Iterative deconvolution of simultaneous 99mTc and 201Tl projection data measured on a CdZnTe-based cardiac SPECT scanner

Abstract

We present a method of correcting self-scatter and crosstalk effects in simultaneous technetium-99m/thallium-201 stress/rest myocardial perfusion (single photon emission computed tomography) SPECT scans. The method, which is in essence a hybrid between the triple energy window method and scatter modelling, is based on a model of spatial and spectral distribution of projection counts in several selected energy windows. The parameters of the model are determined from measurements of thin rod sources in air when no in-object scatter or attenuation effects are present. The model equations are solved using the iterative maximum likelihood expectation maximization algorithm in the projection space to find estimates of the primary photopeak counts of both radionuclides. The method has been developed particularly for a novel dedicated cardiac camera based on CdZnTe pixellated detectors, although it can also be adapted to a conventional scintillator camera. The method has been validated in anthropomorphic phantom experiments. Significant improvement in defect contrast has been observed with only moderate increase in image noise. The application of the method to patient data is illustrated.

Figure 1

Spectra of uniformly scanned thin rod ^99mTc and ²⁰¹Tl sources measured in air for D-SPECT and conventional gamma camera (Prism 3000, Philips, Cleveland, OH)

Figure 2

Count profiles through the static projections of a ²⁰¹Tl rod source in air on a single detector head measured in the main photopeak window 64-77 keV and the lower window 50-64 keV containing mainly tungsten fluorescence X-rays. Solid lines represent Gaussians fitted to measured data points.

Figure 3

The width (FWHM) of ²⁰¹Tl rod source static projections as a function of distance to the detector for the specified energy windows.

Figure 4

Total spectrum of an anthropomorphic phantom filled with ^99mTc and ²⁰¹Tl activities as described in section 2.7 (configuration 1), measured on D-SPECT. The energy windows described in sec. 2.4 are marked.

Figure 5

Simulated distribution of scattered photons for a point ^99mTc source in a cylinder 60×60 cm at varying depths. Dotted lines - scatter in the 130-149 keV photopeak window, solid lines - true scatter in the 105-130 keV window, dotted-dashed lines - estimations of scatter in the 105-130 keV window using Eq. 3 with fixed patrameters α₃ = 0.43 and λ₃ = 0.18 cm⁻¹.

Figure 6

Simulated distribution of scattered photons for a point ²⁰¹Tl source in a cylinder 60×60 cm at varying depths. Dotted lines - scatter in the 64-77 keV photopeak window, solid lines - true scatter in the 50-64 keV window, dotted-dashed lines - estimations of scatter in the 50-64 keV window using Eq. 3 with fixed patrameters α₁ = 0.22 and λ₁ = 0.2 cm⁻¹.

Figure 7

A profile through heart projection of scattered counts in the scatter energy window (105-130 keV) generated from the NCAT phantom using Monte Carlo simulation (points) and an approximated scatter distribution obtained by convolving the scatter projection in photopeak window (130-149 keV) with the kernel given by Eq. (3) with α₃ = 0.43 and λ₃ = 0.18cm⁻¹ (solid line).

Figure 8

Reconstructed images of the phantom with small 100% defects (6 minute scan): short axis slices, horizontal long axis slices and polar map for reconstructions based on raw (windows 2+6 and 4) and deconvolved projections; inter-iterative smoothing with a Gaussian kernel with σ = 1; ^99mTc images - 5 iterations, ²⁰¹Tl images - 10 iterations.

Figure 9

Contrast noise curves for the phantom with large 70% defects. Top graph shows measured true defect contrast (in perfect image it should be 0.7); bottom graph shows contrast in defect regions for non reduced activity (should be 0 in perfect image). Points on each curve correspond to 1, 2, 3, 5, and 10 iterations. Larger false defect contrast in ^99mTc images is mainly due to attenuation effect in the inferior wall.

Figure 10

Polar plots of LV ²⁰¹Tl images of the phantom with large 70% defects (4 minute dual radionuclide scan) reconstructed from raw data ([64 - 77 keV] + [157 - 177 keV] windows), and projections deconvolved with inter-iterative smoothing (Gaussian kernel with σ = 1). Images obtained from ²⁰¹Tl-only scan of the same phantom aer shown for comparison. Activity concentrations (perfusion) calculated by the QPS software.

Figure 11

Simultaneous ²⁰¹Tl rest - ^99mTc stress study of a 73y old male: (a) uncorrected data reconstruction for 20% 140 keV and 15% 70 keV energy windows; (b) reconstruction from data processed by the iterative deconvolution algorithm; (c) rest image obtaied from uncorrected ²⁰¹Tl only scan (prior to injection of ^99mTc ). The arrows indicate the defect in infero-lateral wall, enhanced after crosstalk/scatter correction.