; Issues in PEM Image Reconstruction
Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

Issues in PEM Image Reconstruction

VIEWS: 54 PAGES: 23

  • pg 1
									             Issues in
      PEM Image Reconstruction

                        Pedro de Almeida

         Instituto de Biofísica e Engenharia Biomédica
             Faculdade de Ciências da Universidade de Lisboa




02.07.2002                           Workshop on Positron Emmission Mammography
                                                                  Lisboa - Portugal
                    Aims
• To discuss possible ClearPEM geometries and
  their implication on Image Reconstruction.

• To briefly review some published works on PEM
  and PEM Image Reconstruction Algorithms.

• To identify some possible Image Reconstruction
  Algorithms for ClearPEM.

 02.07.2002              Workshop on Positron Emmission Mammography
                                                      Lisboa - Portugal
                Presentation Outline
  I.         PEM acquisition geometry.

  II.        Some results from the literature.

  III. PEM reconstruction algorithms in the literature.

  IV. Possible image reconstruction algorithms for
      ClearPEM.

  V.         Conclusions.

02.07.2002                          Workshop on Positron Emmission Mammography
                                                                 Lisboa - Portugal
       PEM Acquisition Geometry




Two Static Planar       “Ring Detector”                   Four Planar
   Detectors                                               Detectors

                    Solid Angle = Sensitivity
                    Movement = Flexibility
 02.07.2002                     Workshop on Positron Emmission Mammography
                                                             Lisboa - Portugal
      PEM Acquisition Geometry



 - Seated patients (X-ray coupling).
 - Non-tomographic acquisitions.
 - Reduced sensitivity.
 - Unable to image regions near the
 chest wall.
 - Able to image the lymph nodes.
 - Low cost.
 - 3D acquisitions.
02.07.2002                             Workshop on Positron Emmission Mammography
                                                                    Lisboa - Portugal
       PEM Acquisition Geometry
    - Patient in prone position.
    - Incomplete datasets.




    - Significant spatial variance.
    - 3D acquisitions.
    - Unable to image lymph nodes.
                                                     “Evaluation of a high resolution positron
 “Statistical Image Reconstruction in PET with       emission tomography (PET) scanner for positron
 compensation for missing data.”, P.E. Kinahan et    emission mammography.”, L.L. White, et al,
 al, IEEE TNS 44(4), 1997, pp. 1552-1557.            Proceedings of the IEEE MIC 2001 Conf.
02.07.2002                                          Workshop on Positron Emmission Mammography
                                                                                 Lisboa - Portugal
      PEM Acquisition Geometry


  - Patient in prone position.
                                                                         PEM camera - LBNL
  - Irregular sampling.
  - Flexible.
  - Able to image the lymph nodes.
  - 3D acquisitions.
  - ~ 2.3x106 events/mCi (MCarlo).1                                           Adjustable
                                                                              spacing

1. “List-Mode Maximum-Likelihood Reconstruction applied to
Positron Emission Mammography (PEM) with irregular sampling”,
R. Huesman et al, IEEE TMI, 19(5), 2000, pp 532-537.

02.07.2002                                        Workshop on Positron Emmission Mammography
                                                                               Lisboa - Portugal
               PEM Acquisition Geometry

                                         Whole-body PET PEM
             System Sensitivity                       +                   ++
      System Spatial Resolution                       +                   ++
                Attenuation                           +                   ++
Able to image the base of the breast                  +                     –
   Able to image the lymph nodes                      +                   –+
                   Price                              –                    +
                 Demand                              ––                    +
02.07.2002                        Workshop on Positron Emmission Mammography
                                                               Lisboa - Portugal
    PEM Reconstruction Algorithms in the
                literature
Year         Acq. Geometry        Rec.Algorithm          Reference                Same
                                                                                 Authors
1994         2 planar detectors   Backprojection      Med. Phys 21(4),               
                                                        pp 529-538
1995         2 planar detectors   Backprojection      IEEE TNS 42(4),                
                                                        pp 1012-1017
1996         2 planar detectors     Probably          EJNM 23(7), pp
                                  Backprojection         804-806
1998         2 planar detectors   Backprojection     Med. Phys 25(11),               
                                                      pp 2119-2129
1999         2 planar detectors   Backprojection      IEEE TNS 46(6),                
                                                        pp 2192-2196
02.07.2002                                  Workshop on Positron Emmission Mammography
                                                                         Lisboa - Portugal
    PEM Reconstruction Algorithms in the
                literature
Year         Acq. Geometry        Rec.Algorithm          Reference                Same
                                                                                 Authors
2000         2 planar detectors   Backprojection     Radiology, 251(1),              
                                                        pp 280-285
2000         2 planar detectors   Backprojection        JNM, 41(11),                 
                                                        pp 1851-1858
2000         4 planar detectors   List Mode ML        IEEE TMI, 19(5),              
                                                         pp 532-537
2001         2 planar detectors   Backprojection         JNM, 42(6),                
                                                         pp 960-966
2001         2 planar detectors   Backprojection      IEEE TNS 48(3),
                                                        pp 811-815
02.07.2002                                  Workshop on Positron Emmission Mammography
                                                                         Lisboa - Portugal
    PEM Reconstruction Algorithms in the
                literature
Year         Acq. Geometry        Rec.Algorithm           Reference               Same
                                                                                 Authors
2001         2 planar detectors   Backprojection       IEEE TNS 48(3),               
                                                         pp 913-923
2001         2 curved detectors      RAMLA           IEEE MIC Conf Rec
2001     Cylindrical detector      FORE-2FBP         IEEE MIC Conf Rec
              6 sectors
2001         4 planar detectors   List Mode ML       IEEE MIC Conf Rec               


               PEM is deserving significant research efforts !

02.07.2002                                  Workshop on Positron Emmission Mammography
                                                                         Lisboa - Portugal
             Some results from the literature




                                    J Nucl Med,41, 2000, pp 1851–1858.

02.07.2002                   Workshop on Positron Emmission Mammography
                                                          Lisboa - Portugal
               Some results from the literature

                                                          Bkg = 3,7 kBq/cc – Contrast = 8:1




                                                                1 min.                   3 min.
 PEM
phantom


     1cc
   “lesion”
                                                                6 min.                   9 min.
“First results of a dedicated breast PET imager, BPET, using NaI(Tl) Curve Plate detectors”, Richard Freifelder,
Chris Cardi, Iuliana Grigoras, Janet R. Saffer, and Joel S. Karp, IEEE MIC Conference Records 2001.
  02.07.2002                                              Workshop on Positron Emmission Mammography
                                                                                       Lisboa - Portugal
                 Some results from the literature
8 mm “lesion”
                                                            “Adapted” animal PET using
TBR=6.17                                                    BaF2 detectors and multi-wire
                                                            proportional chambers readout.
                             106 counts
106 counts
    A - 67        A          B             C               11 mm “lesion”

    B - 56
                                                                       VUB - PET
    C - 43
                                 TBR
    “Evaluation of a high resolution positron emission tomography (PET) scanner for positron
    emission mammography (PEM)”, L. L. White, R. J. Ott, P. Carnochan, D. M. Duxbury, P.
    Bruyndonckx, S. Tavernier, IEEE MIC Conference Records 2001.
    02.07.2002                                       Workshop on Positron Emmission Mammography
                                                                                  Lisboa - Portugal
               Image Reconstruction for PEM
      The PEM geometry conditions the choice of the Image
                 Reconstruction Algorithm

                                    Backprojection          FORE-2FBP
NON-ITERATIVE
                                                     3DRP



                                   Algebraic
  ITERATIVE
                                                          2D                    3D
                                   Statistical
Can model the response function
of the scanning instrument via a                 FORE FORE   3D                    List M.
system matrix A.                                 OSEM RAMLA OSEM                    EM
  02.07.2002                            Workshop on Positron Emmission Mammography
                                                                     Lisboa - Portugal
Non Iterative Algorithms

                Image Reconstruction for PEM
  Backprojection


                                          10º        20º
                                                            “Aspects of 3-D Imaging
                                                            by classical tomography
                                                            for Dual Detector PEM”,
                                                            M. F. Smith, et al., submitted.



     – Rapid.
     – Easiest to Implement.
     – Good performance for centered lesions.
     – Introduces blurring from neighbouring slices.


   02.07.2002                          Workshop on Positron Emmission Mammography
                                                                    Lisboa - Portugal
Non Iterative Algorithms

                    Image Reconstruction for PEM
                           FORE1                                                FORE-2FBP
                A                                                           – Rapid.
      B                                                                     – Well tested in PET.
                                                   f        N2 Oblique      – Similar problems
      3D                                                s   Sinograms       to FBP.
  ACQUISITIONS                                                              – 2D on 3D data.
                                 REBINNING

 3D Image
 2N-1 Slices
                                                       f
             2D FBP         2N-1 Ordinary                   s
                              Sinograms

   1”Exact
         and Approximate Rebinning Algorithms for 3D PET Data”, Michel Defrise, P. E. Kinahan, D.
   Townsend, C. Michel, M. Sibomana, D,F Newport, IEEE TMI 16(2), 1997, pp 145-158.
   02.07.2002                                          Workshop on Positron Emmission Mammography
                                                                                    Lisboa - Portugal
Non Iterative Algorithms

                  Image Reconstruction for PEM
     3DRP1
           – Standard method used to reconstruct truncated PET data.
           – Based on the FBP algorithm.
           – Needs to estimate 2D projections in unmeasured regions.
           – Low performance for noisier acquisitions (e.g compared to
           statistical algorithms).

                            3DRP                                     Statistical




   1“Analytical 3D image reconstruction using all detected events”, P.E. Kinahan and J.G. Rogers, IEEE
   TNS, 36(1), 1989, pp. 694-968.
   02.07.2002                                          Workshop on Positron Emmission Mammography
                                                                                    Lisboa - Portugal
Iterative Algorithms

                  Image Reconstruction for PEM
                  FORE                                                    FORE-OSEM
         A                                                     OSEM
 B
                                  f
     3D                                   s
     Data
                                  N2 Oblique   - “Accelerates” MLEM.
                                  Sinograms
                  REBINNING                    - Converges to a solution “close”
                                               to the optimal solution.
                                               - Incorrect noise modelling.


                              f
                                      s

                         2N-1 Ordinary
                           Sinograms

     02.07.2002                                   Workshop on Positron Emmission Mammography
                                                                               Lisboa - Portugal
Iterative Algorithms

                 Image Reconstruction for PEM
                                              3D-OSEM
        Uses the same strategy as OSEM on 3D data (without rebinning):

                                               (j s )                         
                              (j s 1)   
                                                  Aij   i
                                                              A
                                                              
                                                                   i
                                                                      yi
                                                               ij A , ( s )
                                                                                
                                                                                
                                                                                
                                                                                
                                              i


                                                                                (*)


    - Correct modeling of the statistical distribution of the data and of
    detector responses.
    - Too slow for clinical use.
    (*) times for Whole-Body PET.
    02.07.2002                                                Workshop on Positron Emmission Mammography
                                                                                           Lisboa - Portugal
Iterative Algorithms

                  Image Reconstruction for PEM
       Row Action Maximum Likelihood Algorithm (RAMLA)1
                                  Fast alternative to the ML-EM
     The reconstructed image is updated after each projection line (row of Aij)
       - Converges to a unique solution.
       - Depends on the choice of  s.

                                     y             
   (js 1)    (js )   s (js )              1 Ais j
                                          is
                                     A , ( s )
                                     i j           
                                                    
                                                   
    s - relaxation term
    is - processing order
      (related to convergence speed)
    1“A Row-Action Alternative to the EM Algorithm for Maximizing Likekihood in Emission
    Tomography”, J. Browne and A.R. De Pierro, IEEE TMI, 15(5), 1996, pp. 687-699.

    02.07.2002                                      Workshop on Positron Emmission Mammography
                                                                                 Lisboa - Portugal
Iterative Algorithms

                 Image Reconstruction for PEM
                                       ListMode – EM
    – Best storage when low number of counts compared to number of
      possible LORs.
    – Suitable for limited angle tomography after some modifications.
    – Needs List Mode Acquisition capabilities.




     “Influence of the angle of incidence on the sensitiviy of gamma camera based PET”, Stefaan
     Vandenberghe, Yves D’Asseler, Jeff Kolthammer, Rik Vande Walle, Ignace Lemahieu, Rudi A Dierckx,
     Phys. Med. Biol, 47, 2002, pp 289-303.
     “A Fully 3D Maximum Likelihood Estimator Algorithm with Attenuation and Scatter Correction
     Adapted to a Limited Angle Positron Camera”, F. Pönisch, W. Enghardt, K. Lauckner, submitted.

    02.07.2002                                       Workshop on Positron Emmission Mammography
                                                                                  Lisboa - Portugal
                       Conclusions
• Backprojection is the most used algorithm in PEM (but provides
  limited accuracy).
• Optimal reconstruction algorithm depends on acquisition geometry.
• Recently developed algorithms (e.g. List Mode ML) allow 3D high
  quality image reconstruction, but may be slow.
• A well proven algorithm should be used at the initial steps of the
  ClearPEM development (e.g. Backprojection, 3DRP, FORE-
  OSEM).
• At later stages (clinical), the use of statistical iterative algorithms
  may be preferred due to the possibility of system modeling.
• The existing experience within Crystal Clear should be used (3DRP
  on STIR, ListMode EM, etc...).
  02.07.2002                          Workshop on Positron Emmission Mammography
                                                                   Lisboa - Portugal

								
To top