The wide range of Amyloid PET tracers and analysis techniques used in clinical studies can lead to large variability in cut-off values and quantitative assessment of Amyloid load. The Centiloid scale was developed to normalise neocortical retention measured using different tracers and analysis methods. In this study, we evaluate using the Centiloid scale with the PET-only quantification method CapAIBL (


      The 34 YC and 45 AD PiB PET calibration scans from the Centiloid level-1 study were quantified using CapAIBL, using the Centiloid template and reference regions. Additionally, 991 PiB PET scans (639 HC, 161 MCI, 166 AD and 25 others) from the AIBL study were processed using CapAIBL, and compared to the Centiloid values computed using the recommended SPM pipeline.


      Using the calibration scans, the mean whole cerebellum (WC) SUVR values obtained using CapAIBL were within 2% of the published values, (AD:-0.88%, YC:-1.76%). The regression slope (0.99), intercept (0.27) and R2 (0.99) of the Centiloid transformed WC SUVR values were within the accepted range. Using CapAIBL on the AIBL images, the regression slope (0.98) and R2 (0.99) of the Centiloid transformed WC SUVR values (computed using the Centiloid calibration coefficients) were within the accepted range compared to the SPM derived Centiloid values. The intercept (2.03) was however marginally outside the accepted range (<2). Using the AD+ (SUVR>1.5) and HC- (SUVR<1.3) in AIBL, the mean whole cerebellum (WC) SUVR values were within the accepted range (<2%) for the AD+ (-1.13%), but marginally outside for the HC- (-2.70%). This resulted in a mean error in Centiloid values of -0.46% in the AD+ and -7.24% in the HC-. Using the slope and intercept to perform the Centiloid transformation resulted in a reduced mean error of -0.52% in the AD+ and 1.12% in the HC-.


      Reliable Centiloid measures can be obtained for PiB images using the PET-only method CapAIBL using the recommended calibration method. Using the estimated slope and intercept to further correct the estimation of the Centiloid values might however be desirable to reduce the quantification errors.
      Figure thumbnail fx1