《核医学影像中的数据处理课件.ppt》由会员分享，可在线阅读，更多相关《核医学影像中的数据处理课件.ppt（108页珍藏版）》请在三一办公上搜索。
1、核医学影像中的数据处理,中国科学院高能物理研究所北京市射线成像技术与装备工程技术研究中心贠明凯,Modern Nuclear Medical Imaging,Acquire,Process,Apply,Scanners Computers Users,Outline,Data organizationCorrection methodsRebinning Image reconstructionImage registration and fusionDICOM and PACS,Outline,Data organizationCorrection methodsRebinning Imag
2、e reconstructionImage registration and fusionDICOM and PACS,Data organization,List modeHistgramSinogramLinogram,SinogramPET,Sinogram,Projections and Sinogram,SinogramPET,Sinogram,Projections and Sinogram,SinogramSPECT,2D VS. 3D,Septa between crystal ringsLower sensitivityLower randomLower scatter2D
3、reconstructionNo septaHigher sensitivityHigher randomHigher scatter3D reconstruction or hybrid reconstruction,Outline,Data organizationCorrection methodsRebinning Image reconstructionImage registration and fusionDICOM and PACS,Scatter Coincidence,Trues Coincidence,Random Coincidence,True Counts & No
4、ise,Normalization,A,B,C,D,Need to correct the data,Correction methods,random“dead time”normalizationscatterattenuationdecay,Arc correctionDepth of interactionMotion correctionPartial volumeAxial of rotationCamera head tilt,Random,Finite time window withEnergy window Coincidence timing windowActivity
5、,Random,Tail fittingsimplestSmall changes in tail, great changes in estimateEstimation from singles ratesMeasure the single count rate on each detector for a given time windowSubtracting from the prompts between detector pairSingles rate is much larger than that of coincidence eventsSingle rates cha
6、nge in the same way over time,Delayed coincidence channel estimation,One channel is delayed before being sent to coincidence processingSubtracted form prompt coincidencesAdvantage AccurateSame dead time environment as prompt channelDisadvantage Increased system dead timeDoubling of the statistical n
7、oise due to random,Dead time correction,Decaying source experiment is performed,Dead time correction (con),Look up tableUniform sourceKnown quantityShort livedLinear extrapolation from count rate for a given level of activity,Normalization,Causes of sensitivity variationsSumming of adjacent data ele
8、mentsDetector efficiency variationsGeometric and solid angle effectsRotational samplingTime window alignmentStructural alignmentsepta,Summing of adjacent data elements,Geometric and solid angle effects,Rotational sampling,LOR at the edge are sampled less than LOR close to the center,Crystal interfac
9、e factors,Time window alignment factors,Normalization methods (con),Direct normalizationSimplest approachAdequate statistical qualityVery uniform activity sourcesScatter in normalization should be substantially different from normal imaging,Normalization methods (con),Component-based normalization,S
10、catter correction,LORs recorded outside object boundary can only be explained by scatterThe scatter distribution is very broadScattered coincidences fall within the photo-peak window mainly due to scattered once,Scatter correction,Energy spectra distribution of scattered 511KeV photons according to
11、the number of times each photon scatters,Scatter correction,Empirical scatter correctionsFitting the scatter tailsDirect measurement techniqueEnergy window techniquesDual energy window methodsMultiple energy window methodsConvolution and de-convolutionSimulation-based scatter correctionAnalytical si
12、mulationMonte Carlo simulation,Fitting the scatter tails,Simplest approachFit an analytical function to scatter tailsSecond order polynomial or 1D GaussianCoincidences outside the object are entirely scatter eventsNot always well approximated, particularly in thorax,Direct measurement technique,Only
13、 applicable to PET with retractable septaStepsMake a measurement of the same object with and without septaScaling septa extended projections for different efficiencySubtract from projections of polar angle 0Estimate the oblique scatter by interpolation of the direct plane scatter,Dual energy window
14、methods,Dual energy window methods,Multiple energy window methods,Scatter Correction,Analytical simulation,Scatter Correction,Single Scatter - Model based correctionCalculate the contribution for an arbitrary scatter point using the Klein-Nishina equation,BeforeScattercorrection,AfterScattercorrecti
15、on,Attenuation correction,Attenuation in the body is equal to that of source lying along the same LOR,Zaidi H, Hasegawa B. J Nucl Med 2003; 44:291-315.,SPECT,PET,Attenuation correction (con),Measured attenuation correctionCoincidence transmission dataLong-lived positron emitterNormally more than one
16、 rod source are usedSinogram windowing is applied provide location of rodImpractical in 3DSingles transmission dataShielded point transmission sourceSeparate blank scan is neededSignificant scatter and broad beam,Measured attenuation correction,Coincidence measurement using rod sourceTransmission me
17、asurement using point source,CT scan,Advantage High statistical qualityHigh spatial resolutionSignificant reduction in scan timeDisadvantageFaster CT, slower PETSmaller FOV of CTDifficulty in registration values do not scale linearly,Attenuation correction for PET,Types of transmission images,Coinci
18、dent photon Ge-68/Ga-68(511 keV)high noise15-30 min scan timelow biaslow contrast,Single photon Cs-137(662 keV)lower noise5-10 min scan timesome biaslower contrast,X-ray(30-140kVp)no noise1 min scan timepotential for biashigh contrast,Other attenuation correction methods,Calculated attenuationRegula
19、r geometric outlineConstant tissueSegmented attenuationSegment transmission image according to tissue typeAssigning known attenuation coefficientsForward projection,attenuation correction,Attenuation/Scatter correction,University of Pennsylvania PET Center,No AC or Scatter Corr,AC and Scatter Corr,P
20、hilips Allegro,Arc correction,Different sampling distance at different radial positionEqual sampling distance is required in analytical methodInterpolation methodNearest interpolationLinear interpolationB-spline interpolation (negative values!),DOIdepth of interaction,DOIdepth of interaction（con）,Du
21、al Layer,A Point Spread Function (PSF) describes the response of an imaging system to a point source or point object. A system that knows the response of a point source from everywhere in its field of view can use this information to recover the original shape and form of imaged objects. PSFs are us
22、ed in precision imaging instruments, such as microscopy, ophthalmology, and astronomy (e.g. the Hubble telescope) to make geometric corrections to the final image.,Point Spread Function (PSF),Motion correction,Cardiac motion and respiration,Motion correction（con）,Gated framesList mode,Respiratory mo
23、tion is distributed throughout the whole bodyImpact is rarely on detection, but often affects quantitation,Static wholebody,Single respiratory phase(1 of 7, so noisier), 1 cc lesion on CT,Whole-body respiratory gated PET/CT: Patients,Partial volume effect,CharactersObject or structure being imaged o
24、nly partially occupies the sensitive volume of scannerSignal amplitude becomes diluted with signals from surrounding structuresThe degree of underestimation of radioactivity concentration will depend not only on its size but also on the relative concentration in surrounding structuresCorrection meth
25、odsResolution recoveryUse of anatomical imaging data,A Point Spread Function (PSF) describes the response of an imaging system to a point source or point object. A system that knows the response of a point source from everywhere in its field of view can use this information to recover the original s
26、hape and form of imaged objects. PSFs are used in precision imaging instruments, such as microscopy, ophthalmology, and astronomy (e.g. the Hubble telescope) to make geometric corrections to the final image.,Point Spread Function (PSF),Partial volume effectMAP,assumptions:camera moves along circular
27、 orbitorbit is reproducible calibration method finds system geometry,problem 1: tilting detector,AORAxial of rotation,Offset of AORRotation of AORNutation of AOR,Camera head tilt,Heads need to be exactly parallel to axis of rotation,Correct alignment,Head tilt,pinhole calibration,Dirk Bequ, Kathleen
28、 Vunckx,circular orbit,circular orbit + new model,extension 2: circular orbit + arbitrary small deviations,measurement,model,Michel Defrise, Chris Vanhove,extension 2: circular orbit + arbitrary small deviations,old,new,translations,rotations,1mm,-3mm,1.5mm,-1.5mm,1.5mm,-1mm,1o,-2o,1.5o,-1.5o,3.5o,-
29、2.5o,1mm,1.2,1.4,1.6,1.8,2mm,Outline,Data organizationCorrection methodsRebinning Image reconstructionImage registration and fusionDICOM and PACS,Rebinning,Convert 3D data to 2D,SSRB and MSRB,SSRB- Single-slice rebinningDetection: center slice Simple Fast Resolution lossMSRB- Multi-slice rebinningDi
30、stribute along all intermediate slicesDe-blurring along z-axis,Fourier rebinning,Outline,Data organizationCorrection methodsRebinning Image reconstructionImage registration and fusionDICOM and PACS,Image reconstruction,AnalyticalFBPBPFFDK3D RP,IterativeARTMLEMOSEMOSLSMAP,Analytical algorithms,For ex
31、ample, FBP (Filtered Back-projection)Treat the unknown image as continuousPoint-by-point reconstructionRegular grid points are commonly chosenTreat projection process as line integral theoretically,解析重建-FBP,back projection (BP) = summation of projections,filtered back projection (FBP),FDK,Feldkamp、D
32、avis、Kress,FDK,3D RPRe-projection,Steps of 3D RP,Extract 2D sinogramsReconstruct each with 2D FBP and stack to form 3D imageForward project to calculate missing LORsExtract 2D projection data of all oblique slicesTake 2D Fourier transformBack project data through 3D image matrixRepeat for all angles
33、 and oblique slices,What is iterative reconstruction,Discrete measurements, discrete imageOptimization,Attractions of iterative methods,Either consistent or inconsistent is OKComplex geometryPhysical effects and detection processes can be modeledNon-negativity Great reducing streaking artifactsBette
34、r contrast recovery,Classification of iteration reconstruction methods,ART (algebraic reconstruction techniques)MART (multiplicative ART)AART (additive ART)SIRT (simultaneous iterative reconstruction)SMART (simultaneously MART)BI-ART (block iterative ART)BI-SMART (block iterative SMART)RBI-SMART (re
35、scaled BI-SMART),Statistical algorithms,MAP: Maximize the conditional probability P(image|data)MLEM:Maximize the probability P(data|image),Statistical algorithmsGaussian assumption,P is projection column matrix, A is system matrix, F image column matrix, C is the covariance matrix of the data,Assume
36、d all standard deviations are identical and equal to 1, idealized parallel projection, perfect resolution and no attenuation or other degrading affects,Statistical algorithmsPoisson assumption,迭代重建-MLEM&OSEM,MLEM, OSEM, .,Sinogram,Subset 1,Subset 2,Subset 3,Subset 4,1 3 2 4,Subset order,0,1,2,3,4,10
37、,40,ordered subsets,1 iteration of 40 subsets(2 proj per subset),System matrix,Scan geometryCollimator/detector responseAttenuationScatter (object, collimator, scintillator)Duty cycle (dwell time at each angle)Detector efficiencyDead-time lossesPositron rangeNon-colinearityCrystal penetration,Consid
38、erations of system matrix,Quantitative accuracySpatial accuracyComputation time Storage spaceModel uncertaintiesArtifacts due to over simpleifications,System matrix tricks,FactorizeSymmetrySparsenessApproximationPartial Monte Carlo,System matrix model,Reconstruction image of uniform source,FBP VS. O
39、SEM,FBPanalyticalPros：Single pass LinearFastCons：Streak artifactPoor resolutionCorrection not built-in,OSEMiterationPros：Better resolutionBetter contrastLower noiseCons：Extensive time consumingMemory consumingRequired user training,FBP VS. OSEM,Phantom test (left)Clinical results (right),Outline,Dat
40、a organizationCorrection methodsRebinning Image reconstructionImage registration and fusionDICOM and PACS,Image Registration,PET,CT,PET/CT,Voxel based image registration,Image Registration,Image Registration,算法流程图,相似性测量一般用到的函数有：相同模态图像：残差（sum of square difference）不同模态图像：互信息（mutual information）一般用来做配准
41、的优化算法有：六参数或十二参数的优化一般使用 Powell 优化算法多参数优化一般使用LBFGS（ limited-memory BroydenFletcherGoldfarbShanno ）优化算法（由牛顿算法演变而来）,Image Fusion,Alpha Blending basedAdaptive alpha blending,Alpha blending,Adaptive Alpha blending,Outline,Data organizationCorrection methodsRebinning Image reconstructionImage registration
42、and fusionDICOM and PACS,DICOM and PACS,DICOMDigital image and Communication in MedicineCreated by ACR (American College of Radiology) and NEMA (National Electrical Manufacturers Association) in 1985Two components:Communication protocols and image format standards,DICOM image VS. General image,Struc
43、ture:DICOM contains header and image data sections;Other image file such as BMP, JPG, TIFF, which contain also two sections;Size:The header size of DICOM is variable;The header size of many general image is constantContents:DICOM contains additional patients data such as basic information, study inf
44、ormation and so on;General image header describes basic image parameters, such as image size, compression type,DICOM logical layers,Patient information:Patients name, patients ID, patients birth date, hospital information systemStudy information:Study information such as dose, injection time and add
45、itional examination information such as study name study date;Series information:Series ID, manufacturer and institution nameImage information:The size of pixel, image size, pixel value and how it is encoded,PACS,PACSThe Picture Archival and Communication SystemA system for storage of images and transferring images between computers in different facilities through networksPACS is helpful to provide comparative studies among different image modalities,PACS (con),Thanks,Our Group is growing up,TOF-PET,TOF信息,PET原始数据,偶然符合校正,发射图像,衰变校正,死时间校正,弓形几何校正,衰减校正,散射校正,归一化校正,图像重建,数据重组,数据校正,
