# Amyloid quantification in the oldest-old: selecting regions for optimizing correspondence between postmortem pathology and amyloid PET

## TL;DR

This study finds that using specific brain regions in PET scans improves the accuracy of predicting amyloid buildup in the oldest-old individuals.

## Contribution

The study identifies optimal PET regions for predicting amyloid pathology in the oldest-old, improving diagnostic accuracy.

## Key findings

- The PC2+WM SUVR method had the best predictive accuracy for amyloid beta and neuritic plaque positivity.
- White matter outperformed cerebellar gray matter as a reference region for amyloid prediction.
- Using a limited cortical target region improved PET-neuropathology correspondence in older individuals.

## Abstract

Positron emission tomography (PET) is the current gold standard for assessing amyloid burden in vivo and is often quantified using standardized uptake value ratios (SUVRs). We evaluated the performance of four SUVR calculation methods for predicting amyloid deposition at autopsy in a group of oldest-old participants. We analyzed data from 165 participants from The 90 + Study with both florbetapir PET and postmortem assessments. PET scans were re-aligned to an older age template using a custom MRI-free pipeline. SUVRs were computed from two target regions—combined posterior cingulate and precuneus (PC2) and an established cortical summary region—and two reference regions—white matter (WM) and cerebellar gray matter. Their predictive performance for amyloid beta positivity (Thal phase ≥ 3) and neuritic plaque positivity (moderate or frequent CERAD score) was evaluated using receiver operating characteristic analyses, including DeLong’s test for comparisons. Participants had a mean (± standard deviation) age at PET of 94 ± 3 years, and age at death of 97 ± 4 years. Most participants had amyloid at autopsy: 113 (68%) were amyloid beta positive, and 104 (63%) were neuritic plaque positive. The best performing SUVR was the PC2+WM, yielding an area under the curve (AUC) of 0.84 (95% CI 0.78–0.90) for amyloid beta and 0.82 (95% CI 0.76–0.89) for neuritic plaque, with an optimal cutoff of 0.77 for both outcomes. Reference and target region selection was more consequential for predicting amyloid beta positivity, with the WM consistently outperforming the cerebellar gray matter as a reference region for both target regions, while the PC2 significantly outperformed the cortical summary region for the cerebellar gray matter reference (P < 0.001) but only trended for significance for the WM reference (P = 0.069, DeLong’s test). Our findings suggest that using a WM reference and a limited cortical target region may improve PET-neuropathology correspondence, especially in older cohorts.

The online version contains supplementary material available at 10.1186/s40478-025-02198-3.