Multi-Atlas-Based Attenuation Correction for Brain 18F-FDG PET Imaging Using a Time-of-Flight PET/MR Scanner: Comparison with Clinical Single-Atlas- and CT-Based Attenuation Correction

In this work, we assessed the feasibility of attenuation correction (AC) based on a multi-atlas-based method (m-Atlas) by comparing it with a clinical AC method (single-atlas-based method [s-Atlas]), on a time-of-flight (TOF) PET/MRI scanner. Methods: We enrolled 15 patients. The median patient age was 59 y (age range, 31-80). All patients underwent clinically indicated whole-body F-18-FDG PET/CT for staging, restaging, or follow-up of malignant disease. All patients volunteered for an additional PET/MRI scan of the head (no additional tracer being injected). For each patient, 3 AC maps were generated. Both s-Atlas and m-Atlas AC maps were generated from the same patient-specific LAVA-Flex T1-weighted images being acquired by default on the PET/MRI scanner during the first 18 s of the PET scan. An s-Atlas AC map was extracted by the PET/MRI scanner, and an m-Atlas AC map was created using a Web service tool that automatically generates m-Atlas pseudo-CT images. For comparison, the AC map generated by PET/CT was registered and used as a gold standard. PET images were reconstructed from raw data on the TOF PET/MRI scanner using each AC map. All PET images were normalized to the SPM5 PET template, and F-18-FDG accumulation was quantified in 67 volumes of interest (VOIs; automated anatomic labeling atlas). Relative (%diff) and absolute differences (vertical bar%diff vertical bar) between images based on each atlas AC and CT-AC were calculated. F-18-FDG uptake in all VOIs and generalized merged VOIs were compared using the paired t test and Bland-Altman test. Results: The range of error on m-Atlas in all 1,005 VOIs was -4.99% to 4.09%. The vertical bar%diff vertical bar on the m-Atlas was improved by about 20% compared with s-Atlas (s-Atlas vs. m-Atlas: 1.49% +/- 1.06% vs. 1.21% +/- 0.89%, P < 0.01). In generalized VOIs, %diff on m-Atlas in the temporal lobe and cerebellum was significantly smaller (s-Atlas vs. m-Atlas: temporal lobe, 1.49% +/- 1.37% vs. -0.37% +/- 1.41%, P < 0.01; cerebellum, 1.55% +/- 1.97% vs. -1.15% +/- 1.72%, P < 0.01). Conclusion: The errors introduced using either s-Atlas orm-Atlas did not exceed 5% in any brain region investigated. When compared with the clinical s-Atlas, m-Atlas is more accurate, especially in regions close to the skull base.