Abstract
241591
Introduction: Spatial resolution estimates in emission tomography are conventionally measured using two distinct methods: 1) by reconstructing point and/or line source scans using the filtered back-projection algorithm and calculating the full-width half-maximum (FWHM) of a Gaussian fit to line profiles taken through the radial and tangential directions, and 2) by generating a ground truth digital reference object (DRO) of a suitable phantom acquisition and performing a matched-filter analysis (MFA) to find the FWHM of the Gaussian that produces the best match to the reconstructed image. Both methods have their limitations: point/line source profiles do not take object size and object-to-background contrast ratios into account; MFAs, as they are conventionally used, are sensitive to noise and registration errors and provides a single resolution estimate; neither account for locally dependent convergence behavior, which results in different spatial resolutions in different regions. We have developed a novel approach to PET & SPECT image resolution characterization, which we term Recovery Coefficient Equivalent Resolution (RCER), that is more relevant to routine imaging in human subjects, is easy to perform, can be used on existing data (e.g., the NEMA NU-2 IQ phantom), can account for locally dependent convergence, and has potential applications for harmonization, partial volume correction (PVC), and improving routine QC.
Methods: Spill-out recovery coefficients (RCs) were simulated for a range of spherical volumes (diameter 4-52 mm) and resolutions (4-28 mm Gaussian FWHM), and the volume/FWHM^3 and RCout data were plotted and found to lie on a single curve; a three-parameter logistic function was then fit to the data (Figure 1 A); this RCER equation relates the resolution (Gaussian FWHM) required to produce a spill-out recovery coefficient, given the known spherical volume. NEMA NU-2 IQ phantom experiments were performed for a range of PET & SPECT radionuclides and sphere-to-background ratios (SBR) and were reconstructed with various reconstruction parameters (e.g., with and without resolution modelling (RM), filtering etc). The RCs were measured and were corrected for spill-in from the known SBR; the RCER equation was then used to estimate the resolution of each sphere in the reconstructed images. The RCER average spatial resolution was then compared to resolution measurements obtained from MFAs.
Results: Spatial resolution estimates obtained using an MFA and the RCER approach for a range of PET & SPECT isotopes are shown in table 1. The RCER approach produces similar spatial resolution estimates to MFAs, but with the addition of individual sphere measurements. An example showing spatial resolution measurements of 18F and 68Ga from a long axial FOV PET scanner (Siemens Biograph Quadra) is shown in Figure 1 B.
Conclusions: We have formulated a single equation that relates the effective spatial resolution of a spherical object to the measured RC, producing similar resolution estimates compared to MFAs. The RCER methodology requires 3 inputs (sphere volume, measured RCs, SBR) and thus is easier to perform than an MFA, while also revealing additional information such as individual sphere resolution. This may be exploited to investigate a range of factors that influence PET & SPECT resolution; different phantom configurations could be used to probe the variation of resolution across the PET field-of-view, variations in resolution/recovery due to the SPECT collimator-detector response function, and algorithm dependent object convergence. The additional resolution information may also be used to facilitate harmonization efforts which traditionally use an MFA approach. Additionally, the RCER methodology may be used as a PVC model to improve tumor dosimetry estimates; this is being deployed in the MIRDpvc software (under development), an effort from MIRDsoft and the MIRD committee to bring a simple PVC tool to the wider community.
