Performance evaluation of a clinical PET system with uniform axial gaps between individual detector rings

Objectives: To assess the effect of a sparse rings (SR) configuration, involving uniform axial gaps between individual detector rings, on the image quality and lesion detectability in clinical PET.

Methods: The Siemens Biograph mMR PET/MR scanner consists of 64 LSO rings, each of 4mm axial length, in a compact ring (CR) configuration. A sparse detector rings configuration was simulated in previously acquired phantom and clinical mMR studies by setting all coincidence events associated with every other ring (the even rings in this case) to zero in the sinogram space, thus mimicking PET acquisitions with 32 detector rings interleaved with 4mm axial gaps. The performance of the SR mMR model was evaluated for 1 NEMA IEC phantom and 2 human brain ¹⁸F-FDG PET/MR studies. We performed 3D PET image reconstructions (OSEM, 1-5 iterations, 21 subsets, 344x344x127 image matrix, 2.086mm x 2.086mm x 2.031mm voxel size, 4mm FWHM Gaussian post-filtering, no resolution modelling) of the compact ring (CR) and SR PET data using the Software for Tomographic Image Reconstruction (STIR) and applying all corrections (attenuation, scatter, randoms and normalization). For the SR reconstructions, we employed the original mMR system matrix size and recomputed the normalization array to account for the SR axial geometric response. To limit the statistical noise elevation expected in the SR acquisitions we also applied projection-based axial interpolation (PAI) and repeated the reconstructions accounting for the axial interpolation in the normalization. Finally, we compared the SR performance against that of an mMR CR configuration of half the axial FOV (HCR).

Results: For the IEC phantom experiments all spheres were visually detectable with all configurations, although the NEMA background variability for a 37mm region of interest increased from 5.6% in CR to 7.8% in SR+PAI, 8.7% in HCR and 9.2% in the SR images. The NEMA contrast recovery coefficients (CRC) were comparable among all 4 configurations and for all spheres. The CR configuration yielded the highest CRC scores for the 10mm (smallest) and 22mm (largest) diameter spheres (41.6% and 75.4% respectively), closely followed by HCR (40.2% and 74.4%), SR (39.7% and 73.9%) and SR+PAI (38.5% and 73.9%). The contrast-to-noise-ratio (CNR) was impacted from the different noise levels of each configuration. The CR configuration yielded the highest CNR scores for the 10mm and 22mm spheres (4.7 and 15.6 respectively), followed by SR+PAI (3.6 and 14.5), HCR (2.9 and 10.3) and SR (2.5 and 9.4). For the human PET/MR studies the target-to-background (TBR) contrast and CNR in two brain lesions of focal uptake were reduced relative to CR by 8.3% and 23.6% in SR, 7.8% and 20.2% in HCR and 1.7% and 5.4% in SR+PAI respectively. All measurements were conducted on the first OSEM full iteration images. In all cases, the TBR slightly increased from the first to the fifth full iteration while the CNR achieved its maximum value at the first full iteration

Conclusions: The introduction of uniform 4mm axial gaps between the Biograph mMR individual detector rings did not considerably affect visual detectability and contrast recovery of any of the NEMA IEC spheres or brain lesions, but elevated the noise to levels comparable to those observed in compact ring mMR acquisitions of half the axial FOV. Projection-based axial interpolation of the SR data significantly limited the noise and recovered most of the CNR losses. The proposed sparse rings configuration can be utilized to double the PET axial FOV for a given number of detectors or reduce the number of detector elements to half for a given axial FOV.