Info
Background:
Accurate attenuation correction is essential for quantitative PET/MR imaging. In current whole-body clinical PET/MR, Dixon-based MR attenuation correction is widely used because of its efficiency and clinical availability [1]. However, this method does not explicitly account for bone attenuation and may therefore introduce considerable quantification errors in bone structures and adjacent regions [2]. Several approaches have been investigated to improve MR-based attenuation correction, including atlas-based, ultrashort-echo-time, zero-echo-time, and deep learning-based methods [2, 3]. Nevertheless, these methods may require additional dedicated MR sequences, image registration, or large training datasets, which can limit routine clinical implementation. Therefore, a practical method for deriving bone-inclusive attenuation information from routinely acquired Dixon MR images remains of clinical interest.
Hypothesis/Objective:
This study evaluated whether a Dixon MR-derived attenuation map generated using a parametric physical model and a three-compartment bone model could improve PET/MR attenuation correction accuracy compared with conventional Dixon-based attenuation correction while preserving soft-tissue quantification.
Methods:
A Dixon MR-derived attenuation map generated using a parametric physical model and a three-compartment bone model was evaluated in this study. The method uses the water- and fat-image pairs of routine Dixon MR images to estimate tissue composition and derive corresponding attenuation values. By explicitly modeling bone as a separate component, the derived attenuation map overcomes a major limitation of conventional Dixon-based MR attenuation correction, which does not adequately account for bone attenuation. The resulting attenuation map was then used for PET attenuation correction. Twenty patients underwent clinical F-18-FDG PET/CT followed by whole-body PET/MRI. Three attenuation maps were evaluated: standard CT-based attenuation correction (CTAC) as the reference, clinical Dixon MR-derived attenuation correction (MRACD), and Dixon MR-derived attenuation map with three-compartment bone modeling (MRACS). To isolate the effect of attenuation correction, the same PET emission data were separately reconstructed using the three attenuation maps. Quantitative evaluation included voxel-wise comparisons of activity distributions in the torso and volume-of-interest analyses of pelvic bone and soft-tissue structures. Bone VOIs included the sacrum, bilateral iliac bones, and bilateral femoral heads. Soft-tissue VOIs included the bilateral gluteus maximus and quadriceps muscles. SUVmax and SUVmean were compared with the CTAC reference.
Results:
The proposed MRACS reduced torso activity-distribution error compared with the MRACD, with a mean absolute error decreased from 42.69 ± 81.93 to 14.65 ± 6.95. The largest improvement was observed in the femoral heads. In the right femoral head, SUVmax relative error improved from -38.10 ± 8.48% with the CTACD to -5.01 ± 5.17% with CTACS, and SUVmean improved from -42.24 ± 7.34% to -6.97 ± 6.17%. In the sacrum, SUVmax bias improved from -15.78 ± 4.86% to 1.56 ± 4.86%, and SUVmean bias improved from -16.69 ± 4.28% to 1.43 ± 5.24%. Similar reductions in quantitative error were observed in the bilateral iliac bones. In contrast, soft-tissue quantification remained stable, and no marked deterioration was observed with the CTACS.
Conclusion:
This study demonstrated that the proposed MRACS using a parametric physical model and a three-compartment bone model improved PET/MR attenuation correction accuracy in the evaluated torso and pelvic regions, particularly in bone. Compared with conventional MRACD, the proposed method substantially reduced underestimation in bone structures while maintaining soft-tissue quantification. The proposed method may provide a practical bone-inclusive attenuation correction strategy for clinical PET/MR without requiring additional dedicated MR sequences, atlas registration, or large training datasets.
Image/Figure:
Click to View
Image/Figure Caption:
Figure 1. Percent difference maps relative to the CTAC reference. (a) MRAC_D shows marked errors, especially in osseous regions. (b) MRAC_S reduces these errors and shows closer agreement with CTAC.
Author
Tung-Hsin Wu
NYCU