Senjem, Matthew L.

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 Aging and dementia are generally associated with lower metabolism in gray matter on FDG‐PET and have been widely studied. However, white matter (WM) metabolic changes remain understudied. We investigated the metabolic changes in WM associated with aging and dementia and their relationships with established Alzheimer's Disease (AD) biomarkers and cognitive outcomes. Method We analyzed data from 3,169 participants (72 ± 9 years; 53% male; 65% cognitively unimpaired) enrolled in the Mayo Clinic Study of Aging (MCSA) and the Alzheimer's Disease Neuroimaging Initiative (ADNI). All participants underwent baseline MRI, amyloid‐PET, and FDG‐PET scans, as well as neuropsychological evaluations, with at least one clinical follow‐up. Voxel‐wise multiple regression analysis identified WM metabolic changes associated with global cognition in cognitively unimpaired MCSA participants ( n = 1,754). Univariate analyses and structural equation models were applied to elucidate the driving factors underlying the WM metabolic changes in MCSA data. Multiple regression and linear mixed‐effects models evaluated the explanatory utility of WM metabolism for baseline global cognition and cognitive decline in both cohorts. In a subset of MCSA participants, we evaluated neurite density index from diffusion MRI as a surrogate of WM health and glial fibrillary acidic protein (GFAP) as a surrogate of neuroinflammation to understand underlying mechanisms. Result We identified an atypical WM signature (AWM‐Signature) in the anterior and superior corona radiata, where glucose metabolism increased as cognitive performance worsened (Figure 1). Hypermetabolism in AWM‐Signature was related to increased age, amyloidosis, vascular disease, and neuroinflammation, as well as decreasing WM neurite density in the same regions (Figure 2). Hypermetabolism in AWM‐Signature regions predicted faster future cognitive decline, even after accounting for demographics, amyloid, and vascular disease metrics. The results observed in the AWM‐Signature regions were consistently in the opposite direction of the associations seen with gray matter and typical WM regions. Conclusion We found unexpected metabolism increases in corona radiata, a hub that carries significant portion of the brain's neural traffic, with aging and disease which likely indicates a compensatory response. These changes may be useful cognitive resilience measures and require further investigation to understand their role in maintaining cognitive function.

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 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 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 T2*‐weighted imaging is of paramount importance for patient treatment eligibility and safety monitoring in anti‐amyloid pharmaceutical clinical trials and clinical care. Recently with the adoption of multi‐echo GRE (GRE) sequences, susceptibility weighted imaging (SWI), and quantitative susceptibility mapping (QSM) can be obtained in the same acquisition. In 2024, the QSM Consensus Organization Committee released a white paper recommending at least 1 mm isotropic voxel size for QSM. In clinical workflows microbleed detection is performed with higher in‐plane resolution (0.5x0.5mm 2 ), and large out‐of‐plane resolution (∼1.8mm), which was used in the ADNI4 protocol. In the CLARiTI protocol, a 1mm isotropic resolution was adopted. The objective of this work was to compare susceptibility estimates between the ADNI4 (0.5x0.5x1.8mm 3 ) and CLARiTI (1mm 3 ) protocols. Method The ADNI4 and CLARiTI QSM protocols were acquired on forty‐seven subjects (33 cognitively unimpaired (CU), 14 cognitively impaired (CI), in the same scan session. The groups were defined by their Clinical Demential Rating (CDR) global score (CDR global = 0 for CU, and CDR global >= 0.5 for CI subjects). QSM images were generated using publically available software (STI Suite, UC Berkeley). Mean susceptibility values in atlas regions were compared. Result Significant group differences (CU vs. CI) were observed in the fusiform and middle occipital gyrus with the ADNI4 protocol, only in the rectus with the CLARiTI protocol (Figure 1). These three regions are small in size and closer to surface‐air interfaces, which may have impacted estimation accuracy. However, in the substantia nigra and its sub‐regions (pars compacta and pars reticulata), which shows increased susceptibility in Lewy body disease, the ADNI4 and CLARiTI protocols were in good agreement (r 2 > 0.9) (Figure 2). Conclusion Group differences were observed between the ADNI4 and CLARiTI QSM only in 3 small regions close to surface‐air interfaces in this small dataset. The two techniques showed highly correlated (r 2 = 0.93) estimates in the substantia nigra, a region with high susceptibility due to iron deposition. Strong correlations between the two protocols are encouraging. Radiologists have historically preferred submillimeter in‐plane resolution for microbleed detection, which is more time manageable with anisotropic imaging.

Abstract Background Neuroimaging data typically arrives as Digital Imaging and Communications in Medicine (DICOM) files, batched by exam. We use exam or study to denote a single imaging session of a participant, containing up to ∼40 series (the generic term for scans and derived products). Different series have different properties and applications, making their labeling complex. Filenames often change in transit, and Series Description tags (0008, 103E) are arbitrary, operator‐entered, and unreliably coupled to the series type. Furthermore, label preferences vary from site to site, making the application of a common classification system mandatory for multisite studies. Here we describe the system that we have developed and applied to the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and SCAN. Method Although Series Description is unreliable, DICOM includes other tags that parameterize MR pulse sequences and suggest their purpose. The mapping from these features to types is difficult to explicitly specify, especially since different manufacturers and protocols use different conventions. Thus we developed a machine learning classifier using 4060 semiautomatically labeled head and spinal series from 1.5 to 7T MR scanners, including both site‐specific and standard protocols designed to study a wide range of neurological conditions, including dementia, hydrocephalus, and multiple sclerosis. An 80%/20% split, stratified by type, was used for the training/test sets. After initial training, a few cases in the training set with high errors due to rarity were augmented and the model retrained. Performance was evaluated on the clean test set. Result The taxonomy is presented in Table 1. The classifier correctly identified 807 out of 812 (99.4%) series, with the error cases coming from outside the ADNI or SCAN protocols (Table 2). The confusion matrix is shown in Figure 1. Conclusion We implemented an automatic series type classifier with labels chosen to be generally useful for the brain MRI community. Some errors were noted, but the test set was deliberately challenging and the errors appear to be ones that could easily be made by a human. The classifier is expected to have good tolerance for as yet unseen head MR protocols but is likely unsatisfactory for body MRI.

Abstract INTRODUCTION M0 images were missing in Siemens ASL data in Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, prohibiting cerebral blood flow (CBF) quantification. METHODS A conditional latent diffusion model was trained and evaluated on in‐house datasets, then applied to the Siemens data in ADNI‐3. Regional CBF differences by Alzheimer's disease (AD) stages, their accuracy for AD classification, and CBF trajectory slopes were compared between generated data (Siemens) and acquired data (General Electric). RESULTS The diffusion model generated M0 images with high fidelity (SSIM = 0.918 ± 0.023, PSNR = 31.361 ± 2.537) and minimal CBF bias (mean difference is 0.21 ± 1.58 mL/100 g/min). Both generated and acquired CBF showed similar spatial patterns and decreasing trends with AD progression in specific AD‐related regions. Generated CBF also improved accuracy in classifying AD stages compared to qualitative perfusion images. CONCLUSION This study shows the potential of diffusion models for imputing missing modalities in large‐scale studies exploring the use of ASL as a biomarker of AD. Highlights Using latent diffusion model, we can generate M0 image from control image in arterial spin labeling (ASL) with high fidelity. The generated M0 can be used for cerebral blood flow (CBF) quantification in Alzheimer's Disease Neuroimaging Initiative dataset. The performance of classification between Alzheimer's disease (AD) patients and cognitive normal people is better when using generated CBF maps than using non‐quantitative perfusion images. ASL CBF decreases with AD progression in key AD‐related brain regions.

Importance Advances in Alzheimer disease (AD) have shifted research focus to earlier disease stages, necessitating more scalable approaches to identify cognitively unimpaired individuals with amyloid β (Aβ) pathology. Objective To assess the utility of plasma phosphorylated tau 217 (p-tau217) for classifying Aβ status in cognitively unimpaired individuals, both as a stand-alone test and in a 2-step approach where positive plasma results were confirmed using a second modality (Aβ positron emission tomography [PET] or cerebrospinal fluid [CSF]). Design, Setting, and Participants This cross-sectional cohort study used data collected between June 2009 and March 2024. We included 2916 cognitively unimpaired participants from 12 international independent observational cohorts in the US, Europe, Australia, and Canada with available plasma p-tau217 levels and CSF or PET Aβ biomarkers. Performance comparisons between mass spectrometry and immunoassay-based p-tau217 measurements were also performed (n = 964). Exposures Plasma p-tau217 levels measured by immunoassay. Main Outcome and Measures Aβ status, determined by CSF or Aβ PET biomarkers. Results Participants had a mean (SD) age of 66.9 (9.9) years; 971 (33.3%) were Aβ positive by either CSF or PET, 1667 (57.2%) were women, and 1108 (38.1%) carried at least 1 APOE ε4 allele. As a stand-alone test, plasma p-tau217 achieved a positive predictive value (PPV) of 79% (95% CI, 74-84) and an overall accuracy of 81% (95% CI, 80-82). In a 2-step workflow, the PPV and accuracy significantly increased to 91% (95% CI, 86-95). While this approach required screening of 677 individuals with plasma p-tau217 to identify 100 Aβ-positive individuals, compared to 536 participants when using PET alone, it reduced the need for PET testing to 124. Immunoassays demonstrated comparable PPVs to mass spectrometry (80% [95% CI, 74-86] vs 85% [95% CI, 81-90]; P = .12) but significantly lower overall accuracy (82% [95% CI, 79-84]% vs 88 [95% CI, 86-90]; P &amp;lt; .001) and true Aβ-positive detection rate (49% [95% CI, 43-55] vs 69% [95% CI, 64-75]; P &amp;lt; .001). Conclusions and Relevance The findings highlight the potential of plasma p-tau217 as a stand-alone test—or when used in a sequential 2-step approach alongside PET or CSF testing—as a cost-effective, scalable, and minimally burdensome strategy for identifying preclinical AD. Tailored screening workflows that incorporate p-tau217 can improve efficiency in participant selection for preclinical AD trials and, in the future, help guide access to disease-modifying treatments.