O’Bryant, Sid E.

Importance Amyloid positron emission tomography (PET) is increasingly used in research and clinical settings to determine the etiology of cognitive decline and eligibility for amyloid-targeting therapies. To assist with amyloid PET evaluation and to guide clinical decision-making, images can be quantified in a standardized unit called Centiloid, the interpretation of which can vary according to the method and threshold used. Objective To collect Centiloid values from available studies and determine robust positivity cutoffs using data-driven methods and correspondence with visual reads. Data Sources PubMed search (October 2024) identified studies with Centiloid values. Corresponding authors were invited to share individual participant data. Additional data were obtained through access-controlled repositories and conference outreach (July 2024-July 2025). Study Selection Studies were included if they provided Centiloids, radiotracer, age, and sex. Data Extraction and Synthesis Each study was analyzed using a unified statistical pipeline; study estimates were pooled using random-effects meta-analysis. Main Outcomes and Measures Gaussian mixture models (GMMs) were fitted to Centiloid values for each study. In studies with a bimodal distribution (per integrated completed likelihood), single cutoffs for positivity were set as mean plus 2 SDs of the lower gaussian component. Using GMMs, a double-cutoff approach defined a lower certainty range using a 90% posterior probability cutoff for assignment to the low (amyloid-negative) vs high (amyloid-positive) component. An alternative Centiloid cutoff was derived from maximizing the correspondence (Cohen κ) with the binary visual reads when available. Results This meta-analysis included cross-sectional amyloid PET scans acquired with 5 radiotracers from 49 227 participants across 53 studies from 15 countries (mean age, 71 years; 54% female, 62% cognitively impaired). The data-driven GMM approach identified a bimodal distribution in 51 studies (n = 48 786), resulting in a single cutoff for positivity of 18 Centiloids (95% CI,16-19; I 2 = 97%). The double-cutoff approach revealed high confidence for interpreting scans as negative when Centiloid values were lower than 11 (95% CI, 9-13; I 2 = 95%) and interpreting scans as positive if Centiloid values were higher than 26 (95% CI, 24-28; I 2 = 95%). In analyses of correspondence with binary (positive or negative) visual reads of amyloid PET scans (n = 35 045; 36 studies), Centiloids were highly predictive of visual positivity (Cohen κ, 0.86; 95% CI, 0.83-0.89; I 2 = 96%) with a cutoff of 27 Centiloids (95% CI, 24-30; I 2 = 80%). Conclusions and Relevance In this individual participant data meta-analysis, positivity cutoffs converged around 18 Centiloids (data-driven) and 27 Centiloids (visual reads). Findings from a double-cutoff analysis suggest that scans in the 11 to 26 Centiloid range should be interpreted with caution depending on the context of use.

Abstract INTRODUCTION The Centiloid scale is the standard for amyloid (Aβ) PET quantification in research and clinical settings. However, variability between tracers and scanners remains a challenge. METHODS This study introduces DeepSUVR, a deep learning method to correct Centiloid quantification, by penalizing implausible longitudinal trajectories during training. The model was trained using data from 2,129 participants (7,149 Aβ positron emission tomography [PET] scans) in the Australian Imaging, Biomarkers and Lifestyle Study of ageing (AIBL)/Alzheimer's Disease Neuroimaging Initiative (ADNI) and validated using 15,807 Aβ PET scans from 10,543 participants across 10 external datasets. RESULTS DeepSUVR increased correlation between tracers, and reduced variability in the Aß‐negatives. It showed significantly stronger association with cognition, visual reads, neuropathology, and increased longitudinal consistency between studies. DeepSUVR also increased the effect size for detecting small treatment related slowing of amyloid accumulation in the A4 study. DISCUSSION DeepSUVR substantially advances Aβ PET quantification, outperforming all standard approaches, which is particularly important for clinical decision making and to detect subtle or early changes in Aβ. Highlights Novel artificial intelligence (AI)‐method that penalizes biologically implausible longitudinal trajectories, enabling the model to learn standardized uptake value ratios (SUVR) correction factors without requiring longitudinal data at inference time. Improves Centiloid consistency across tracers and studies, significantly enhancing cross‐sectional and longitudinal amyloid positron emission tomography (PET) quantification. DeepSUVR‐derived Centiloids show stronger associations with cognition, visual reads, and neuropathology. Longitudinal variability is reduced three‐ to five‐fold, enabling more reliable tracking of amyloid accumulation and better detection of treatment effects. Novel reference and target masks derived from DeepSUVR replicate most of the model's performance, offering a practical alternative for integration into existing pipelines.

Abstract Background Ten years after the original publication, the Centiloid framework is now broadly used to harmonize amyloid‐PET quantification, facilitate data sharing and comparison across cohorts, and even assist visual interpretation in clinical settings. We evaluated the global implementation of Centiloids by comparing their distribution and corresponding positivity thresholds across cohorts. Methods We gathered data from publicly available cohorts and reached out to investigators across the world to collect cross‐sectional Centiloids, demographic and clinical information, and visual read data. Gaussian mixture models (GMM, k=2) were fitted to Centiloid values for each cohort and cutoffs were calculated as mean + 2SD of the lower Gaussian. When visual reads were available, we determined Centiloid cutoffs that maximized correspondence with visual reads (Cohen's kappa). Data was combined across cohorts using random effects meta‐analyses. Results As of January 2025, we included 37 cohorts ( n = 41,678 participants) with heterogeneous pipelines, radiotracers, and clinical and demographic characteristics (Table‐1). The low Gaussian peaks ranged from ‐9 to 10CL; the meta‐analysis identified a common peak at 1CL. The second peaks were more heterogeneous (range=38‐102CL; meta‐analysis outcome=64CL). Across cohorts, the proportion of cognitively unimpaired versus impaired participants impacted the position of both peaks, with better separation in cohorts enriched in impaired individuals (Figure 1C). The meta‐analysis indicated a GMM‐based cutoff of 19CL (95%CI: 16‐21CL, Figure 1B); subgroup analyses showed no evidence of significant effect between single versus multicenter settings (17 versus 20CL, p = 0.30), MRI‐based or PET‐only processing (18 versus 19CL, p = 0.88), and no evidence of difference across radiotracers (Flutemetamol: 16CL; PIB: 17CL, Flutafuranol: 18CL, Florbetaben: 19CL, Florbetapir: 20CL, p = 0.78). In a subset of 29,496 participants with visual reads available, binary visual reads corresponded well to Centiloids (common kappa=0.86, Figure 2A). The visual read‐based cutoff of 24CL (95%CI: 21‐27CL, Figure 2B) maximized correspondence between visual read and quantification and was slightly higher than the GMM‐based cutoff. All meta‐analysis models showed high non‐random heterogeneity (I 2 >80%) across studies, suggesting non‐random differences in peaks and cutoffs. Conclusions Meta‐analysis‐based cutoffs align well with thresholds from the existing literature. High heterogeneity among studies underscores the need to investigate contributing factors, raising concerns about applying common cutoffs.

Abstract Background Sex, education and race/ethnicity are all associated with risk of Alzheimer's disease dementia. Here, we assess the effects of self‐reported sex, educational attainment and race/ethnicity on amyloid‐positivity, and tau‐PET‐positivity in 12,048 (7,394 cognitively unimpaired [CU], 2,177 MCI, and 2,477 dementia) individuals from 42 cohorts worldwide. Method Logistic generalized estimating equations were used to estimate frequency of amyloid‐positivity (using cohort‐specific thresholds for amyloid‐PET [84%] or CSF) and tau‐PET‐positivity (cohort‐specific thresholds of 2SD above mean temporal uptake in amyloid‐negative controls). We assessed: i) sex and APOEε4 ( N = 10,098) associations, to complement earlier findings of a higher frequency of tau‐positivity in females, ii) effects of lower/higher education ( N = 10,970; cohort‐specific median‐split), and iii) effects of race/ethnicity (non‐Hispanic White [hereafter: White], N = 4880; Asian, N = 116; Black or African‐American [hereafter: Black], N = 353; Hispanic, N = 356, only from Northern‐American cohorts). Outcomes were frequency of amyloid‐positivity in CU individuals only, and tau‐PET‐positivity in both amyloid‐positive (AB+) CU and cognitively impaired (CI, i.e. MCI and dementia) individuals. Interaction effects on the relationship between age and amyloid/tau‐positivity were assessed and only retained in the models when significant. Result Female sex was associated with an APOEε4 ‐independent increased frequency of amyloid‐positivity (β=0.51[0.22], p = 0.02) in CU and increase of tau‐positivity in both AB+CU (β=0.27[0.08]) and AB+CI (β=0.37[0.08], both p <0.01). Remarkably, tau‐positivity frequencies of female APOEε4 non‐carriers were equivalent to male APOEε4 carriers in AB+CI (Figure 1). No significant sex* APOE interactions were observed. In CU, higher education was associated with lower amyloid‐positivity frequency (β=‐0.12[0.05], p = 0.02). In contrast, among AB+CU, there was an age*education interaction effect that indicated more pronounced age effects on tau‐positivity in individuals with higher education (age*education:β interaction =0.03[0.01], p <0.01). There were no education effects in AB+CI (Figure 2). In CU, an age*race/ethnicity interaction effect was observed across all non‐White groups compared to White (Hispanic:β interaction =‐0.05[0.01], p <0.01; Black:‐0.04[0.01], p <0.01; Asian:‐0.02[0.01], p = 0.04). This suggests that the impact of age on amyloid‐positivity was less pronounced in non‐White groups. Furthermore, in AB+CI, Hispanic ethnicity was related to higher tau‐positivity frequency than White (β=0.51[0.22], p = 0.02; Figure 3). Conclusion In this multi‐center initiative comprised of clinical and community‐based cohorts, we observed that self‐reported sex, educational attainment and race/ethnicity were related to positivity‐frequencies of Alzheimer's disease pathology.

Abstract Background Use of software to “de‐face” research brain images has grown as advances in automated face recognition have made it increasingly possible to re‐identify participants. Imaging studies of Alzheimer's Disease and related diseases have included more participants from historically underrepresented groups, but de‐facing software have primarily been validated using images from mostly non‐Hispanic White participants, and its performance has not been directly compared across ethnoracial groups. Method From The Health and Aging Brain Study: Health Disparities (HABS‐HD) study, we constructed a sample of 305 T1‐weighted MRI from HABS‐HD baseline visits, matched across three ethnoracial groups (non‐Hispanic White (NHW, n = 104), non‐Hispanic African American (AA, n = 104), Mexican American (MA, n = 97)) according to age‐decade, sex, cognitive status (impaired/unimpaired), and MRI scanner model. We ran mri_reface 0.3.5 on each image, and we ran FreeSurfer 7.4.1 on each unmodified image and on each de‐faced output. We then compared regional gray matter (GM) volume and cortical thickness measurements by subtracting each de‐faced image measurement from the corresponding unmodified image measurement. We compared these differences between de‐faced and unmodified images across the three matched groups using GLM analyses adjusted for age, sex, and cognitive status, and Bonferroni corrected. Result When aggregating across all brain regions, differences in cortical thickness measures due to de‐facing were significantly different between the NHW and MA groups, and between AA and MA, but not between NHW and AA (Figure 1). Differences between these group means were <0.005mm. For GM volume, differences were significant between AA and MA (magnitude < 0.2%), but not between NHW and AA or NHW and MA (Figure 2). Examining each region individually (Figure 3), cortical thickness differences were significant for only 4 regions (all with magnitudes <0.03mm), and GM volume differences were significant for only 6 regions (all with magnitudes <2%). Conclusion Effects of mri_reface on FreeSurfer brain measurements were significantly different across ethnoracial groups. Our future work will optimize mri_reface to eliminate these differences. However, the magnitudes of these differences are sufficiently small that they are unlikely to have a practical effect on analyses of data using current versions.

Abstract Background Ten years after the original publication, the Centiloid framework is now broadly used to harmonize amyloid‐PET quantification, facilitate data sharing and comparison across cohorts, and even assist visual interpretation in clinical settings. We evaluated the global implementation of Centiloids by comparing their distribution and corresponding positivity thresholds across cohorts. Methods We gathered data from publicly available cohorts and reached out to investigators across the world to collect cross‐sectional Centiloids, demographic and clinical information, and visual read data. Gaussian mixture models (GMM, k=2) were fitted to Centiloid values for each cohort and cutoffs were calculated as mean + 2SD of the lower Gaussian. When visual reads were available, we determined Centiloid cutoffs that maximized correspondence with visual reads (Cohen’s kappa). Data was combined across cohorts using random effects meta‐analyses. Results As of January 2025, we included 37 cohorts ( n = 41,678 participants) with heterogeneous pipelines, radiotracers, and clinical and demographic characteristics (Table‐1). The low Gaussian peaks ranged from ‐9 to 10CL; the meta‐analysis identified a common peak at 1CL. The second peaks were more heterogeneous (range=38‐102CL; meta‐analysis outcome=64CL). Across cohorts, the proportion of cognitively unimpaired versus impaired participants impacted the position of both peaks, with better separation in cohorts enriched in impaired individuals (Figure‐1C). The meta‐analysis indicated a GMM‐based cutoff of 19CL (95%CI: 16‐21CL, Figure‐1B); subgroup analyses showed no evidence of significant effect between single versus multicenter settings (17 versus 20CL, p = 0.30), MRI‐based or PET‐only processing (18 versus 19CL, p = 0.88), and no evidence of difference across radiotracers (Flutemetamol: 16CL; PIB: 17CL, Flutafuranol: 18CL, Florbetaben: 19CL, Florbetapir: 20CL, p = 0.78). In a subset of 29,496 participants with visual reads available, binary visual reads corresponded well to Centiloids (common kappa=0.86, Figure‐2A). The visual read‐based cutoff of 24CL (95%CI: 21‐27CL, Figure‐2B) maximized correspondence between visual read and quantification and was slightly higher than the GMM‐based cutoff. All meta‐analysis models showed high non‐random heterogeneity (I 2 >80%) across studies, suggesting non‐random differences in peaks and cutoffs. Conclusions Meta‐analysis‐based cutoffs align well with thresholds from the existing literature. High heterogeneity among studies underscores the need to investigate contributing factors, raising concerns about applying common cutoffs.

Abstract Background Use of software to “de‐face” research brain images has grown as advances in automated face recognition have made it increasingly possible to re‐identify participants. Imaging studies of Alzheimer’s Disease and related diseases have included more participants from historically underrepresented groups, but de‐facing software have primarily been validated using images from mostly non‐Hispanic White participants, and its performance has not been directly compared across ethnoracial groups. Method From The Health and Aging Brain Study: Health Disparities (HABS‐HD) study, we constructed a sample of 305 T1‐weighted MRI from HABS‐HD baseline visits, matched across three ethnoracial groups (non‐Hispanic White (NHW, n = 104), non‐Hispanic African American (AA, n = 104), Mexican American (MA, n = 97)) according to age‐decade, sex, cognitive status (impaired/unimpaired), and MRI scanner model. We ran mri_reface 0.3.5 on each image, and we ran FreeSurfer 7.4.1 on each unmodified image and on each de‐faced output. We then compared regional gray matter (GM) volume and cortical thickness measurements by subtracting each de‐faced image measurement from the corresponding unmodified image measurement. We compared these differences between de‐faced and unmodified images across the three matched groups using GLM analyses adjusted for age, sex, and cognitive status, and Bonferroni corrected. Result When aggregating across all brain regions, differences in cortical thickness measures due to de‐facing were significantly different between the NHW and MA groups, and between AA and MA, but not between NHW and AA (Figure 1). Differences between these group means were <0.005mm. For GM volume, differences were significant between AA and MA (magnitude < 0.2%), but not between NHW and AA or NHW and MA (Figure 2). Examining each region individually (Figure 3), cortical thickness differences were significant for only 4 regions (all with magnitudes <0.03mm), and GM volume differences were significant for only 6 regions (all with magnitudes <2%). Conclusion Effects of mri_reface on FreeSurfer brain measurements were significantly different across ethnoracial groups. Our future work will optimize mri_reface to eliminate these differences. However, the magnitudes of these differences are sufficiently small that they are unlikely to have a practical effect on analyses of data using current versions.

Abstract The Centiloid scale is the standard for Amyloid (Aβ) PET quantification, widely used in research, clinical settings, and trial stratification. However, variability between tracers and scanners remains a challenge. This study introduces DeepSUVR, a deep learning method to correct Centiloid quantification, by penalising implausible longitudinal trajectories during training. The model was trained using data from 2,098 participants (6,762 Aβ PET scans) in AIBL/ADNI and validated using 15,806 Aβ PET scans from 10,543 participants across 10 external datasets. DeepSUVR increased correlation between tracers, and reduced variability in the Aβ-negatives. It showed the strongest association with cognition, highest AUC against visual reads and best longitudinal consistency between studies. DeepSUVR also increased the effect size for detecting lower Centiloid increase per year in the A4 study. DeepSUVR advances Aβ PET quantification, outperforming standard approaches, which is particularly important for consistent decision making and to detect subtle and early changes in clinical interventions.