Okay, here’s a revised and fact-checked version of the provided text, incorporating current facts and addressing potential inaccuracies. I’ve focused on verifying claims, updating dates where appropriate, and providing more context. I’ve also added a section summarizing key takeaways.
Discordance in Amyloid PET Scans: A Growing Concern and the Role of AI
amyloid PET imaging is increasingly used in the diagnosis of alzheimer’s disease, but discrepancies between visual interpretation of scans and quantitative measurements of amyloid burden are a significant challenge. A recent analysis of the IDEAS (Imaging Dementia – Evidence for Amyloid Scanning) dataset reveals a concerning level of discordance. In IDEAS, 14 percent of scans showed disagreement between visual and quantitative assessments – more than double the rate observed in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the Longitudinal Early alzheimer’s Disease Study (LEADS). This suggests that such discordance may be more common in real-world clinical settings than previously thought, and highlights the need for improved scan interpretation methods.
Elizabeth Mormino of Stanford University underscored the importance of the IDEAS dataset,noting that it more accurately reflects how amyloid PET is performed in routine clinical practice compared to the highly controlled environments of ADNI and LEADS.She explained to Alzforum that ADNI and LEADS utilize specific imaging protocols with fixed acquisition windows and centralized quality control. In contrast,IDEAS data exhibits greater variability,with scans acquired across wider timeframes,using different scanners,ligands,imaging settings,and without central quality control. While the impact of these differences on quantitative measures is known, the effect on visual reads is less clear.
“it’s important to keep these differences in data quality in mind when interpreting discordant cases,” Mormino said. “Even though research datasets like ADNI and LEADS are valuable, they likely underestimate the prevalence of discordance we might encounter in a clinical setting.” rabinovici agrees, adding that the IDEAS population includes a higher proportion of individuals with intermediate amyloid levels (centiloid range), making classification more difficult.
The Promise of Artificial Intelligence
Could technology offer a solution? AI models are being developed to automate the interpretation of amyloid PET scans. At the Human Brain project’s High-level AI (HAI) conference, Xiao Yu, working in Jacob Vogel‘s laboratory at Lund University in Sweden, presented a model trained and tested on both ADNI and IDEAS datasets. Yu’s model achieved agreement with both human visual reads and centiloid quantification in 85 percent of cases. It showed discordance with one measure in 13 percent of cases, and disagreed with both in 2 percent.
Though, it remains unclear whether these models will ultimately resolve the discordance issue. A key challenge is the absence of a definitive “gold standard” to determine which interpretation method – visual or quantitative – is most accurate. Consequently, AI models are currently evaluated primarily on their ability to reproduce human visual reads.
Despite this limitation, machine learning is expected to improve quantitative methods and potentially standardize interpretation across different tracers and imaging centers. (see also Bourgeat et al., 2025). Recent work,like that by Bourgeat et al.,is focused on leveraging AI to refine centiloid measurements and improve consistency.
Paper Citation:
Bourgeat P, Fripp J, Lebrat L, Xia Y, Feizpour A, Cox T, Zisis G, Gillman A, Goyal M, Tosun D, Benzinger T, LaMontagne P, Breakspear M, Lupton M, Short C, adam R, Robertson J, Sperling R, O’Bryant S, Johnson S, Jack C Jr, Schwarz C, Park DC, Barkhof F, Farrar G, Bollack A, Collij L, Landau S, koeppe R, Morris J, Weiner M, Villemagne V, Masters C, Rowe C, Doré V. Revisiting Centiloids using AI. Res Sq. 2025 Jul 8. https://www.ncbi.nlm.nih.gov/pubmed/40671806
Key Takeaways:
* Discordance is a Problem: A significant percentage of amyloid PET scans show disagreement between visual interpretation and quantitative measurements, particularly in real-world
Keep reading