Changing Voxel Volume in PET Reconstruction Necessitates Re-Optimization: A Demonstration in a Breast Cancer Therapy Response Assessment Trial

Objectives In therapy response assessments using PET/CT, precise measurement of SUV is critical. Recently, the utilization of thinner slice, smaller voxel volume reconstruction has been proposed with varying benefits. An increase in image noise was seen as the major deterrent from clinically utilizing smaller voxel volumes. Therefore, we embarked upon an effort to assess the impact of reconstruction parameters on quantification using 2x2x2 mm3 voxels sizes (8 mm3 voxel volume) compared to the system default 4x4x4 mm3 (64 mm3 voxel volume).

Methods Using the list mode acquired time of flight (TOF) raw data from 10 patients enrolled in a breast cancer FDG PET/CT clinical trial performed using a Gemini 64 TF, Astonish PET/CT, 2mm isotropic voxel datasets (144x144 matrix size using a 576 FOV) were reconstructed using 33, 29 and 15 subsets with 3 iterations. Using RECIST 1.1 criteria, up to 5 measurable lesions were analyzed from baseline to two follow-up time-points. The IntelliSpace Portal System was utilized and 3D ROIs placed while being blinded on the measurable lesions as well as healthy liver and para-spinal musculature analyzed for background uptake measurements. A total of 4 reconstruction runs from each dataset was performed (4mm with default 3i33s and 2mm with 3i33s, 3i29s and 3i15s) reconstruction parameters. The image quality was assessed by three independent, blinded readers. Data were tabulated and statistically analyzed including using correction methodology (Bonferriori) for multiple comparisons.

Results Changing only the voxel size in reconstruction led to substantially noisier images that were rated predominately not preferable. Substantial difference was noted in the 2mm voxel reconstructions between the 3 parameter sets, with the 15 subset consistently rated to be the preferable. Performing an overall assessment of the readers preferences and using the system default 4mm 3i33s reconstruction as the benchmark, the 2mm datasets using the 3i33s were not rated preferable, the 3i29s as preferable but limited due to increased noise level and background inhomogeneity with the 3i15s rated statistically significant (P≤.01) preferable. In the quantitative assessment for 102 lesions across the 30 PET exams in 10 patients, the SUVMax varied compared to the conventional (default) 4mm voxel size by 20%, 20% and 15% respectively for the three different reconstruction parameter sets (TOF 2mm 3i33s, 3i29s and 3i15s). The ranges of difference were 1%-50%, 1%-50%, 1%-44% respectively for the three 2mm voxel volume reconstructions. For lesions that measured SUVMax ≤ 5 in the TOF4mm reconstruction, average % changes were 29%, 29% and 21% respectively with a range of 8%-50%, 10%-50% and 5%-44% respectively. On comparing the background uptakes, SUVMean absolute values differed only by 0.08, 0.09 and 0.1 in Liver; 0.07, 0.08 and 0.08 in Para-Spinal muscle respectively in the TOF2mm 3i33s, 3i29s and 3i15s reconstructions.

Conclusions Smaller voxel volume and this higher definition reconstruction is feasible using current generation photomultiplier tube detector time of flight PET/CT system leading to improved image quality, reader preference and lesion details. When reducing the voxel volume, reconstruction parameters, especially the number of subsets has to be adjusted and optimized to account for the reduced count density and different iterative convergence.