Comparison and cutoff values of two amyloid PET scaling methods: centiloid scale and amyloid-β load.

OBJECTIVE

To enhance amyloid-β (Aβ) positron emission tomography (PET) diagnostic accuracy and adjust for tracer differences, the amyloid-β load (AβL) and centiloid (CL) scale were developed. However, the correlation and superiority of these indicators remain unclear. This study aimed to elucidate the correlation and determine cutoff values.

METHODS

Data from 281 consecutive participants (144 males; mean age, 69.6 years) in the Parkinson's and Alzheimer's Disease Dimensional Neuroimaging Initiative study were analyzed. Initial data, including Aβ PET, structural MRI, and Global Clinical Dementia Rating (G-CDR) scores, were reviewed. CL was calculated using the cerebral cortical and striatum Volume of Interest templates (VOI) and whole cerebellum VOI. AβL was calculated using voxel-wise regression with two canonical images representing non-displaceable and specific binding. Visual estimation was performed by five radiologists, with two independently classify as visual-only Aβ-negative or Aβ-positive. The discrepancies were resolved through majority consensus involving a third observer. Additional visual estimation was performed for cases with CL value under 10 but visual-only Aβ-positive, and those over 30 but Aβ-negative, by two nuclear medicine radiologists. The discrepancies were resolved through discussion and classify as final Aβ diagnosis negative or positive.

RESULTS

CL and AβL correlated linearly (CL = AβL × 2.10-9.40, R2 = 0.923). In predicting final Aβ diagnosis negative versus positive, both CL (AUC = 0.996) and AβL (AUC = 0.997) were significant, with no significant differences (P = 0.882). The Youden Index was 23.3 for CL and 15.6 for AβL. In predicting cognitively normal participants (G-CDR = 0) from others, both CL (AUC = 0.687) and AβL (AUC = 0.667) were significant, with no significant differences (P = 0.379). The Youden Index was 11.9 for CL and 11.6 for AβL. Using a non-biased Gaussian mixture model, the normal group's mean and SD were -2.5 (SD = 6.5) for CL and 3.7 (SD = 3.0) for AβL, resulting in 95% limits (mean + 2SD) of 10.4 for CL and 9.7 for AβL.

CONCLUSIONS

This study demonstrated a high correlation and non-inferiority between CL and AβL.