Comparison of static and dynamic analysis techniques for longitudinal analysis of amyloid PET

Background Amyloid PET is a robust biomarker of cortical β‐amyloid accumulation and a candidate endpoint for Alzheimer’s prevention trials. Quantification typically uses Standard Uptake Value Ratio (SUVR) measures acquired over 10‐20 minutes at steady‐state. SUVR may be more susceptible to altered blood flow than modelling of dynamic uptake data from injection to steady state. This may influence quantification, particularly change over time. Here we compared cross‐sectional measures and longitudinal rates of β‐amyloid change from static and dynamic analyses in individuals free of dementia. Method All participants were from Insight46, a sub‐study of the British 1946 birth cohort. Florbetapir PET scans were acquired on a single Biograph mMR PET/MR scanner and reconstructed using pseudo‐CT attenuation maps. Static SUVR images were acquired 50‐60 minutes post‐injection. Distribution Volume Ratio (DVR) was obtained from the Simplified Reference Tissue Model (SRTM) over 0‐60 minutes post‐injection. Both analyses used the same reference (cerebellar grey) and target (composite cortical grey) regions with no partial volume correction. Amyloid positivity was determined from a Gaussian Mixture model of the baseline data, taking the 99th percentile of the amyloid negative distribution. Amyloid accumulators were individuals with annual rates of change above 1.8%/year, representing the mean annualised rate of change in 37 cognitively normal CSF Aβ42 positive ADNI participants (doi:// 10.2967/jnumed.114.148981). Result 390 participants had suitable PET data for cross‐sectional static and dynamic analysis; 195 participants had longitudinal data. DVR and SUVR exhibited high agreement cross‐sectionally (r2=0.79) with similar cut‐points (Table 1) and amyloid positivity rates (94% concordant classification, Figure 1). Longitudinal DVR resulted in more converters and accumulators in the amyloid negative group (Figure 2 and 3). However, comparing baseline amyloid positive/negative groups, longitudinal SUVR provided a larger effect size (Cohen’s d=1.02) than longitudinal DVR (d=0.28) (Table 1), with a sample size of 241 needed to detect a 25% reduction in SUVR per arm (compared to amyloid negatives) at 95% significance, 80% power. Conclusion Static and dynamic PET measures provide similar “state measures” at baseline, but have different longitudinal properties. SUVR may be more sensitive for quantifying accumulation in amyloid positive individuals; dynamic measures may be better