Possible errors with MR-based attenuation correction for brain imaging in PET/MR scanners

Objectives Template-based methods for attenuation correction (TAC) of brain studies in PET/MR scanners are easy to handle and deliver reliable attenuation maps (AM) in short time. Since a template is not expected to match measured attenuation data exactly, the source of potential errors were analysed in this study using TAC recently introduced by our group [1]: i) choice of template reference head (size, form, gender); ii) skull anomalies of the specific patient; iii) template smoothness influence on the reconstructed images.

Methods Eight normal subjects (4 f, 4 m) underwent a 2-bed-position transmission scan in the Siemens ECAT HR+ PET covering the neck region. i) 8 templates were created with each one of the 8 heads as reference. Eight patients (5 f, 3 m; 4 with, 4 w/o brain lesions) were measured in the Siemens BrainPET/MR scanner. AM with the 8 templates were generated and used for reconstruction. A two-sample t-test was performed between all possible map pairs within each patient group and VOI-atlas was used to analyse the reconstructed images. ii) PET data of two tumor patients with a skull anomaly were reconstructed with one template as described in i) and with the same, but adapted one with simulated skull lesion. iii) For one normal subject CT data were available. After converting the CT into a mu-map, 3D Gaussian kernels (3, 6, 10 mm) were applied. The PET data were reconstructed with TAC and the 3 CT-AC.

Results i) The 8 AM did not differ significant (p<0.001). The VOI-atlas analysis yielded an absolute mean variation of 2.0% among all possible image pairs, with single exceptions of 4%. ii) Only the area extremely close to the skull lesion showed a maximum error of 3%. iii) The mean difference between the VOIs reconstructed with CT-AC 3mm and 10mm was 0.21±0.46%, while the mean deviation between TAC and CT-AC 3mm was -2.10±1.89% with a maximum of -4.86%.

Conclusions The present error analysis confirms that our attenuation template method provides a reliable attenuation correction of brain PET data measured in PET/MR scanners. [1] Rota Kops E et al. IEEE NSS/MIC Conf Rec, pp 4327, 200