Page 1
Fromfacedetec,ontothefacesofscien,ficimages:
ScalingAnaly,csforImageDatafromExperiments
LawrenceBerkeleyNa.onalLaboratory,Berkeley,CA,USA
DaniUshizima,HarinarayanKrishnan,TalitaPerciano,DulaParkinson,PeterErcius,WesBethelandJamesSethian
Page 2
DataAnaly,cs&Visualiza,onDAVGroup
Wes Bethel, Daniela Ushizima, Gunther Weber, Dmitriy Morozov, Hank Childs, Talita Perciano, Mark Howison, Oliver Ruebel, Burlen Loring, David Camp, Hari Krishnan
2
Page 3
Collaborators
CustomUI
DomainProcessing
Page 4
Embedded
Lightweight,Collabora,on
TailoredVis
Page 5
ClimateScience• Customizeuserinterfaces:
– Interface–Lat/Long2DGrid,3Dglobe,Con,nentalOverlays.– Op,ons–ZonalMeanAverages,ExtremeValues,PeaksOverThreshold,etc..
• Collabora,onandProvenance:– Collaborate,Control,&CommunicateresultwithpeersRecordandrecreateworkflows.
• ExtendCapabili,es:– ExtendExtremeValueAnalysisorPeaks-Over-Thresholdalgorithmorwritecustomanalysisrou,nesto
exploredata.
Page 6
DataBrowser(SDM,ACS),DeepVadoseZone(PNNL),JohnPeterson,SusanHubbardEnvironmentalScience
Page 7
Astrophysics(YT),Climatescience(R),VTK(python)
Astrophysics
Page 8
6/16/16 8
ImageProcessing,Reconstruc,on,Segmenta,on,andAnalysis
Restofthetalk…
Page 9
Overview1. Inves,ga,ngimage-basedexperiments:
a. MaterialScience-focusedimageanalysis;
b. Health-focusedimageanalysis;
2. Computermethodsandresults;
3. Scalingthroughpartnerships;4. Imageintheexascalelandscape
6/16/169
OURTOOLS
Page 10
CAMERACenterforAppliedMathema,csforEnergyResearchApplica,ons
FractureAnalysisofHigh-resImages 10
Page 12
6/16/1612
Nanoparticle Ocular fundus Head CT Radar image
Page 13
6/16/16 13
SAIDE projects – from nano to meter scale
SIAM2010
ISBI’14+15
AdvancedFunc.Materials2015
UXMagazine2013 PSOC-NCI2011
Real->meImagingActaMicroscopica ACS2014
DemoNCEM2013 DemoESD2012
DemoLSD2013
DemoEETD
IEEEBigData2014
Chemical+
Electronic+
Structural
Chemical+
Structural StructuralChemical
+Structural
Chemical+
Structural
Chemical+
Structural
Chemical+
Structural
Chemical+
StructuralStructural Structural
Chemical+
Structural
Chemical+
Structural
Page 14
ImageAcrossDomains
14
Specimens• Materials,composites,compoundsandbiologicalsamples.
Formats• Tiff,jpeg,hdf5,featurevectors,mul,-resolu,onpyramids,binaries.
DataAnalysis• Morphometry;• Spectralcontent;
• Mul,modal;• Templates.
DataUnderstanding• Clustering;• Classifica,on;• Randomizedschemes;
• Visualiza,on.
Reproducibleresearch• Datarepositories;• Sokwarerepositories;
• Collabora,on.
Page 15
1.Imageanalysis@UCB-BIDS/LBL
6/16/16
Geologicalsamples
Resistantcomposites
Free-sokware,open-source,git,reproducible
Cervicalcells
Pilliden,fier
Page 16
1.a.Imageanalysis@LBL/UCB-BIDS
6/16/16
Geologicalsamples
Resistantcomposites
Page 17
ECRP,CAMERAandDAV
• Scidac2012– Geologicalsamples– Carbonsequestra,on
• MathFoundry2013– MicroCT-imagedsamples– ConfocalandPS-OC
• CAMERA2014– ASCR+BES
6/16/16 17
rock
bone
composite
Page 18
The science question: material resilience sample (CMC) and instrument (microCT) ▪ DetectcracksandfiberbreaksfrommicroCTimagesfromALSto
quan.fytherobustnessandresilienceofnewmaterials:noautomatedmethodsexistforthistypeofanalysis;
▪ Constraints:(1)exis,ngsokwaretoolsincapableofmee,ngthroughputrequirementsandscaletofull-resolu,onoftheexperiment(raw~60GB)(2)unabletoprovidereal-,mefeedback.
WorkwasperformedatLawrenceBerkeleyNa,onalLaboratorybytheCRDCenterforAppliedMathema,csinEnergyResearchApplica,ons(CAMERA)andonALSBeamline8.3.2.Opera,onoftheALSissupportedbyU.S.DepartmentofEnergy,OfficeofBasicEnergySciences.CAMERAissupportedbyjointlybyU.S.DepartmentofEnergy,
AdvancedScien,ficCompu,ngResearchandOfficeofBasicEnergySciences.
t
Pressure & temperature
Page 19
Micro-CT Pattern RecognitionProblem: quantify micro-structural damage of ceramic matrix composites using time-resolved data for full exploration of the micro-tomographycontent; Goal:- Iden,fymaterialfailureanddeformi,esfrommicro-CT,forexample,to
inspectfiberreinforcedCMC,anddendritespermea,ngbaseries;- Real-,mefeedbackaboutdatacollec,onandsamplecondi,on;
Approach: • Develop scalable pattern
recognition algorithms to find defects from 3D images;
• Createsokwaretoolstobeserinterfacehumanstoinstrumentswithhighresolu,onhigh-throughput.
19
19
DOEEarlyCareerResearchProject:ScalingAnaly.csforImageDatafromExperiments(SAIDE)D.Ushizima(P.I.),T.Perciano,H.Krishnan,D.Parkinson(ALS),R.Richie(UCB),E.W.Bethel(LBNL)&J.Sethian(CAMERA)
Page 20
FractureAnalysisofHigh-resImages 20
93N 133N 151N
Deformation evolution
Page 21
FractureAnalysisofHigh-resImages
approachTemplate matching
Apply F3D filtersto improve contrast
to extract compositeApply F3D filters
Template matching
Intersection with"Base Result"
with high tolerance
Template matching with low tolerance
Union
For each slice in the stack Prototype examples
Identification of structures
Rawdata
Page 22
Templatematching
FractureAnalysisofHigh-resImages 22
MSE(x, y) = 1n
[p(i, j)− f (x + i, y+ j)]2i, j∑
NCCC(x, y) =p(i, j)− p(i, j)
i, j∑ f (i, j)− f (i, j)
i, j∑
( p(i, j)− p(i, j)i, j∑ f (i, j)− f (i, j)
i, j∑ )2
#
$%%
&
'((
12
1)Similaritybetweenprototypesandlocalregions:
2)Determinethebestmatches:
approachTemplate matching
Apply F3D filtersto improve contrast
to extract compositeApply F3D filters
Template matching
Intersection with"Base Result"
with high tolerance
Template matching with low tolerance
Union
For each slice in the stack Prototype examples
Page 23
F3Dplugin• Accelerate key image
processingalgorithms• Enablesegmenta,onand
analysisofhighresolu.onimagedatasets
• Requirement:parallel-capablealgorithmstoaccommodatelargedatasizesandtoallowreal-,mefeedback
hVps://github.com/CameraIA/F3D
• Non-linearedgepreservingfilters• Morphologicaloperatorswithvaryingstrel
Image processing at high-resolution
DOEEarlyCareerResearchProgram
Page 24
FractureAnalysisofHigh-resImages 24
Quantitative results
Page 25
FractureAnalysisofHigh-resImages 25
17Xfaster
●●●●●●●●●●● ●● ● ● ● ●
●●
●
020
4060
80
Performance
Data Size (Gb)
Tim
e (m
in)
0 2 5 7 12 19 30
●●●●●●●●●●● ●● ● ● ● ●●
●
●
●●●●●●●●●●● ●●●
●
●
●
●
●
●
F3DF3D Virtual StackSacha
Performance evaluation: comparison between proposed filter and only tool previously available in Fiji
• IntelXeonCPUE5-2660-20GHz• 3NVIDIATeslaK20X+1K40m
Page 26
Terabyte-sizeimagerepresenta,on• Problem:
– Largedatasets(originally16GBperframe)• Solu,on:
– Mul,resolu,onpyramidsatfourdifferentscalesstoredasHDF5chunkedmul,-dimensionalarraysthroughBig-DataViewer;
– Pluginoriginallyoffersinterac,vearbitraryvirtualreslicingofmul,-terabyterecordings,sothattheusercaninspecttheexperimentaldataefficiently;
– Compressfilesandallowencapsula,onofterabyte-sizeimagedatasets,includingmetadata,andop,mizedaccesstomul,plescalesofthedata,bothforvisualiza,onaswellasforprocessing.
– OtheradvantagesofBigDataViewerformaung:a)increasedcompu,ngperformance,b)decreasedcluseringoftheexperimentalarchives,andc)poten,alforparallelI/O.
6/16/16 26
Ref:T.Pietzsch,S.Saalfeld,S.Preibisch,andP.Tomancak.Bigdataviewer:visualiza,onandprocessingforlargeimagedatasets.NatureMethods,2015.
Page 27
Tes,ngfileswithdifferentsizes
6/16/16 27
Scalabilityofthemul,-dimensionalrepresenta,onusingHDF5withincreasingdatasize.
Page 28
Advanced technique: team work
DOEEarlyCareerResearchProgram
Page 29
6/16/16 29
Inven,ngnewcodesforcharacteriza,onofreinforcedcomposites
Page 30
SignificanceandimpactScien=ficAchievement§ Analysisofthinfilmsbyusingscanning
transmissionelectronmicroscopy(STEM)tomographyimagesinsupportofmaterialarchitectureenhancement;
§ Quan,fyporestructureevolu,oninordertocontrolqualityoffabricatedfilms.
§ ResultsusingporosimetryfromSTEMimagescorroboratediniden,fica,onoffabrica,oncondi,onsthatledtothelowesteverdielectricconstantsfortheneededfilms.
§ Collabora,onwithIntel,LBLNCEMandOrganicandMacromolecularSynthesisattheMolecularFoundry,andSLAC,
Researchdetails*§ ReportedlowesteverdielectricconstantsforPMOmatrix
material,usedinmicroelectronics;§ Textureanalysisusingsecond-ordersta,s,csofimageintensity
varia,onstomeasurefilmroughness;§ Newtoolsadaptedto3DstacksforNCEMinstruments;§ Newdevelopments:porosityanalysisusingnewmaterial
architecturedrivers(withT.WilliamsandB.Helms)andspectralanalysisofcataly,cprocesses(withK.Bus,lloandP.Ercius).
Ref:Willsetal,“BlockCopolymerPackingLimitsandInterfacialReconfigurabilityintheAssemblyofPeriodicMesoporousOrganosilicas”,FuncionalMaterials2015.
Imageanalysisforqualitycontrolofmaterialarchitecture
Image-based porosimetry for quality control during assembly of films CAMERA and Molecular Foundry
*WorkwasperformedatLBNLbytheCRDDAVandCAMERA.DAVissupportedbyASCRandCAMERAjointlybyASCRandBES.
Page 31
FeaturedesignforSTEMimagedata
(A)(bluetraces)and(B)(redtraces).
DOEEarlyCareerResearchProgram
Page 32
Finalremarks
• Scalingthroughpartnerships:
6/16/16 32
Page 33
Finalremarks
• Algorithmsinanexascalelandscape– I/Oawareness
– Datareduc,onandin-situanalysis
– Machinelearning
– Experimental/observa,onaldatasets
– Digitaltwin
6/16/16 33