Sci-Fri AM: Imaging - 06: The role of body mass and gender in atlas construction for attenuation correction in PET/MRI.

ABSTRACT

Attenuation correction (AC) in PET/MRI is difficult as there is no clear relationship between MR signal and 511 keV attenuation coefficients (µ) as there is with CT. One approach is to register a pre-defined atlas of µ to the PET/MRI for AC. However, the design of the atlas may strongly influence the quantitative accuracy of the AC. Here we compare 3 different atlas design approaches and evaluate their performance in an oncology patient population. The 3 strategies were use of BMI-dependent atlases; use of gender-dependent atlases, and use of a gender- and sex-independent atlas. Seventeen patients were imaged with FDG PET/CT and subsequently scanned with 3T MRI. MR and PET/CT images were coregistered, CT scans converted to µ-maps, and the resulting MRI/µ-map paired data were used to construct 6 atlases averaged male and female atlases, averaged BMI-specific atlases (obese >30, overweight 25-29.9, Normal 18.5-24.9), and a single atlas comprised of all patients averaged together. The atlases were then used for PET AC for patients not included in the construction of the atlas in a leave-one-out manner. Resulting PET images were compared to each other and to the gold-standard CT-based PET reconstructions across all voxels and tissue-specific regions (soft-tissue, bone, lung). Sex-specific atlases yielded best results (average relative percent error over the 3 VOIs = 0.4509) & BMI-based atlases yielded highest average relative percent error at 0.9340. In all cases, highest errors were in the VOIs located in the livers.