A comparative evaluation of scatter correction techniques in high by gdf57j


									IEEE IST 2004 Intcmational Workshop

on Imaging Systems and Tcchniqun
Strcsa, Italy, 14 May 2004

     A Comparative Evaluation of Scatter Correction Techniques in high Resolution
                Detectors Based on PSPMTs and Scintillator Arrays
                             E. Karalil, G. Loudos', N. Sakelios', K. S. Nikita', N. Giokaris2
      ' School of Electrical and Computer Engineering, National Technical University of Athens, 9 Iroon Polytechniou Str.,
        Zografos, 15780, Athens, Greece, Phone:+30(2 10)7722285, Fax:t30(2 10)7723557, email:knikita@cc.ece.ntua,gr
                  21nstituteof Accelerating Systems and Applications, P.0.Box 17214, 10024, Athens, Greece
                    2NationalKapodistrian University of Athens, Panepistimiou 30, 10679, Athens, Greece

Abstract - SPECT images suffer from low conlrast (1s a result o   f         Window Subtraction Technique (DEWST), the Convolution
photons scatter. The standard method for excluding scatter                  Subtraction Technique (CST) and a Deconvolution
component in pixelized scintillators b the application of an energy         Technique. All techniques are compared to the standard
window around the central photopeak channel o each crystal cell,
                                                  f                         method.
but small angle scattered photons still appear in the photopeak
window and they are included in the reconstructed images. A
number o scatter correction techniques have been proposed in
          f                                                                              11. MATERIALS AND METHODS
order to estimate the scatter component but they have not yet been
applied in pixelized scintillotors, where most groups use the               A . Data acquisition
standard one-photopeak windowfor scatter correction.
   In this work we have assessed three subtraction techniques that              The gamma camera that was used for the evaluation of
use a different approach in order to calculate the scatter component        DEWST, CST, DT and the standard method is based on a
and subtract itfrom the photopeak image: The Dual Energy Window             Hamamatsu E 4 8 6 PSPMT and a pixelized CsI(TI)
Subtraction Technique (DEWS?) the Convolution Subtraction                   scintillator, 4.6cm in diameter, 4mm thick, with cell size
Technique (CS?) and a Deconvolution Technique (Or). AN these
techniques are compared to the standard method
                                                                            1.13x1.13mm2. The spatial resolution of the system has been
                                                                            measured and found <2mm in planar imaging [3].
Keyworrls -scatter correction, pixelized scintillators
                                                                            B. Dual Energy Window Subtraction Technique
                       I. INTRODUCTION
                                                                               The Dual Energy Window Subtraction Technique
     Small field of view detectors based on Position Sensitive              (DEWST) proposed by Jaszczak et al. [7] involves data
Photomultiplier Tubes (PSPMTs) coupled to pixelized                         collection in a lower energy window, which provides a
scintillators offer improved spatial resolution compared to the             reasonable approximation for the scatter component in the
standard Anger camera and have been used in the last decade                 photopeak window. Two images are used for scatter
in dedicated small field of view systems for Single Photon                  compensation; the photopeak image P(u) and the lower
Emission Computed Tomography (SPECT) [1][2]. Their                                                ()
                                                                            energy window image L u . If C(u) is tbe corrected image
main applications are planar and tomographic imaging of
                                                                            then: C ( u ) = P ( u ) - U ( u )                      (1)
small animals in laboratory environment and clinical
scintimammography for the early detection of small breast                   where U is the vector of image pixels coordinates and k is
tumors [3]. However SPECT images still suffer from low                      the weighting factor. The average optimal value for k was
contrast as a result of photons scatter, which affects the                  found to be equal to 0.5 [4].
quantification of SPECT images [4]. The standard method for
excluding scatter component in pixelized scintillators is the               C. Convolution Subtraction Technique
application of an energy window around the central
photopeak channel of each crystal cell, but small angle                          The Convolution Subtraction Technique (CST) proposed
scattered photons still appear in the photopeak window and                  by Axehson et al. [5] assumes that the scatter component in
they are included in the reconstructed images. A number of                  the photopeak window can be estimated as the convolution of
scatter correction techniques [5][6][7][8][9] have been                     the measured photopeak data P(u) with a characteristic
proposed in order to estimate the scatter component. These                  function f ( x ) that can be modeled as an exponential
techniques have not yet been applied in pixelized                           function, which is derived Bom the system's Line Spread
scintillators, where most groups use the standard one-
                                                                            Function (LSF) in a scattering medium. In order to determine
photopeak window for scatter correction. In this work we
                                                                             f ( x ) a capillary I . l m inner diameter filled with a v c
have assessed three subtraction techniques that use a different
approach in order to calculate the scatter component and                    solution was placed in the center of a water filled cylindrical
subtract it &om the photopeak image: The Dual Energy                        pot, 15cm in height and lOcm in diameter. Data were
                                                                            collected and line profile of the capillary activity was drawn.

0-7803-859 I-8/04/$20.00 2004 E E E                                    18
This line profile stands for the system's LSF. Using a least            lOcm in diameter, containing a 3001111 99mTc solution,
squares method, the one dimensional exponential function                2mCi/ml. The activity ratios of the hot spots to the
that best fitted the theoretically linear part of log(LSF) was:         background were 1O:l (SI) and 5:l (S2) respectively. The
fix) = 0. 0247e"01x1                                  (2)               detector was placed at a lOcm distance from the bottom of
                                                                        the pot. The breast phantom was imaged for 16min and
where x denotes distance in pixels from the center of the
                                                                        -790,000 counts were collected.
camera's field of view (FOV).
                                                                                              111.      RESULTS
D. Deconvolufion Technique
                                                                             The results from the application of the three methods
     Deconvolution techniques (DT) [IO] assume that the                 (one-photopeak, DEWST, CST and DT) are presented in
measured data P(u)derive from the convolution of the true               Fig.1 and Tables 1-111. As it can be seen &om the images and
data C(u)and a characteristic two-dimensional function                  the corresponding h e profiles, all the techniques offer
 h(u) that can be modelled as an exponential function. The              significantly improved results when compared with the one-
                                                                        photopeak method. DEWST subtracts data corresponding to
latter is derived from the system's Point Spread Function
                                                                        non photopeak photons and thus causes the highest reduction
(PSF) in a scattering medium:
                                                                        to the S N R . CST and DT perform scatter correction at an
    P(u)= C(u) h(u)                                     (3)             image level and assume a detector with as uniform response
Various methods have been applied to solve (deconvolve)                 as possible. On the other hand, they both depend on the
equation 3. We have used the blind deconvolution technique              scattering medium and the distance between the camera and
that is based on the Expectation Maximization (EM)                      the phantom.
Algorithm [II], which assumes that the true data Ck+,(u),in
a specific time k + 1 is expressed as:

with w(u) = FFT-'{                       I
                      I FFUh(u))2I. +Y
                           . ...     .
where FFT denotes the Fast Fourier Transform and y is the
squared noise-to-signal ratio (NSR).
  In our implementation of DT h(u)was defined as:
h ( u ) = 0.0247e~0~071"                        (6)
and the value of the parameter y was experimentally
determined as y = 0.0002 for all cases.

E. Phantom Data

     A hot, a cold and a breast phantom have been used for
the assessment of the three methods. The hot phantom
consisted of three capillaries 7cm long, with 1.5mm inner
diameter and 1.6mm outer diameter, placed at 5mm and 7mm
distances fiom each other and filled with a 9m"rc solution,
8mCi"l. The phantom was placed at a lOcm distance from
the collimator and a pot filled with 300ml of water was
placed between the phantom and the detector. The phantom
was imaged for 150secs and -301,OOOcounts were acquired.
     The cold phantom was a metallic cylinder with 1.5cm
outer diameter, 0.8cm inner diameter and 0.7cm height,
placed at the bottom of a thin plastic pot, 6cm in diameter and
8cm high. The pot was filled with 30ml of a 99mTc      solution,
O.l4mCi/ml. The detector was placed at a lOcm distance
ftom the bottom of the pot. The cold circular phantom was
imaged for 3 minutes and -600,000 counts were acquired.
     The breast phantom consisted of two hot quantities of a               Fig. I : Hot (left column), cold (middle column) and breast phantom
%Tc solution, both 0.5ml in volume, placed under a pot,                    (1.4 column) imaging. (a) The image in the 20% photopeak window.

(b) The corrected image using DEWST. ( c ) The corrected image                             Although scatter correction methods have been used in
using CST. (d) The corrected image using DT. (e) Normalized line                     SPECT imaging, the results presented here are limited in
                                                                                     scintigrdphic mode. However, since these methods are
                                                                                     usually applied prior to reconstruction [14] their use for
Table I Image contrast using three different ratios for ROIs in the cold
  (C) and background (B) region for the standard 20% photopeak                       scatter correction in planar mode seems a correct approach.
             window technique. DEWST, CST and DT.                                          The DEWST uses the energy of each detected photon in
                                                                                     order to determine weather it is located in the selected lower
I    Method      I    C/B        I   (B-C)/B       I     IB-CMB+C)          I        energy window or not. The assumption that this window can
    20%              0.7498           0.2502                0.1430                   be used in order to provide an estimate for the scattered
    DEWST            0.6466           0.3534                0.2146                   photons in the primary photopeak window is independent
                                                                                     from the acquisition mode (planar or tomographic). The
                     0.5399           0.4601                0.2988                   DEWST, as it has been modified, is related to the pixelized
                     0.5547           0.4423                0.2839                   scintillators physics since it uses the energy spechum of each
                                                                                     crystal cell. All system’s non-uniformities do not affect the
                                                                                     method since calibration is performed in pixel level and the
                                                                                     method is applied in pixel level as well. Each crystal cell is
                                                                                     treated as an independent detector with uniform response and
                                                                                     in the case of the used crystal, the response of the system is
     Method           SUB            (B-S 1)/B         (B-S l)/(B+S 1)               considered to be uniform only within each pixel, which has
                                                                                     an area of 1.13x1.13mm2. However this method subtracts
    20%              1.3881          0.3881                 0.1625                   non-photopeak data from photopeak data thus reducing image
    DEWST            1.6191          0.6191                 0.2364                   quality, as it can been seen in Fig. I(%).
    CST         I    1.5113     1    0.5113        I        0.2036                         The CST uses a function f ( x ) that depends on the
    DT          I    1.8191     I    0.8191        I        0.2906                   properties of the detection system to be used. Since this
                                                                                     function is desirable to be independent of the position of the
     Table 111: Image contmt using three different ratios far ROls in the
                                                                                     source within the investigated object, Axelsson et al. [14]
     spot with ratio to background 5:l (S2) and the background (B) region            have extensively investigated the determination of the most
    for the standard 20% photopeak window technique, the DEWST, CST                  suitable f ( x ) in SPECT mode. In their work several
                                  and DT.
                                                                                     positions of the used line source in the field of view and
     Method                          (B-S2)/B           (B-S2)/(B+S2)                variable distances from the detector were used. The function
    20%              1.1326           0.1326               0.0622                     f ( x ) was derived by the tails of LSF, when plotted in semi-
    DEWST            1.2319           0.2319               0.1039                    logarithmic scale, which had minor changes in most
                                                                                     positions. In our approach the capillary was placed at IO
                     1.1758           0.1758               0.0808
                                                                                     different distances from the detector. The used parameters of
                     1.3898           0.3898               0.1631                    the function f ( x ) where the mean values of the 10 different
                                                                                     parameters estimated for each corresponding distance. The
                          IV. DISCUSSION                                             obtained results were superior, when compared with the
                                                                                     standard one-photopeak window method, as it is shown in
     The necessity for pixelized scintillators in dedicated
                                                                                     tables I-Ill. However a unique function f ( x ) that could
SPECT systems that are based on PSPMTs is uncontested
since the thickness of the PSPMT glass window together with                          produce optimum results for different energy windows could
a large intrinsic spread of charge distribution hamper the use                       not be determined.
of planar scintillation crystals [12]. In addition the use of a                            DT seems to drastically reduce scatter component and
thick crystal with good detection efficiency involves a large                        increase image contrast. Moreover DT compensates for any
spread of light distribution with respect to the size of the                         factor that decreases image contrast and resolution, like the
PSPMT that decreases spatial resolution [13]. However,                               collimator sensitivity. In Fig.2 we have applied DT in an
calibration steps and data manipulation are more complicated                         image of a small animal (small mouse) head. As it can be
than in standard Anger type cameras and in many cases                                seen fiom Fig.2b DT causes a significant scatter reduction
processing and correction methods are still an open research                         and improves image contrast.. On the other band DT like
field. Scatter correction has not been extensively investigated                      CST depends on the used energy window and system’s
and the use of an energy window around the photopeak                                 energy resolution, the density of the scattering medium and
channel of each crystal cell is a simple technique, which does                       the measured total counts that determine the S N R
not reject the scatter component that is included in the

                                                                                     [7] R.J. Jaszcnuk, K.L. Grcer, C.E. Floyd, C.C. Harris, R.E. Colcman,
                                                                                             “Improved SPECT Quantitication Using Compcnsation for Scattered
                                                                                             Photons”, J. Nue. Mcd., Vol. 29, pp.893-900, 1984.
                                                                                     [8] Buvat, M. Rodriguez-Villafucnc. A. Todd-Pokmpck, H. Benali, R. Di
                                                                                             Pa&, "Comparative Asscssmcnt of Ninc Scancr Comction Methods
                                                                                             Based an Spcctral Analysis Using M o m Carlo Simulations”, J Nuc.
                                                                                             Med,Vol. 36 (S), pp.1476-1488, A u y s t 1995.
                                                                                     [9] M.C. Gilardi, V. Bcltinardi, A. Todd-Pokropck, L. Milancsi, F. Funio,
                                                                                             “Asscssmcnt and Comparison of Three Scancr Corrcction Techniqucs
                                                                                             in Singlc Photon Emission Computed Tomography”, I. Nuc. Med., Val.
                                                                                             29 (12). pp.1971-1979, 1988.
                                                                                     [IO1 M. Mignotlc, I. Mcunier, J. Sausy, C. Janicki, “Comparison of
 8                                                                                           dcconvolution tcchniqun using a dismbution mixurc parameter
                                                                                             estimation: Application in singlc photon cmission computcd
                 (a)                                  (b)                                    tomography imagcry”. Joumal of ElectraNc Imaging, Val. I I, No. 1,
                                                                                             Jan. 2002.
      Fig. 2. Image of a small animal head. a)The image in the 20%                   [ I l l G. Kontaxakis, L. Straus, “Maximum Likelihood Algorithms for
        photopeak window, b) the comcted image using DT.                                     Image Rcconsrmctian in Positron Emission Tomography”,
                                                                                             Radionuclidcs for Oncolaw- C u r ” Stam and Future Asoects.      r   .

                         V. CONCLUSION                                                       MEDITTERA Publishcn.          73-106, Athens, 1998.
                                                                                     [I21 R. Paoi. R. Sea% R. Pcllcgrini, A. Soluri. G . Tratta, L. Indovina, M. N.
                                                                                             Cinti and G. Dc Vinccntis, Scintillation arrays chanctcrizatian for
      The presented results indicate that the three used                                     photon emission imaging, Nuclear lmtrumcnts and Methods in Physics
techniques (DEWST, CST and DT) can play a very important                                     Rescarch Scction A Aceelcrators, Spcctromctcrs, DCtCClOK and
role in scatter rejection in scintillator array detectors.                                   Associatcd Equipment, Volumc 477, lssucs 1-3, 21 January 2002,
Scintimmamography is a research area where these methods                                     Pages 72-76.
                                                                                     [I31 J.H. Kim. Y. Choi, K.S. loa, B.S. Sihn, J.W. Chang, S.E. Kim, K.H.
would improve image contrast and allow the early detection                                   Lee, Y.S. Choc, B.T. Kim. “Devclopmcnt o f a MiniaNrc Scintillation
of small tumors. All these techniques are for the present                                    Camen Using an Nal(T1) Scintillator and PSPMT for
being evaluated in SPECT mode and applied in clinical data.                                  Scintimammagraphy”, Physics in Mcdicinc and Biology, Volumc 45,
Finally the performance of these techniques can be explored                                  Issue I I, pp.3481-3488, Novcmbcr 2000.
                                                                                     [I41 M.C. Gilardi, V. Bettinardi, A. Todd-Pohpek, L. Milancsi, F. Fazio,
in other array detectors, where each detector element is                                     “Asscssmcnt and Comvarisan of Three Scatter Correction Tcchniaucs
individually processed. CdTeZn cameras provide an ideal                                      in Singlc Photon Emission Computed Tomography”, J. Nuc. Med., Vol.
field where these techniques could be applied.                                               29(IZ),pp.1971-1979,1988.


   The authors would like to thank S. Majewski and D.
Weisenberger from “Jefferson Lab” and R. Pani from
Univeristy of Rome “La Sapienza” for offering part of the
used equipment and their experience.


     Weiscnbergcr AG, Kross B, Majewski S, Wojcik R, Bradlcy E, Saha
     M. “Design fcalures and performance of a Csl(Na) array based gamma
     camcra for small animal gcnc mearch”, IEEE Trans Nucl Sei. 1998;
     45(6): 3053-3058.
     R. Pani, R. ScaR, R. Pellegrini. A. Solun, G. Tratta, L. Indovina, M. N.
     Cinti and G. De Vinecntis, “Scintillation arrays charactcrizatian for
     photon emission imaging, Nuclear Insmnncnts and Mclhods in Physics
     Rcscarch Section A ACCClcratocs, Spectrometers. DC~CC~OK            and
     Assaciatcd Equipmcnr‘, Volume 477, Issues 1-3, 21 Janualy 2002,
     Pages 72-76.
     G . K.Loudos, K.S. Nikita, N.D. Gialraris, ct. al “A 3D High Rcsolution
     gamma-camcra far Radiophmaccuticsl Sludics with Small Animals”,
     Applicd Radiation and Isotopes, Volumc 58, Issue 4, April 2003, Pages
     G. Laudos, N. Sakclias, N. Giokaris, K. Nikita, N. Umoglu, D.
     Main-, ”A modification of lhc Multiple Encrgy Window Subhaclion
     Mcthod for Scancr Compcnsation in Pixclizcd Scintillators for
     SPECT’, acccptcd for publication in NIM. 2004.
     B. Axelsson, P. Msaki, A. Israclsson, “Subtraction of Compton-
     Scattered Photons in Single-Photon Emission Computed Tomography“,
     J Nuc. Med.., 25.pp.490494. 1984.
     M.A. King, G.J. Hademcnos, S.J. Gliek, “A Dual Window Method for
     Scatter Comction”, J NUC.M d . Vol. 33 (4). pp.605-61 I , April 1992.


To top