MR-Based Attenuation Correction Using Ultrashort-Echo-Time Pulse Sequences in Dementia Patients

Attenuation correction (AC) is a critical requirement for quantitative PET reconstruction. Accounting for bone information in the attenuation map (mu map) is of paramount importance for accurate brain PET quantification. However, to measure the signal from bone structures represents a challenging task in MR. Recent F-18-FDG PET/MR studies showed quantitative bias for the assessment of radiotracer concentration when bone was ignored. This work is focused on F-18-FDG PET/MR neurodegenerative dementing disorders. These are known to lead to specific patterns of F-18-FDG hypometabolism, mainly in superficial brain structures, which might suffer from attenuation artifacts and thus have immediate diagnostic consequences. A fully automatic method to estimate the mu map, including bone tissue using only MR information, is presented. Methods: The algorithm was based on a dual-echo ultrashort-echo-time MR imaging sequence to calculate the R-2 map, from which the mu map was derived. The R-2-based mu map was postprocessed to calculate an estimated distribution of the bone tissue. mu maps calculated from datasets of 9 patients were compared with their CT-based mu maps (mu map(CT)) by determining the confusion matrix. Additionally, a regionof- interest comparison between reconstructed PET data, corrected using different mu maps, was performed. PET data were reconstructed using a Dixon-based mu map (mu map(DX)) and a dual-echo ultrashort-echotime- based mu map (mu map(UTE)), which are both calculated by the scanner, and the R-2-based mu map presented in this work was compared with reconstructed PET data using the mu map(CT) as a reference. Results: Errors were approximately 20% higher using the mu map(DX) and mu map(UTE) for AC, compared with reconstructed PET data using the reference mu map(CT). However, PET AC using the R-2-based mu map resulted, for all the patients and all the analyzed regions of interest, in a significant improvement, reducing the error to -5.8% to 2.5%. Conclusion: The proposed method successfully showed significantly reduced errors in quantification, compared with the mu map(DX) and mu map(UTE), and therefore delivered more accurate PET image quantification for an improved diagnostic workup in dementia patients.