From image reconstruction to clinical prognosis: A comprehensive analysis of centiloid-based amyloid PET assessment.

Abstract

Purpose: To investigate the sources of variability in Centiloid (CL) calculations, particularly the influence of image reconstruction and reference region selection, and to examine the relationship between baseline CL scores, visual interpretation and subsequent disease progression.

Methods: 162 aMCI patients who underwent amyloid PET at a single center were retrospectively analyzed. Visual assessment was performed by two nuclear medicine physicians and Centiloid scoring was determined using syngo.MI Neurology Cortical Analysis, using different reference regions (RR) and image reconstruction settings. The CL values were compared against visual interpretation, using a ROC analysis. The value of CL in predicting the onset of Alzheimer's dementia was assessed.

Results: The use of the whole cerebellum as RR provided the most robust and consistent CL values across reconstruction methods. The RR was critical in the case of flutemetamol, as CL varied in more than 20 units between pons and whole cerebellum. Visual classifications and CL values showed strong concordance (area under the ROC curve: 0.9786) and the CL cut-off value that maximized agreement with visual reading was 28 CL. During follow-up, 49% of patients progressed to AD dementia and CL-based amyloid positivity was a significant predictor of progression.

Conclusion: Standardized CL quantification using the whole cerebellum as RR enhances the reliability of amyloid PET interpretation across tracers and reconstruction settings. CL values strongly correlate with visual assessment and are predictive of clinical progression. These findings suggest the potential utility of CL quantification in both clinical and research settings.