Concordance between amyloid-PET quantification and real-world visual reads: results from IDEAS

Amyloid-PET detects fibrillar β-amyloid deposits, a defining feature of Alzheimer’s disease. This technology has been used in research for over 20 years, and is now used in clinical practice to guide patient diagnosis and management. However, the real-world use of amyloid-PET may differ from research settings due to less standardized acquisition protocols, less experienced visual readers, and patients with more comorbidities, potentially compromising test accuracy. We evaluated the performance of amyloid-PET as used in a real-world clinical setting utilizing data collected via the Imaging Dementia—Evidence for Amyloid Scanning (IDEAS) study.

The study collected clinical amyloid-PET scans of older adults with cognitive decline acquired in 294 imaging facilities using three FDA-approved radiotracers. We centrally processed these scans using a PET-only processing pipeline and derived summary quantification of cerebral radiotracer retention measured on the Centiloid scale. We applied an a priori autopsy-based threshold of 24.4 Centiloids to quantitatively define scan positivity and compared this quantitative classification with binary visual reads performed by local radiologists or nuclear medicine physicians.

10,350/10,774 scans (96%) passed quality control and were successfully quantified. Median patient age was 75 (interquartile range 71, 80) years, 51% were females, 87% self-identified as White, 63% had mild cognitive impairment (vs. 37% with dementia), and 61% of scans were visually positive based on local reads. Centiloid values were higher in visually positive scans (median=74 [46, 99]) compared to scans locally read as negative (−2 [−13, 12]; p<0.001). Patients with dementia had higher Centiloids than those with mild cognitive impairment (57 [8, 91] vs. 34 [−2, 79]; p<0.001), consistent with a higher visual positivity rate (70% vs. 56%, respectively; p<0.001). Agreement between local visual reads and quantitative classification was 86.3% (95%CI: 85.7%, 87.0%), corresponding to Cohen’s Kappa of 0.715 (95%CI: 0.701, 0.729; p<0.001). Overall, 5,519 [53% of total] scans were positive by both visual read and quantification (V+/Q+); 3,416 [33%] were negative by both (V-/Q-), 813 [8%] were V+/Q-, and 602 [6%] were V-/Q+. Female sex, White race, and use of the radiotracers ¹⁸ F-flutemetamol and ¹⁸ F-florbetaben (compared to ¹⁸ F-florbetapir) were associated with higher visual-quantitative concordance.

In conclusion, local visual reads showed high concordance with central quantification of clinical amyloid-PET scans, supporting the validity of amyloid-PET as used in the real world and the use of quantification in clinical settings.