Purpose

Pseudo-CT imaged based on Zero-Echo Time (ZTE) has been proposed as an alternative attenuation method for PET-MR¹. Previous studies² have shown that PET-MR data corrected using a ZTE-based AC map, with no filtering, correlates well with an AC gold standard; the transmission scan acquired using rotating 68Ge rod sources. However, the filtering effects on the ZTE attenuation correction map have not been explicitly discussed, in relation to effects on reconstructed and corrected images, and is what we aim to explore in this work.

Methods

Four patients underwent a 10 min 15O-water scan on an integrated time-of-flight PET/MR (Signa PET/MR, GE Healthcare, Milwaukee, WI). . After generating a ZTE based attenuation map for each subject, accordingly to the method proposed by Wisienger et al.3, Gaussian filtering was applied with four different full widths at half maximum (FWHM) 0, 3, 5, and 10 mm. PET images were reconstructed using time-of-flight ordered subset expectation maximisation (OSEM) with 4 iteration, 28 subsets, a reconstruction matrix size of 265 x 256, 5 mm Gaussian post-filter, using both ZTE and atlas-based attenuation maps. Volumes of interest were defined using a probabilistic template over the co-registered T1w image and transferred to all PET datasets. Total activity concentration values were calculated for 4 major brain regions: anterior cortical (ACR - cingulated and frontal gyrus), posterior cortical (PCR – occipital, parietal and somatosensory-motor cortex), limbic (LR - amygdala, hippocampus, hypothalamus, thalamus) and basal ganglia (BG - caudate nucleus, putamen), as well as whole brain, for each of the four filters. Relative differences ($Diff = X1 – Xref / Xref *100$), between the ZTE corrected images with a filtered AC map and with an unfiltered (0 mm), ZTE corrected with a filter AC and the vendor’s atlas method (HA), and Bland-Altman analysis were computed.

Results

Considering the entire brain and the comparison between ZTE corrected images with a filtered AC and non-filtered, there is a general overestimation (~0.8%) of activity for all volumes and filter sizes. The smallest difference and bias was seen in the 5 mm filtered image (0.50% and 0.51% respectively) while the highest was observed for the 3 mm filtered image (1.10% and 1.09). For both BG and LR, the 5 mm filter exhibited a difference of 1.38 % and 1.56%, respectively, and a bias of around 1.4% (1.37% and 1.55% respectively). In regard to ACR and PCR, the 3 mm filtered image appeared to have the smallest difference (-0.80% and -3.38%) and bias (-0.83% and -3.46% respectively). Even more, when comparing the Diff between the ZTE corrected images, both with filtered and non-filtered AC maps, and the manufacturer’s HA method, we depict a higher Diff (0.56%) for the image in which the 3mm filter was used and the lowest in the one using the 5mm Gaussian filter (-0.03%). All comparisons can be found in Table 1, Table 2, and Figure 1.

Discussion/Conclusion

There is a non-significant overestimation between the volumes reconstructed using a filtered and those with an unfiltered AC map, and also between ZTE and HA corrected images. The volume reconstructed using the AC map smoothed with a 5 mm Gaussian filter was the one showing the lowest difference and bias compared to the standard head atlas (HA) approach. We believe this can be explained by the ZTE method giving a thicker skull estimate than HA. Applying a slight smoothing to match the reconstructed resolution might be preferred to mitigate effects of AC map resampling, but optimal values has to be further investigated.

Acknowledgements

No acknowledgement found.