Citations

Abstract Background Brain age models are increasingly being used as measures of brain health and neurodegeneration. These models can be deep learning (DL) based and/or classical methods and provide a brain age gap (BAG) which is the difference between chronological and predicted brain age. Our goal was to develop DL‐based scanner invariant brain age models that can be applied in aging and dementia studies even when there are scanner changes that may hinder longitudinal follow‐up. Method We identified 3374 cognitively unimpaired (CU) participants with T1 scans from the Mayo Clinic Study of Aging (MCSA). A subset of 863 CU amyloidosis negative (PiB‐PET cutoff < 25 Centiloid) and no history of diabetes and hypertension was selected (CU‐noVAβ). We built two models – Model‐A trained on CU‐noVAβ, and Model‐B trained on all CU. We included 453 MCSA MCI participants as test set for both models. We used 113 participants (74 CU, 28 MCI, 10 dementia) from MCSA and Mayo ADRC with scans from two different vendors (scanned within a week of each other) for testing cross‐vendor/scanner differences (cross‐vendor data). The two unique features of our model included histogram matching and scanner type as an input to the DenseNet model for training (Figure 1). Result In the data, the inclusion of scanner as an input decreased the prediction of mean age differences between the scanners (Model‐A=2.17 years; Model‐B=1.71 years) (Figure 2). The mean absolute error (MAE=absolute value of BAG) for Model‐A in MCI was 5.72 years and Model‐B in MCI was 4.06 years, consistent across scanners (Figure 3). Longitudinal prediction increased with age (Model‐A slope=0.53; Model‐B slope=0.63). Conclusion We found that training on all CU (Model B) rather than a subset CU‐noVAβ (Model A) allowed for the DenseNet model to backpropagate on more variable data which lowered MAE in prediction error between scanners and lowered MAE for MCIs. Histogram matching and scanner as an input to the brain age model accounted for scanner changes and resulted in scanner invariant longitudinal brain age estimation.

Abstract Background The objective of this study was to determine the predictive value of amyloid‐PET versus plasma %p‐tau217 (defined as the ratio of phosphorylated to non‐phosphorylated Tau217 × 100), Aβ42/40, and the amyloid probability score (APS2; which combines of Aβ42/40 and %p‐tau217) in tau‐PET transitions (T‐ to T+). Method Participants included were from the Mayo Clinic Study of Aging (MCSA) and had plasma markers measured by mass spectrometry (IP‐MS; C2N), had an amyloid‐PET scan, and had a tau‐PET negative scan (SUVR<1.29, temporal meta‐ROI) at the baseline, along with one or more follow‐up tau‐PET scans. Cox proportional hazard models were used to evaluate the relationship between C2N plasma %P‐Tau217 and APS2 and incident tau‐PET positivity (SUVR>=1.29) adjusting for age, sex, and APOE ε4. An analysis was also performed using amyloid‐PET centiloid as the predictor. Both the plasma markers and amyloid‐PET centiloid were scaled so that a one‐unit change corresponds to the respective interquartile range (IQR) for comparability of hazard ratios (HR). The analyses were repeated in the BioFINDER‐2 cohort for validation. Result 255 MCSA participants (122 female [47.8%] were included with a median age of 71.9 (IQR: 65.8, 78.8) years old, of which 37 progressed to tau‐PET positivity. At baseline, 238 (93.7%) were cognitively unimpaired. Higher baseline %p‐tau217 (HR: 1.52 [95% CI: 1.28, 1.8]), amyloid‐PET centiloid (HR: 1.47 [95% CI: 1.20, 1.79]), and APS2 (HR: 1.62 [95% CI: 1.22, 2.16]) were associated with an increased risk of progressing to tau‐PET positive (Figure 1). Aβ42/40 (HR: 1.06 [95% 0.604–1.867] was not associated with incident tau positivity Among 605 participants who were tau negative at index date in BioFINDER‐2 (316 (52.2%) cognitively unimpaired), with a median age of 70.2 (IQR: 60.5‐76.4), 33 progressed to tau positivity with higher %p‐tau217 (HR: 1.8 [95% CI: 1.5, 2.17]), amyloid‐PET centiloid (HR: 2.3 [95% CI: 1.77, 2.97]), and lower Aβ 42/40 (HR: 0.42 [95% CI: 0.21, 0.85]) at index date all associated with an increased risk of progressing to tau‐PET positive. Conclusion In two independent cohorts , %p‐tau217 provides predictive information about those who will convert from tau negative to tau positive on PET scan with the associations being similar to amyloid‐PET.

Abstract Background Some data suggest that neuroinflammation may be associated with pro‐inflammatory states in the body. Systemic, proinflammatory cytokines (TNFα, IL‐1, IL‐6, etc.) are produced by a variety of comorbidities. Neuroinflammation has been shown to be associated with the accumulation of amyloid in the brain (Becher, Heneka). We hypothesized that systemic cytokine production could be associated with neuroinflammation and thereby be related to brain amyloid accumulation. This study aims to evaluate the relationship between neuroinflammation, amyloid accumulation and systemic cytokines. Method 90 participants from the Mayo Clinic Study of Aging (MCSA) received a fasting cytokine panel over a 5‐year span where IL‐1, IL‐6, IL‐10, TNFa levels were measured. Participants were divided into four groups: cognitively unimpaired with normal levels of amyloid (CU A‐, n = 25), cognitively unimpaired with elevated amyloid (CU A+, n = 25), and mild cognitively impaired (MCI) with amyloid positive (MCI A+, n = 25), and Alzheimer's disease (AD) with elevated amyloid (AD A+, n = 15). To compare regional standardized uptake value ratio (SUVr) from neuroinflammation PET based on amyloid status (A+ vs A‐) and the four cytokines, Spearman's rank correlation and Fishers r‐to‐z transformations were used to assess pairwise group differences. Result Using Spearman's rank correlation and Fishers r‐to‐z transformations, no strong relationship was found between systemic inflammation and neuroinflammation. While comparing CUA‐ vs CUA+ in Figure 1, three regions (supplementary motor area, frontal superior medial lobe, and middle cingulate cortex) showed a significant difference ( p < 0.05). However, Figure 2 showed an overarching trend bias towards significance across most regions in IL10, IL6, and TNFa. Conclusion Although significant relationships between systemic inflammation, brain amyloid and neuroinflammation were limited, the data show an overarching trend bias in broad brain regions, suggesting that further analyses are needed, possibly with larger sample sizes, to assess the relationship of regional neuroinflammation and amyloid to systemic inflammation. Future analyses could also regress out factors like sex, age, and genotype to further proinflammatory cytokine research.

Abstract Background Cognitive resilience is the discrepancy in individuals with intermediate to high Alzheimer's disease neuropathologic change (ADNC) and normal cognition. Currently, the role of microglial activation in cognitive resilience is understudied. Method In the MCSA brain bank, high‐cognitively unimpaired (high‐CU) participants were defined as those performing in the upper quartile of their respective age groups based upon antemortem global cognition z‐scores, computed from memory, language, executive function, and visuospatial domain z‐scores. From the high‐CU group ( N = 118, global z‐scores from 1.55 to 1.78), cognitively resilient (CR) individuals were identified as those with intermediate to high ADNC upon neuropathologic evaluation. First, regression models were fit to evaluate global cognition as a function of age, sex, amyloid PET, and microglial‐specific genes of interest, including MS4A gene cluster SNPs. Second, a subset of CR participants was analyzed for postmortem ARM expression in the dorsolateral prefrontal cortex. Single immunohistochemistry was performed on formalin‐fixed, paraffin‐embedded sections using antibodies against beta‐amyloid (6F3D) and CD163 (10D6). Digital image analysis was performed on scanned sections (40x, Leica Aperio GT 450), and ARM (CD163%area) and beta‐amyloid (%area) measurements were obtained using QuPath. Cognitively impaired (CI) control group was matched on age, sex, ADNC level, and APOE4 status. Spearman correlations were performed between postmortem ARM and antemortem MRI, PET, plasma, and global cognition metrics. Result A third (39/118) of high‐CU cases showed intermediate/high ADNC (CR group). Furthermore, 28% of CR individuals harbored an MS4A6A SNP, which showed a protective effect on cognition via amyloid‐PET ( p = 0.043, Figure 1). Moreover, CR individuals harbored significantly higher ARM:amyloid than the CI group ( p <0.001, Figure 2) and fewer neuritic plaques ( p = 0.03, Figure 3). Antemortem white matter hyperintensities and global cognition were significantly correlated to ARM expression (r=0.556 [p=0.042] and r=0.648 [p=0.02], respectively). Conclusion Microglial specific MS4A6A SNP, which modulates microglial activation, was seen in a third of CR individuals. Postmortem analyses in CR individuals revealed increased ARM:amyloid, which was also associated with higher antemortem cognitive performance. These findings underscore the role of microglial activation in cognitive resilience. Future work is needed to investigate microglial protective mechanisms and links between vascular disease and microglial activation.

Abstract Background Neuropathologic evaluation is essential for diagnosing and treating dementia, but definitive diagnoses is only available after post‐mortem brain autopsy. Positron emission tomography (PET) offers a potential in vivo proxy for predicting neuropathology. This study proposes a semi‐supervised learning (SSL) model using a single FDG‐PET image to predict underlying neuropathology. Method The SSL model leverages a self‐organized graph structure based on a patient similarity network (PSN) created from 1,495 FDG‐PET images across the aging and neurodegenerative spectrum. Each patient is represented by a node, with edges indicating image similarity derived from principal component analysis features and cosine similarity weights. The PSN structure connects 20 nearest neighbors. Neuropathologic labels were available for 204 autopsied patients. The Poisson Learning Method was employed to propagate labels from these annotated nodes to unlabeled nodes, guided by the PSN structure. FDG‐PET images were collected from clinical and research participants at the Mayo Clinic and processed using an MRI‐free pipeline. Neuropathologic labels were assigned based on widely accepted diagnostic criteria. Leave‐one‐out cross‐validation (LOOCV) was used to assess model performance. Results The confusion matrix demonstrated high accuracy in predicting Alzheimer's disease, frontotemporal lobar degeneration with tau inclusions, and Lewy body disease pathology. A Sankey diagram and label distribution analysis revealed strong correlations between clinical and predicted neuropathologic labels. Conclusion This study highlights the potential of a graph‐based SSL pipeline to predict neuropathological changes using FDG‐PET imaging data, despite limited pathology labels. The findings underscore the value of SSL in bridging the gap between abundant neuroimaging data and scarce neuropathologic labels, contributing to improved dementia diagnosis and understanding.

Abstract Background Alzheimer’s Disease (AD) progression is conceptualized as a cascade of sigmoidal (S‐shaped) curves, with biomarkers shifting sequentially from normal to abnormal states. Estimating these curves requires sophisticated modeling, with challenges including (1) selecting appropriate sigmoidal‐type functions, (2) achieving algorithmic convergence, and (3) producing clinically useful parameters. We compare commonly used sigmoidal‐functions and effects on novel estimates of early and late AD‐biomarker progression ages. Method Using 3,999 Amyloid PET (PiB SUVR) measurements from 1,367 Mayo Clinic Study of Aging participants, we compared model fits (AIC, BIC) between logistic and hyperbolic functions using Non‐Linear Mixed Effect (NLME) models with subject‐specific age‐shifts. We defined novel “initial” and “maximum” acceleration points (IAP and MAP) to compare clinically “earlier” and “later” AD‐progression ages between sigmoidal‐curves. The MAP uses the maximum second‐derivative point, reflecting the age of fastest acceleration in biomarker accumulation. The IAP is defined as the first point where the second derivative of the sigmoidal‐curve exceeds a clinical threshold (here, 5% velocity increase per year), indicating onset age of early accelerated biomarker accumulation. Estimates for C 2 N plasma p ‐tau217 were also attempted (1,306 measurements on 457 participants). Models adjusted for sex and APOE4‐positivity. Result In the larger PET sample, the hyperbola had lower AIC/BIC (better fit), but showed a decreasing early PiB‐SUVR trend (Table 1A, Figure 1). Ages of late MAP were comparable, as were sex and APOE4+ related PiB‐SUVR‐curve shifts. Age of early AD‐onset was estimated 1.5 years earlier for logistic (IAP=63.6) versus hyperbola (IAP=65.1) functions. The logistic function always converged, while the hyperbolic function did not converge for p ‐tau217 in the smaller C 2 N sample. Early‐onset and late‐progression ages were 7‐8 years later for p ‐tau217 versus PiB‐SUVR using logistic‐curves. Conclusion Late AD‐progression and covariate‐related shifts in progression were comparable across models, suggesting robustness to sigmoid choice. In contrast, estimates of early AD‐onset may be more dependent on functional form. While logistic‐sigmoids inherently increase monotonically, hyperbolas do not, here estimating a decline in initial amyloid trend. Future work may explore adding biologically informed constraints to flexible sigmoidal‐functions, and alternative Bayesian approaches, to optimize both model fit and estimation of early AD‐onset age.

Abstract Background Multi‐site aging and dementia studies are challenged by data heterogeneity caused by scanner (manufacturer and protocol) related variability. Quantitative analysis of white matter hyperintensity (WMH) from pooled samples, a traditional marker of cerebrovascular disease, is sensitive to machine variability. Our goal was to compare Deming (symmetric) linear regression and ComBat, a common method used for harmonizing neuroimaging measurements, for harmonizing WMH measurements on participants scanned on two different scanners (GE and Siemens) and protocols (2D FLAIR on GE and 3D FLAIR on Siemens). Methods Cognitively unimpaired (CU) ( n =53, age (mean (SD))=67 (15) years) and cognitively impaired (CI; MCI/dementia) ( n =34, age (mean (SD))=82 (9) years) participants from the Mayo Clinic Study of Aging and Mayo ADRC were scanned on GE and Siemens scanners within a week of each other (typically done on the same day). WMH as a percentage of total intracranial volume was measured using multispectral segmentation of FLAIR and T1‐weighted images (Gunter et. al. AAIC 2023). We used Deming regression as it includes errors in both input dimensions. ComBat was applied using age, sex, and diagnosis as covariates, with scanner vendor and protocol as the batch variable. We assessed the inter‐scanner similarity using Maximum Mean Discrepancy (MMD) analysis, where a value of zero indicates identical distributions, p >0.05 indicates no significant difference under H 0 . (Gretton et al., JMLR 2012). Results Our results showed that the distribution of WMH measurements, which were different in the CU group, were more similar after Deming than ComBat harmonization (Figure 1). This was confirmed by the lower MMD in the CU group from Deming regression (‐0.01 ( p =0.76) from 0.03 ( p =0.05)) compared to ComBat (0.04 ( p =0.02))(Figure 1). However, ComBat and Deming were unable to create distributions as compatible as original measurements for the CI group. Conclusion Using a unique dataset of the same participants scanned close in time on different scanner and protocols, we found that (1) MMD measurements were useful to test the need/improvement due to harmonization; (2) ComBat was not substantially better at creating compatible distributions between scanner vendors than Deming regression, and both were sometimes inadequate. Future work will require alternative models to linear scale and shift transformations.

Abstract Background Amyloid related imaging abnormality edema (ARIA‐E) occurs in about 19% of individuals with autosomal dominant Alzheimer disease (ADAD) treated with an anti‐amyloid‐b monoclonal antibody (Salloway et al., 2021). In a previously reported case, ARIA‐E appeared to colocalize with decreases in PiB‐PET uptake (Joseph‐Mathurin, Llibre‐Guerra, et al, 2022), suggesting an association between amyloid‐b (Aβ) removal and ARIA‐E. Here, we compare longitudinal neuroimaging and corresponding neuropathology in an ADAD individual treated with gantenerumab, who experienced multiple ARIA‐E episodes as sulcal effusions that resolved by the end of the four‐year trial. Methods PiB‐PET and MRI were collected pre‐ and post‐randomization (2 and 4 years). The participant consented to brain donation, which was received a year after trial completion. Four tissue samples were taken from the location of the ARIA‐E in the parieto‐occipital left hemibrain as identified by MRI (one block contained the sulcus affected by ARIA‐E; the rest of the coronal section was captured in the remaining three blocks). Area fractions (AFs) of Aβ (10D5 immunohistochemistry (IHC)), tauopathy (PHF1), microglia (Iba1), and astrocytes (GFAP) in one sulcal region of interest (ROI) from each block and PiB‐PET SUVRs of corresponding ROIs were extracted ( n = 4, Figure 1). Additionally, we included PiB‐PET SUVRs corresponding to ARIA‐E findings observed in the right hemisphere at baseline, two‐year, and four‐year visits ( n = 6). Results Overall PiB‐PET uptake increased during the first two years (before ARIA‐E) and decreased during the last two years (including ARIA‐E episodes) (0.03±0.05 vs. ‐0.04±0.04 SUVR/year, p ‐value=0.01, Figure 2). The decrease seemed more pronounced in ARIA‐E ROIs versus normal‐appearing sulcus ROIs (‐0.05±0.04 vs. ‐0.02±0.03 SUVR/year). Lower PiB uptake at last visit appeared associated with lower Aβ AF (estimates=0.03±0.01, p ‐value=0.07). The ARIA‐E ROI had an Aβ AF of 0.009 while normal‐appearing sulcus ROIs had a mean of 0.02+0.003 (Figure 3). AFs for tauopathy, microglia, and astrocytes were within the range of those of normal‐appearing sulcus ROIs, suggesting an Aβ‐specific effect, although unexamined markers may play a role. Conclusions Our preliminary findings indicated that ARIA‐E is associated with longitudinal PiB‐PET decrease and Aβ AF, supporting the link between ARIA‐E, changes in PiB‐PET, and local Aβ removal observed at autopsy. Funding : K01AG080123; U01AG042791; R01AG046179; R01AG053267

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 Diffusion MRI (dMRI) measures are increasingly being used as biomarkers for aging and dementia in multi‐center studies. Measurements in longitudinal studies are often acquired on different systems, which can result in systematic differences of machine origin (e.g. vendor, hardware, algorithms). This may warrant harmonization prior to pooling multi‐site data. However, harmonization algorithms may introduce a risk of removing valuable information content and reducing sensitivity to physiological factors (e.g. sex, age, disease). Here, we tested the hypothesis that protocol standardization may minimize machine‐origin differences, obviating a need for post‐acquisition harmonization in a two‐site (Mississippi‐MS and Minnesota‐MN) cohort of the longitudinal Mayo Clinic Study of Aging (MCSA). Methods Two data sets were analyzed: first, a crossover dataset with 80 participants (scanned within 2 days) bridging two different protocols on GE and Siemens scanners; second, participants scanned using standardized protocols from the ongoing MCSA ( N MN‐MCSA =204; N MS‐MCSA =52). Parameters are listed in Figure 1. For crossover data, analyses were repeated with GE data resampled to 2‐mm isotropic resolution, and discarding Siemens‐only b‐shells to artificially reduce protocol disparities. Analysis encompassed visualization of overlain subjectwise histogram‐derived densities obtained from whole‐brain masks for dMRI measures; and median dMRI measures from within the genu of the corpus callosum (GCC) with nonparametric statistical testing. When available, both DTI and NODDI measures were compared. Results Pronounced visual distributional differences seen between MD and FA in crossover data were reduced with resampling of GE data and including only b‐shells shared across acquisitions. For the resampled/reshelled data, significant differences remained for all GCC‐FA comparisons, but not all GCC‐MD (Figure 2). For the ongoing MCSA data from two sites, MD and FA whole‐brain densities were more closely aligned, with GCC‐FA and GCC‐MD showing no significant differences (Figure 3). NODDI measures showed more discordant densities than DTI measures, including shifts in most probable values and differences in distributional arrangement. Conclusions Our work provides further evidence supporting protocol and processing standardization for minimizing machine‐origin differences prior to considering harmonization. Differences in dMRI metrics seen when comparing across unstandardized protocols were reduced when protocols were standardized. NODDI metrics had more apparent sensitivity to site than DTI measures.