Optical surface information-based respiratory phase-sorting and motion-incorporated reconstruction for SPECT imaging.

BACKGROUND

Respiratory motion during the single photon emission computed tomography (SPECT) acquisition can cause blurring artifacts in the reconstructed images, leading to inaccurate estimates for activity and absorbed doses.

PURPOSE: To address the impact of respiratory motion, we utilized a new optical surface imaging (OSI) system to extract the respiratory signals for phase sorting and verified its effectiveness through simulation and patient data. Additionally, we implemented GPU-accelerated motion-incorporated reconstruction algorithms for the SPECT projections, integrating motion information to produce motion-free images from all acquired data.

METHODS

We used the 4D XCAT Phantom to generate attenuation maps and activity images across different respiratory phases, with activity distributions based on patient images. SPECT projections were simulated using the SIMIND Monte Carlo program with Poisson noise. The OSI system was modeled by introducing Gaussian noise into the point clouds on the body surface within the attenuation map. The body surface images were registered across phases using a Gaussian mixture model combined with principal component analysis. The extracted respiratory signals were compared to the center-of-light (COL) approach, with or without filtering and kidney masking. The OSI method was further validated by comparing respiratory signals derived from a real patient using OSI to simultaneous cone-beam CT (CBCT) projections. Two motion-incorporated techniques, namely, 4D reconstruction (4D-Recon) and post-reconstruction registration and summation (post-Recon), were compared with non-motion-corrected images (non-MC) and single-phase gating (Gating). The quantitative evaluation of image quality utilized recovery coefficients (RC), contrast recovery coefficients (CRC), and uncertainty estimation.

RESULTS

In simulation, the correlation between the ground-truth and OSI-based signals remained high and stable (0.99 ± 0.004, p-value < $<$ 0.001 vs. COL-filter with kidney masking). While the kidney mask improved performance (0.87 ± 0.07 without filtering and 0.90 ± 0.06 with filtering, p-value < $<$ 0.001), it was less effective and more uncertain than the OSI method. Validation with patient data showed high consistency in breathing frequencies and phase alignment between CBCT-based and OSI-based signals. For reconstruction, both 4D-Recon and post-Recon significantly enhanced RC and CRC compared to non-MC, with less uncertainty than Gating. In addition, 4D-Recon outperformed post-Recon in certain aspects.

CONCLUSIONS

Our novel respiratory signal extraction approach based on OSI demonstrated superior accuracy and reliability compared to a data-driven method. Applying motion-incorporated SPECT reconstruction using these accurate breathing signals has the potential to enhance image quality and improve absorbed dose quantification in radiopharmaceutical therapy. The relevant reconstruction algorithms are also made available for public use in the open-source library PyTomography.