CT-based evaluation of segmented head regions for attenuation correction in MR-PET systems

Attenuation correction (AC) is an important prerequisite for quantitative brain PET in MR-BrainPET systems. The new knowledge-based method segments attenuation-differing head regions solely based on the routinely acquired T1-weighted MR data set of the patient´s head. The original approach (O) was extended (E1-E3) with regard to the MR image quality at different bandwidth/voxel (130 HZ/voxel, 610 Hz/voxel) obtained at a 3T MR TimTrio system with BrainPET insert installed. Based on the Dice coefficient, the automatically obtained MR-based segmentation results for bone and soft tissue were compared with segmented CT data as gold standard modality data. So far, registered multi-modality data (MR, CT, PET) are available for one female volunteer F and two tumor patients T1, T2 with CT data of different image quality. Best results were obtained for BW130-E3 and BW610-E2. For F, the Dice coefficient for bone is up to 0.776 for BW130-E3 and up to 0.723 for BW610-E2 in the best part of the cranial region. The Dice coefficient for soft tissue is 0.867 for BW130-E3 and 0.868 for BW610-E2 in the whole data set used for AC. The SegMR-(SBA) and CT-based AC (CBA) were compared w.r.t. PET-based AC (PBA) for a HR+ PET device. AC with SBA yields very similar results to the gold standard CBA.