Considerations for Performing Level-2 Centiloid Transformations for Amyloid PET SUVR values.

Area of Science:

• Nuclear Medicine
• Neuroimaging
• Biostatistics

Background:

• The Centiloid Project provides a standardized method for analyzing amyloid positron emission tomography (PET) data.
• Standardization is crucial for comparing PET measurements across different research sites and studies with varying acquisition and analysis techniques.
• Existing methods use linear regression for data transformation to a standard Centiloid unit scale.

Purpose of the Study:

• To evaluate multiple post-hoc data transformation variations for amyloid PET measurements.
• To compare the performance of standard linear regression against Deming regression for Centiloid standardization.
• To assess the impact of separate regressions for acquisition and analysis differences versus a single regression approach.

Main Methods:

• Examined several variations of post-hoc data transformation techniques for amyloid PET.
• Compared standard linear regression with Deming regression, which accounts for errors on both axes.
• Investigated the efficacy of single-regression versus separate regressions for acquisition and analysis method differences.

Main Results:

• All tested transformation approaches demonstrated very similar performance in standardizing amyloid PET data.
• The performance of all methods was within the recommended tolerance thresholds for data comparison.
• Deming regression and separate regression approaches did not yield significantly superior results compared to standard linear regression.

Conclusions:

• Various post-hoc transformation methods for amyloid PET data exhibit comparable performance.
• The choice of regression technique (linear vs. Deming) and regression strategy (single vs. separate) has minimal impact on standardization outcomes.
• Current transformation approaches are adequate for enabling reliable data comparisons across diverse studies and sites.