Jack, Clifford R.

AbstractBackgroundChronic medical conditions increase dementia risk, and the mechanism may be captured computationally by measures of accelerated brain aging. We hypothesized that quantifying the impact of chronic conditions on GM and WM aging will aid in measuring the differential impact of each condition on brain aging and shed light on the mechanisms through which they increase dementia risk. We specifically focused on microstructural brain changes with Neurite Orientation Dispersion and Density Imaging (NODDI) based on advanced diffusion MRI ‐ neurite density index (NDI), free water fraction (FWF), and orientation dispersion index (ODI).MethodWe identified 1123 participants [913 cognitively unimpaired (CU), 114 with MCI, 96 with AD] from the Mayo Clinic Study of Aging (MCSA) and Mayo ADRC with multi‐shell diffusion MRI. MCSA participants had chronic medical conditions’ diagnosis (based on US Dept of HHS) extracted from the electronic health care records. Using the BorutaSHAP algorithm on NDI, ODI, and ISOVF features from GM and WM separately, we constructed two distinct brain age models for GM and WM. First, we evaluated the usefulness of the predicted age gap (PAG), which is the difference between predicted age by model and chronological age, for both WM and GM NODDI models in CU/MCI/AD. We then developed separate GM and WM models for each condition using samples from condition‐negative participants to compare PAG of GM and WM for that condition.ResultThere was greater PAG in GM (avg = ) than WM (avg = 9.15 yrs) in MCI/AD participants (Fig. 1A). Each condition shifted the GM PAG by 0.10 to 3.56 years and WM PAG by 0.38 to 3.72 years with stroke and congestive heart failure showing the greater PAG overall for GM and WM PAD respectively (Fig. 2). COPD, stroke, chronic kidney disease, hypertension had comparatively greater GM PAG than WM PAG; while hyperlipidemia, congestive heart failure and coronary artery disease had greater WM PAG than GM PAG (Fig. 1B).ConclusionChronic conditions are associated with increased GM and WM Brain aging models can aid in quantifying the impact of chronic conditions on dementia risk and understanding mechanisms.

AbstractBackgroundThe TAR DNA‐binding protein of 43 kDa (TDP‐43) is linked to hippocampal volume loss and faster rates of hippocampal atrophy in Alzheimer's disease and primary age‐related tauopathy (PART). Hence, TDP‐43 is becoming an important player in age‐related neurodegeneration. To advance our understanding of TDP‐43 effect on hippocampus, we conducted an imaging‐pathological study to determine which specific hippocampal subfields are affected by TDP‐43 in cases of PART.MethodOne‐hundred fifteen autopsied cases from Mayo Clinic Alzheimer Disease Center, Study of Aging and Neurodegenerative Research Group were analyzed. All had completed antemortem brain MRIs, underwent neuropathological examination per NIA and Alzheimer’s Association (NIA‐AA) criteria and had additional assessment for TDP‐43 deposition. Only cases meeting criteria for PART (Thal phases 0‐2 and Braak NFT stages I‐IV) were included. Hippocampal subfield segmentation was performed using FreeSurfer‐7. Brain volumes were normalized to total intracranial volume (TIV) when comparing TDP‐43 groups. Statistical analyses including logistic regression analyses were utilized to explore associations between TDP‐43 status and hippocampal subfield volumes.ResultForty‐nine of the 115 cases (43%) with PART were female. Thirty‐seven (32%) cases were TDP‐positive. Mean age at last MRI was 84 years. There were no differences in pathological characteristics between both groups except for presence of hippocampal sclerosis which was more frequent in the TDP(+) group (P <0.001). Whole hippocampal volume was significantly smaller in the TDP(+) group (176 ±31 vs. 188 ±27, p=0.049). After stratification of hippocampus into head, body, and tail, TDP‐positive cases had smaller volumes of hippocampal body (p=0.031) and tail (p=0.021) but did not reach significance for hippocampal head (p=0.129). When regions were subdivided into hippocampal subfields, TDP(+) cases showed smaller volumes of molecular layer of the hippocampal body (p=0.046), CA1 body (p=0.031), subiculum body (p=0.023), and presubiculum body (p=0.046) compared to TDP(‐) cases. Logistic regression analysis adjusting for TIV confirmed these associations.ConclusionThis study underscores the association between TDP‐43 and the hippocampus and provides further evidence that TDP‐43 targets multiple regions and subfields of the hippocampus in patients with PART. Intriguingly, regions targeted are those where TDP‐43 deposition is microscopically observed, suggesting a direct “hit” by TDP‐43 deposition.

Abstract Background Many proposed clinical decision support systems (CDSS) require multiple disparate data elements as input, which makes implementation difficult, and furthermore have a black‐box nature leading to low interpretability. Fluorodeoxyglucose Positron Emission Tomography (FDG‐PET) is an established modality for the diagnosis of dementia, and a CDSS that uses only an FDG‐PET image to produce a reliable and understandable result would ease both of these challenges to clinical application. Method A deep variational autoencoder (VAE) was used to extract a latent representation of each image through prior training from FDG‐PET brain images (n=2000). This unsupervised VAE has a novel graph convolutional architecture that makes it applicable to masked template space images. A logistic regression model was used to classify each image. A parametric study of the latent space revealed the imaging features that were used in the logistic regression model to differentiate each class. A separate dataset of participants labeled with their clinical diagnosis (n=1239) was used to assess the logistic regression model’s diagnostic accuracy of cognitively unimpaired (CU) (n=679) and between common dementia subtypes: Alzheimer’s disease (AD) (n=310), Lewy body dementia (DLB) (n=151), and behavioral variant frontotemporal dementia (bvFTD) (n=99). The logistic regression classifier was evaluated using receiver‐operator characteristic (ROC‐AUC) curves. Result ROC‐AUC curves for CU and each dementia subtype are illustrated in Figure 1a. A k nearest neighbors model was used to develop a graphical representation of the latent space where nodes are images and edges are drawn between nearest neighbors, illustrated in Figure 1b. The graph of the latent space demonstrates a comparison between dementia subtypes and CU images. Conclusion In this study, we developed a proof‐of‐concept VAE that can achieve differential diagnosis with 84.8% balanced accuracy. The model architecture provides visual interpretability, and shows that the hippocampi, parietal lobes, and cerebellum, were useful for distinguishing AD, the frontal lobe, posterior cingulate cortex, and hippocampi were useful for distinguishing DLB, and prefrontal pole, subcortical structures, and occipital lobe were useful for distinguishing bvFTD. These regions of metabolism align with clinical pathology and metabolic patterns.

AbstractBackgroundRecent advances in automatic face recognition have increased the risk that de‐identified research imaging data could be re‐identified from face imagery in brain scans.MethodAn ADNI committee of independent imaging experts evaluated 11 published techniques for face‐deidentification (“de‐facing”) and selected four algorithms (FSL‐UK Biobank, HCP/XNAT, mri_reface, and BIC) for formal testing using 183 longitudinal scans of 61 racially and ethnically diverse ADNI participants, evaluated by their facial feature removal on 3D rendered surfaces (confirming sufficient privacy protection) and by comparing measurements from ADNI routine image analyses on unmodified vs. de‐faced images (confirming negligible side effects on analyses).ResultEffects of the de‐facing methods on FreeSurfer at UCSF (Figure 1‐top), ASHS at University of Pennsylvania (Figure 1 bottom‐left), and longitudinal atrophy at Mayo Clinic (Figure 1 bottom‐right) were comparable with each other and sufficiently small vs. unmodified scans to recommend de‐facing in ADNI‐4. The FSL and mri_reface methods were more thorough in removing facial features, partly because these also removed ears (Figure 2‐top). mri_reface was recommended for ADNI‐4 primarily because fewer (zero) images had modified brain (Figure 2‐bottom), and ADNI leadership approved de‐facing with mri_reface. ADNI performed a second validation of its latest version on another diverse sample of 100 ADNI‐3 participants. These analyses included: boundary shift interval at University College London, where average differences in measurements were all <0.05%; cortical white matter hyperintensities at UC Davis, where they found “minimum impact” (R>0.97); and others.ConclusionADNI‐4 de‐faces all indicated image types (Table 1) before subsequent pre‐processing, analyses, and public release. Sites upload images to the central repository (LONI), where they are queued to the MRI Core at Mayo Clinic for de‐facing. Image types requiring de‐facing are de‐faced using mri_reface. Trained analysts inspect de‐faced images to confirm complete removal of the face and complete non‐modification of the brain, and any flagged images are re‐processed manually until issues are resolved. De‐faced images are converted back to DICOM and re‐uploaded for further distribution. Image types not requiring de‐facing are indicated to proceed directly to subsequent steps. In summary, ADNI‐4 has adopted mri_reface because it successfully de‐faces without substantially affecting brain measurements.

AbstractBackgroundPhase four of the Alzheimer’s Disease Neuroimaging Initiative (ADNI4) began in 2023. This time‐period corresponded to MRI vendors introducing product sequences with compressed sensing (CS), cross‐vendor adoption of arterial spin‐labelling (ASL) and multi‐band slice excitation, and hardware improvements (head‐coils, increased gradient amplitudes). These advances enabled the acquisition of new imaging measures and reduced scan times. The ADNI4 MRI protocol aims to maintain longitudinal consistency across two decades of data acquisition, while adopting new technologies. Here we describe the design and justification behind the ADNI4 protocol and list the target imaging biomarkers.MethodsThe ADNI4 MRI protocol includes 9 MRI sequences; i) T1‐weighted (T1w) MP‐RAGE, ii) a compressed sensing T1w MP‐RAGE, iii) 3D T2w FLAIR, iv) structural isotropic 3D T2w v) T2*‐weighted susceptibility imaging vi) multi‐post labelling delay (PLD) 3D ASL, vii) 2D multi‐band‐multiple‐shell diffusion (dMRI), viii) 2D multi‐band EPI BOLD ix) high resolution, limited field of view, imaging centered on the hippocampus. Table 1 gives a detailed summary of the key protocol parameters and lists some of the imaging biomarkers that can be obtained from each series.ResultsStructural imaging was kept consistent across imaging sites. Variations in vendor specific sequences, software upgrades, and hardware performance capabilities, resulted in variations in advanced imaging strategies. Table 2 highlights the advanced sequences introduced in ADNI4, which will gradually be implemented as the required upgrades and licenses are obtained at each site. The main advanced sequences that are new in ADNI4 from ADNI 3 are i) CS for T1w images, ii) pseudo‐continuous‐ASL (pCASL) being implemented on all 3 vendors (GE, Siemens, Philips), iii) multi‐PLD ASL, which allows for the generation of quantitative maps of cerebral blood‐flow (CBF) and arterial transit times (ATTs), iv) and isotropic 1mm3 resolution for 3D FLAIR. When ADNI4 began, product sequences that utilized artificial intelligence were not widely available and could not be adequately tested before the rollout of ADNI4.ConclusionsThe ADNI4 database will continue to help the neuroimaging community, not only extract valuable imaging biomarkers, but provide the data necessary to test the impact of advanced imaging strategies on diagnostic accuracy and disease sensitivity.

AbstractBackgroundAutosomal dominant Alzheimer disease (ADAD) is characterized by genetic mutations affecting the beta‐amyloid (Aß) pathway. However, vascular and immune factors play important roles which are not completely understood. Understanding the function of the neurovascular unit (NVU) comprised of neurons, glial cells, and vasculature, at different disease stages appears ideal to developing and evaluating therapeutics. Omics approaches can inform NVU changes and their disease associations.MethodsCSF proteomic data using the Somalogic® platform was generated from 485 participants in the DIAN study, who also had PiB‐PET and MRI assessing Aß burden, white matter hyperintensity (WMH), and cerebral microhemorrhage (CMH). From previously published AD studies, we identified 33 NVU‐associated proteins with corresponding Somalogic aptamers (Figure 1). We used Principal Component Analysis (PCA) on participants with complete data (n = 262, 166 mutation‐carriers and 96 non‐carriers) to find groups of proteins associated with mutation status and disease stage as measured with estimated year to symptom onset (EYO), PiB‐PET, and clinical status. Linear‐mixed effect models and two‐part zero‐inflated negative binomial mixed models further evaluated the link between significant principal components (PCs) and WMH and CMH, respectively. Models accounted for age, APOE‐e4 status, sex, and education.ResultsThe first ten PCs explained >60% of the variance in NVU‐related protein levels (Figure 2A) with only PC5 (4.8%) associated with mutation*EYO, suggesting relationship with ADAD‐specific disease progression (***p<0.0001, Figure 2B). While asymptomatic PiB‐negative individuals were not significantly different from non‐carriers, asymptomatic PiB‐positive individuals had higher PC5 scores than non‐carriers and symptomatic individuals scored still higher (**p<0.005, ***p<0.0001, Figure 2C). The main proteins composing PC5 were: neurofilament light‐chain, angiopoietin‐2, SMOC1, LRP1, E‐selectin, and sTREM2, suggesting strong association with proinflammatory glial response from neuronal injury (Figure 2D). PC5 was also associated with WMH volumes (*p<0.05, Figure 3A) and with CMH count in participants with CMH (**p<0.005, Figure 3B).ConclusionsThis PCA approach indicates that some groups of NVU‐associated proteins are related with mutation status and disease stage, and that they are differentially associated with features of ADAD such as white matter integrity as measured by WMH volume. Weighted Correlation Network Analysis will be of interest to further understand these associations. Funding: K01AG080123, RF1‐AG044546, UF1AG032438

AbstractBackgroundFew normative data for computerized measures administered in unsupervised remote environments are available. We aimed to determine what variables to include in normative models for remote self‐administered assessments, develop normative data for measures administered through Mayo Test Drive (MTD, a multi‐device remote cognitive assessment platform) and evaluate application of norms.Method1240 adults ages 33‐100 (96% White) from the Mayo Clinic Study of Aging and Mayo Alzheimer’s Disease Research Center met normative sample inclusion criteria that included a concordant Cognitively Unimpaired (CU) diagnosis (3 independent raters all diagnosed CU) and CDR = 0 (see Table 1 for sample characteristics). Raw test scores were converted to normalized scaled scores. We derived regression‐based normative data that adjusted for age, age2, sex and education (base model); alternative models are also provided (age+age2+sex; age+ age2). We evaluated whether additional model terms beyond the base model were needed using an a priori cut‐off of 1% variance improvement in adjusted R2 over and above the base model. We examined rates of low test performance defined as performance below ‐1 SD in independent validation samples (n = 167 concordant CU, n = 64 mild cognitive impairment (MCI), n = 14 dementia). Rates are considered significantly different than expected when 95% confidence intervals (CIs) do not include the expected 14.7% base rate frequency.ResultNo additional potential model terms met our a priori cut‐off for inclusion beyond the base model (see Table 2), including other quadratic and cubic terms or interactions, device type, response input type, potential session interference, and national area deprivation index (ADI available in subset; N = 776). Application of norms showed expected rates of low test performance among CU participants (95% CI includes expected base rate) and significantly greater than expected rates of low test performance for all variables in MCI (e.g., 62.5% for composite) and dementia (e.g., 71.4% for composite; see Table 3).ConclusionTypical normative models appear appropriate for remote self‐administered measures and are sensitive to cognitive impairment. Device type and response input did not explain enough variance to warrant inclusion in normative models but are important for individual level interpretation. Future work will improve representation of individuals from under‐represented groups.

AbstractBackgroundWe recently developed a novel tau‐PET summary measure THETA, capturing regional heterogeneity and identifying tau status, using ground truth visual assessments from a large single‐center cross‐sectional dataset and validated on independent cohorts [1, 2]. In this study, we aimed to evaluate the performance of THETA on longitudinal and histopathology data.MethodWe included longitudinal tau‐PET ([18F]flortaucipir) data from 696 Mayo Clinic Study of Aging (MCSA) and ADRC participants, with histopathology in n = 90. Fig. 1 shows the model that uses regional standard uptake value ratios (SUVR) and a target of binary class of tau positivity for prediction of THETA. This model was applied to predict tau status on each followup scan. Slopes of the meta‐ROIs’ tau SUVRs and THETA were evaluated by diagnostic group and as a function of baseline amyloid SUVR in discordant CU participants (where meta‐ROI and visual assessments did not match at baseline) and in incident‐THETA+ CU participants (whose THETA moved from below to above 1 over serial tau‐PET scans). We evaluated tau measurements as a function of Braak stages for neurofibrillary tangles.ResultLongitudinal plots of meta‐ROIs for diagnostic groups were very similar, but the expanded range of THETA based on visual positivity prediction clearly identified tau positive or tau‐negative scans. In discordant‐CU (n = 97) and incident‐THETA+ CU participants (n = 14 all amyloid positive at follow‐up but only 65% at baseline), the relationship between baseline amyloid and rate of tau increase was stronger for THETA than temporal meta‐ROI (Fig. 2). Separation between baseline and follow‐up was greater for THETA (t‐statistics = 90) compared to temporal meta‐ROI (t‐statistics = 20) (p<0.01) in the incident‐THETA+ CU participants. THETA showed a slightly stronger association with Braak stage than meta‐ROIs (rho = 0.87 vs. ≤0.83, p<0.05), with better separation of clinical diagnoses (Fig. 3).ConclusionTHETA remained clearly negative or positive in MCI and AD, providing consistent information on underlying etiology of impairment at both baseline and follow‐up. Although binary in its construction, THETA both provided separation of values based on tau status and its change correlated with baseline amyloid burden especially in discordant‐CU where tau deposition is not in typical meta‐ROIs. Further work is needed to confirm if THETA captures early tau changes.

AbstractBackgroundBiomarker cutpoints have received significant importance in AD research. Reliable worsening method based on serial biomarker data has been proposed (Jack et al., 2017 Alz & Dem). Advanced machine learning models (ML) and explainable AI provide an opportunity to enhance existing techniques. We aimed to estimate biomarker cutpoints on cross‐sectional data using ML by identifying important transitions in the biomarker values.MethodWe first trained an ML model with the biomarker of interest as an input including other available biomarkers, demographics, and diagnosis, and predicted global cognition z‐score. We then obtained SHAP values (Shapley Additive exPlanations), fit a curve between the SHAP values and biomarker values, and calculated the first derivatives (slope) of the curve. Then, a CUSUM (Cumulative Sum) analysis was used to identify SHAP change points and cross‐referenced to values where the slopes were steepest near the zero‐crossing location of the curve. A zero SHAP is an important indicator of a feature’s progressive importance for model prediction because a feature’s influence on prediction significantly shifts when its SHAP values go from positive to negative or vice versa. However, relying on zero‐crossing alone may be insufficient when the relationship is non‐linear. Consequently, we calculated the second derivatives to identify inflection points (peaks and troughs) along the curve. Lastly, we compiled all the potential cutpoints and performed post‐processing as shown in Fig. 1.ResultThe algorithm identified cutpoints at 1.44 SUVR for all and at 1.42 SUVR for CU participants (Fig. 2). At age 70, both cutpoints estimated the percentage of abnormals at less than 40% (Fig. 3).ConclusionTPE amyloid‐PET SUVR cutpoints for CUs equal the reported RW values (1.42 SUVR). A strength of the TPE is its application on cross‐sectional data making it versatile for various biomarkers across different cohorts.

AbstractBackgroundPrior longitudinal studies among older adults have documented associations between hearing loss and changes in brain morphology. Whether interventions involving hearing aids can reduce age‐related atrophy is unknown. A substudy within the Aging and Cognitive Health Evaluation in Elders (ACHIEVE, Clinicaltrials.gov Identifier: NCT03243422) randomized controlled trial tested the effect of a best‐practices hearing intervention versus health education control on three‐year change in cortical thickness among older adults with hearing loss.MethodThe ACHIEVE study enrolled 977 community‐dwelling adults aged 70‐84 years at baseline (2018‐2019) with untreated hearing loss (better ear pure tone average [0.5‐4 kHz] ≥30 and <70 dB HL) and without substantial cognitive impairment from four sites across the U.S. (Jackson, MS, Forsyth County, NC, Minneapolis, MN, Washington County, MD). Participants were randomized to a hearing intervention (provision of hearing aids and related technologies, counseling, and education) or a health education control (individual sessions with a health educator covering topics relevant to chronic disease and disability prevention). Three‐dimensional magnetic resonance imaging was performed on 3 Tesla Siemens scanners in a subsample of 445 participants at the ACHIEVE baseline and three‐year follow‐up. Linear mixed effects models were used in intention‐to‐treat analyses to estimate three‐year change in cortical thickness. All models adjusted for baseline measures of hearing loss, recruitment source, site, age, sex, and education. Missing outcome and covariate data was imputed to mitigate bias caused by informative attrition.ResultAt baseline, 224 participants were women (50.3%), 52 participants were Black (11.7%), and the mean (SD) age was 76.4 (4.0) years old (Table 1). Compared to the health education control, the hearing intervention exhibited a nominally protective effect on three‐year change in average cortical thickness (Figure 1). The greatest effect size for cortical thickness was observed in the occipital lobe, while the smallest effect size was detected in the temporal lobe. Statistically significant effects were detected in the pars orbitalis, rostral anterior cingulate, posterior cingulate, and isthmus cingulate (Figure 2).ConclusionHearing aid use may reduce decline in cortical thickness among older adults. The effects of hearing aids may be greatest in regions other than those associated with the auditory cortex.