge FeaturesZernike Moment FeaturesHaralick Texture FeaturesSlide 28Pixel Resolution and Gray LevelsFourier featuresFrequency representationSlide 32Slide 33Slide 34Slide 35Demonstration spreadsheetMATLAB demonstrationSlide 38Wavelet Transformation - 1D2D Wavelets - intuitionSlide 41Slide 42Daubechies D4 decompositionWavelet Feature CalculationWaveletsFeature selectionFeature Selection MethodsMatlab demonstrationsSlide 49Computational Biology, Part 23Segmentation and Feature Calculation for Automated Interpretation of Subcellular PatternsComputational Biology, Part 23Segmentation and Feature Calculation for Automated Interpretation of Subcellular PatternsRobert F. MurphyRobert F. MurphyCopyright Copyright  1996, 1999, 2000-2009. 1996, 1999, 2000-2009.All rights reserved.All rights reserved.This is a micro-tubule patternAssign proteins to major subcellular structures using fluorescent microscopyInitial GoalInitial GoalPreprocessingPreprocessingCorrection for/Removal of camera defectsCorrection for/Removal of camera defectsBackground correctionBackground correctionAutofluorescence correctionAutofluorescence correctionIllumination correctionIllumination correctionDeconvolutionDeconvolutionPreprocessingPreprocessingRegistrationRegistrationNot critical if only using DNA or membrane Not critical if only using DNA or membrane referencesreferencesIntensity scaling (constant scale or contrast Intensity scaling (constant scale or contrast stretched for each cell)stretched for each cell)Feature levels and granularityFeature levels and granularityObjectfeaturesSingleObjectSingleCellSingleFieldCellfeaturesFieldfeaturesGranularity: 2D, 3D, 2Dt, 3DtAggregate/average operatorCell SegmentationCell SegmentationSingle cell segmentation approachesSingle cell segmentation approachesVoronoiVoronoiWatershedWatershedSeeded WatershedSeeded WatershedLevel Set MethodsLevel Set MethodsGraphical ModelsGraphical ModelsVoronoi diagramVoronoi diagramSeedEdgeVertexGiven a set of seeds, draw vertices and edges such that each seed is enclosed in a single polygon where each edge is equidistant from the seeds on either side.Voronoi Segmentation ProcessVoronoi Segmentation Process•Threshold DNA image (downsample?)Threshold DNA image (downsample?)•Find the objects in the imageFind the objects in the image•Find the centers of the objectsFind the centers of the objects•Use as seeds to generate Voronoi diagramUse as seeds to generate Voronoi diagram•Create a mask for each region in the Voronoi Create a mask for each region in the Voronoi diagramdiagram•Remove regions whose object that does not Remove regions whose object that does not have intensity/size/shape of nucleushave intensity/size/shape of nucleusOriginal DNA imageAfter thresholding and removing small objectsAfter triangulationAfter removing edge cells and filteringFinal regions masked onto original imageWatershed SegmentationWatershed SegmentationIntensity of an image ~ Intensity of an image ~ elevation in a elevation in a landscapelandscapeFlood from minimaFlood from minimaPrevent merging of Prevent merging of “catchment basins”“catchment basins”Watershed borders Watershed borders built at contacts built at contacts between basinsbetween basinshttp://www.ctic.purdue.edu/KYW/glossary/whatisaws.htmlhttp://www.ctic.purdue.edu/KYW/glossary/whatisaws.htmlWatershed SegmentationWatershed SegmentationIf starting image has intensity centered on the cells (e.g., If starting image has intensity centered on the cells (e.g., DNA) that you want to segment, invert image so that DNA) that you want to segment, invert image so that bright objects are the sourcesbright objects are the sourcesIf starting image has intensity centered on the boundary If starting image has intensity centered on the boundary between the cells (e.g., plasma membrane protein), don’t between the cells (e.g., plasma membrane protein), don’t invert so that boundary runs along high intensityinvert so that boundary runs along high intensitySeeded Watershed SegmentationSeeded Watershed SegmentationDrawback is that the number of regions may not Drawback is that the number of regions may not correspond to the number of cellscorrespond to the number of cellsSeeded watershed allows water to rise only from Seeded watershed allows water to rise only from predefined sources (seeds)predefined sources (seeds)If DNA image available, can use same approach to If DNA image available, can use same approach to generate these seeds as for Voronoi segmentationgenerate these seeds as for Voronoi segmentationCan use seeds from DNA image but use total protein Can use seeds from DNA image but use total protein image for watershed segmentationimage for watershed segmentationSeeded Watershed SegmentationSeeded Watershed SegmentationOriginal imageSeeds and boundaryApplied directly to protein image (no DNA image)Note non-linear boundariesFeature ExtractionFeature ExtractionMorphological FeaturesMorphological FeaturesMorphological features require some method for Morphological features require some method for defining objectsdefining objectsMost common approach is global thresholdingMost common approach is global thresholdingAlternatives include locally adaptive thresholdingAlternatives include locally adaptive thresholding2D FeaturesMorphological Features2D FeaturesMorphological FeaturesDescriptionDescriptionThe number of fluorescent objects in the imageThe number of fluorescent objects in the imageThe Euler number of the imageThe Euler number of the imageThe average number of above-threshold pixels per objectThe average number of above-threshold pixels per objectThe variance of the number of above-threshold pixels per The variance of the number of above-threshold pixels per objectobjectThe ratio of the size of the largest object to the smallestThe ratio of the size of the largest object to the smallestThe average object distance to