Correcting photon attenuation in brain PET-MR using a ZTE sequence and comparison to CT-based attenuation correction.

Objectives One of the main challenges of PET imaging on hybrid PET-MR systems is the absence of the CT for photon attenuation estimation and correction. For brain imaging, the standard method to estimate the attenuation map is based on an atlas approach. However, this approach usually does not include children or postoperative subjects and, consequently, is not optimal for these populations. To overcome this limitation, we propose in this study a segmentation-based approach using a zero echo time (ZTE) sequence [Weiger et al., NMR Biomed 2013], a silent 3D radial MR sequence that measures signal in short T2 tissue. From the ZTE image, an attenuation map containing 3 segments (soft tissue, air and bone) is obtained and compared to the attenuation map derived from a CT acquired on the same patient. An analytical simulation tool was used to evaluate the impact of this segmentation-based approach on PET data.

Methods The same patient underwent a [18F]-FDG PET-CT (Siemens Biograph 6) scan, followed one hour later by a PET-MR (GE Signa) scan. A ZTE sequence was acquired on the head during the MR acquisition. Sequence parameters were as follows: resolution: 1.6x1.6x1.6 mm3, FOV: 24x24 cm2, FA=1°, BW=31.5kHz, NEX=2, scan time=1min. The ZTE image was bias-corrected using ITK N4-bias filter [Wiesinger et al, MRM 2015.]. Then a mask was created (-1[asterisk]image+1) and a histogram-based segmentation was done to create a soft-tissue mask and an air mask. Morphological processing was done to get the final masks. Then the bone mask was deduced by subtracting the air and soft mask form the original mask. A Gaussian filter was applied to smooth the edge between segments. Corresponding attenuation coefficient values were attributed to each segment. The reference attenuation map was derived from the CT image, using the bilinear relation method. For a given patient, synthetic emission PET data were obtained with an analytical simulation tool, using an emission map and an attenuation map derived from the patient CT images. The PET data were attenuation-corrected using either the reference attenuation map or the ZTE derived attenuation map, followed by an analytical reconstruction.

Results Attenuation maps generated from CT and ZTE are visually very similar. The air-bone interface is well segmented in the sinus, nasal and the sphenoid cavities as well as in the airways. The ZTE attenuation map was used for attenuation correction in analytically simulated PET emission data and compared to the CT mu-map used as reference. The mean error over the brain in the emission map reconstructed with a mu-map that ignores bone is 7%, the error is reduced to less that 3% when using a ZTE mu-map accounting for bone attenuation.

Conclusions We showed a method to generate a brain attenuation map from an MR ZTE image containing three tissue segments: bone, air and soft tissue. The ZTE mu-map compares well to the reference CT attenuation map. This method is very promising for its specificity, its simplicity in implementation, its low scan time. It offers an alternative to the standard head attenuation correction based on an adult human atlas, especially for pediatric and non-human primates applications.

• Download figure
• Open in new tab
• Download powerpoint