Comparison of processing methods for 18F-florbetapir PET amyloid quantification

Objectives: The IMMUNE-AD study (NIH R01-AG022304) is investigating potential mechanisms by which physical activity counteracts the negative effects of the APOE ε4 allele, a genetic risk factor for late onset Alzheimer’s disease. Among other parameters, the IMMUNE-AD study measures amyloid accumulation over a two-year period based on 18F-florbetapir PET scans. Although the Centiloid score has emerged as a standard measure of amyloid burden, alternate processing methods might provide higher precision and statistical power. To achieve this goal, we analyzed florbetapir PET images by different methods and compared statistics of ε4+ and ε4- cases.

Methods: 18F-florbetapir PET images were acquired and reconstructed under the ADNI3 protocol (4x5min frames, 50-70min post-injection) at a single site with a single scanner (Biograph mCT 4-ring). Participants are cognitively normal elders (65-80yrs), of which 18 ε4+ (ε3ε4) and 17 ε4- (ε3ε3) have had initial PET scans to date. Centiloid scores were computed using a Level-3-validated SPM8 pipeline (ref: Navitsky et al, 2018), which mapped 50-60min data (filtered 5mm Gaussian) to the MNI152 template and computed SUVR using standard Centiloid VOIs (cortex, referenced to whole cerebellum). In addition, cortical SUVR was computed from 50-70 min PET data using patient-specific VOIs in native space based on MRI T1 and FLAIR images. MRI segmentation was done in FreeSurfer v6 using the Human Connectome Project pipeline (-FLAIRpial and -no-conf2hires flags). PET analysis was done using PETSurfer in two ways: (1) “Standard” processing (filtered 5mm, co-registered summed images, without partial volume correction) and (2) “Enhanced” processing (unfiltered, frame-by-frame processing, with partial volume correction 5mm kernel). In both cases, SUVR was computed twice, using whole cerebellum (WC) or cerebellar gray matter (CGM) as reference. For consistency in comparisons, SUVR was rescaled to Centiloid units via linear regression to standard Centiloid scores.

Results: Normal PET studies (Centiloid<24: 14 ε4+ and 16 ε4- cases) were compared statistically (see Table). The normal ε4- cases are expected to be tightly clustered, and thus the standard deviation (SD) indicates the noise associated with PET analysis. Native-space processing had smaller SD than template-based Centiloid processing. Although CGM-reference had larger SD than WC-reference for both Centiloid and native-Standard processing, CGM-reference had the lowest overall SD for native-Enhanced processing. Next, the likelihood that the normal ε4+ and ε4- cases are distributed similarly was evaluated (Mann-Whitney test). The Centiloid and native-Standard results support this hypothesis. However, the native-Enhanced results reject it (p<0.05), indicating that the “normal” ε4+ cases have greater amyloid burden than the normal ε4- cases.

Conclusions: These results indicate that the precision of amyloid PET measurements is improved with Native-Enhanced processing, compared to standard Centiloid processing. Native-Enhanced processing appears to have greater ability to stratify cases with low amyloid levels. Further study is planned with dual-timepoint scans, when available, to evaluate longitudinal precision of these processing methods.