Speaker
Description
Photon scatter remains a major source of image degradation in Single Photon Emission Computed Tomography (SPECT), leading to reduced contrast recovery and quantitative accuracy. Conventional reconstruction frameworks typically reject these scattered events. This study investigates the utility of rejected scattered data as a "scatter-informed" initial estimate for Maximum Likelihood Expectation Maximization (MLEM) [1] reconstruction to improve convergence and image quality in low-count conditions. Simulations were performed using the SIMIND [2] Monte Carlo toolkit, modelling a high-resolution SPECT system with a Jaszczak phantom containing six hot spheres (radii: 0.80 - 3.00 cm) and a 4:1 activity concentration ratio of $^{99m}Tc$ radionuclide. The primary photopeak energy of $^{99m}Tc$ is 140 keV. Scatter was estimated using the Triple Energy Window (TEW) method, with the primary photopeak at 130–150 keV and adjacent scatter windows at 120–130 keV and 150–160 keV.
We compared attenuation corrected reconstructed images from a uniform initial estimate and scatter-derived initial estimate generated by back-projecting the TEW scatter profiles for 25 MLEM iterations using CASTOR [3] software. The Fig. 1 shows central trans-axial slices of the Jaszczak phantom reconstructed at different MLEM iterations for the two initialization strategies. Results demonstrate that the scatter-derived initialization yields notable improvements at early iterations; as illustrated in Fig. 1 ($4^{th}$ iteration), it achieves visibly reduced background noise while preserving contrast compared to the uniform initialization.
Fig.1 The top row shows images reconstructed using uniform initial estimate, and the bottom row shows images reconstructed using TEW-based scatter estimate. The images are normalized.
Quantitative evaluation was performed using NEMA-based metrics [4], including Contrast Recovery Coefficient (CRC), Coefficient of Variation (CoV), and Contrast-to-Noise Ratio (CNR) as defined below:
$$ CRC=\frac{\left(\frac{C_{hot}}{C_{bkg}}-1\right)}{A-1}, COV=\frac{\sigma_{bkg}}{C_{bkg}}\ , CNR=\frac{C_{hot}-C_{bkg}}{\sigma_{bkg}}\ . $$ where, $C_{hot}$ is measured average counts in the region of interest (ROI), $C_{bkg}$ is corresponding average background counts, $\textbf{A}$ is the true hot to background Ratio and, $\sigma_{bkg}$ is standard deviation of the background ROI counts. At the $4^{th}$ iteration, CRC improves by approximately 12-18\% across sphere sizes, while CNR shows an enhancement of 15-20\% compared to uniform initialization at comparable noise levels. Fig. 2 shows that the scatter-informed approach achieved a peak CNR at the $4^{th}$ iteration for largest sphere. In contrast, the uniform estimate required 14 iterations to reach its peak CNR. These improvements are accompanied by better visual delineation of structures and reduced background contamination. Fig.2 CNR as a function of MLEM iteration number for reconstruction using uniform and scatter initial estimates. Maximum CNR for biggest sphere are marked with arrows. However, at higher iterations, the benefit diminishes due to noise amplification inherent in the scatter derived image. This is evident in Fig. 1 ($12^{th}$ iteration), where pronounced boundary artifacts begin to emerge. This study highlights that scattered photons, typically discarded, contain valuable spatial information that can be leveraged to enhance early-iteration image quality. The proposed scatter-informed initialization provides a practical pathway for improving reconstruction performance in low-count and high-scatter imaging scenarios, with potential implications for accelerated and dose-efficient SPECT imaging.
[1] Lawrence A Shepp and Yehuda Vardi. Maximum likelihood reconstruction for emission tomography. IEEE transactions on medical imaging, 1(2):113–122, 2007.
[2] Michael Ljungberg et al. The simind monte carlo program. Monte Carlo calculation in nuclear medicine: Applications in diagnostic imaging, pages 145–163, 1998.
[3] Thibaut Merlin et al. Castor: a generic data organization and processing code framework for multi-modal and multi-dimensional tomographic reconstruction. Physics in Medicine & Biology, 63(18):185005, 2018.
[4] NEMA et al. Nema standards publication nu 1-2018: Performance measurements of gamma cameras. Rosslyn Virginia: National Electrical Manufacturers Association, 2018.