On Enhancing Monte-Carlo Scatter Correction for Y90 Bremsstrahlung SPECT using Guided Filtering

Objectives: Monte-Carlo (MC) scatter correction methods for quantitative yttrium-90 (Y90) bremsstrahlung SPECT imaging with wide, continuous energy spectrum have yielded promising results, especially for hot regions. However, MC scatter corrections still have shortcomings such as relatively poor performance for cold regions and heavy computation complexity. Moreover, improving the accuracy of MC scatter estimation becomes more important for Y90 image reconstruction with wider energy windows. In this work, we investigate how to enhance MC scatter correction for Y90 SPECT image reconstruction using guided filtering (GF).

Methods: Long simulation times are associated with high quality MC-based scatter estimation in iterative reconstruction, which is a challenge especially when multiple updates of the estimate are desired. GF is an approach to yield the filtering output by considering the content of a guidance image based on a local linear model and can be used as an edge-preserving smoothing operator. For given high count measurement, we proposed to use GF with the whole measurement as guidance image for filtering out low count MC scatter estimates. Then, filtered MC scatter estimate was applied to SPECT image reconstruction with OS-EM. To test our approach, for a Y90 phantom measurement with hot and cold spheres, 12 MC scatter estimates were generated on 12 processors for about 8 hours as the last scatter update for multiple scatter updates. Then, 1, 2, 4, 8, 12 of them were randomly chosen and were averaged. After local scaling to the level of the measurement and blurring, GF was applied to these averaged MC scatter estimates to yield the final GF MC scatter estimates. Both non-GF and GF MC scatter estimates were used for single spectral reconstruction (SSR) with a narrow energy window (105-135keV) and joint spectral reconstruction (JSR) with six narrow energy windows covering a wide range (105-285keV, 30keV width for each window).

Results: In SSR with different scatter corrections for phantom experiment, our proposed GF MC methods yielded around 4% higher average recovery coefficient (ARC, ideally 1) in hot regions and about 8% lower residual count error (RCE, ideally 0) in cold regions than MC scatter corrections using 1, 2, 4, 8 estimates. All GF MC scatter corrections yielded better quantitative results than MC scatter correction with 12 estimates. Thus, it seems possible to achieve similar performance of scatter correction with up to 12 times faster computation. In JSR, GF MC scatter correction with 4 estimates yielded 82.16% ARC and 76.4% RCE as compared to MC scatter correction with 12 estimates (81.71% ARC, 77.3% RCE). Thus, 3 times faster MC simulation with GF seems to yield similar performance to MC simulation without GF.

Conclusions: We propose a new MC scatter correction method using GF with the total measurement as guidance image for more accurate quantitative Y90 SPECT reconstruction. Our proposed GF MC scatter correction yielded enhanced quantitative results for both hot and cold spheres in a phantom experiment. Further investigation will be needed to improve quantitative results and/or to reduce MC simulation run time with similar performance. Research support: This work was supported in part by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2017R1D1A1B05035810) and in part by grant R01 EB022075, awarded by National Institute of Biomedical Imaging and Bioengineering, National Institute of Health, U.S. Department of Health and Human Services.