Kremers, Walter K.

Abstract Background Brief self‐administered cognitive assessments deployed remotely and combined with plasma biomarkers of Alzheimer’s disease (AD) are well suited to assess clinical symptoms and underlying biology in large‐scale decentralized studies. We investigated whether associations of self‐administered cognitive measures and plasma biomarkers were comparable to associations of in‐person cognitive measures and plasma biomarkers. Method Participants included 327 adults from the Mayo Clinic Study of Aging (mean age = 70.8; 48% female; mean education = 16.0 years; 96.9% white). Most (94.5%) were cognitively unimpaired; 5.5% had consensus diagnosis of mild cognitive impairment. Participants completed Mayo Test Drive (MTD; Stricker et al., 2022), a remote, self‐administered, multi‐device compatible web‐based cognitive assessment comprised of the Stricker Learning Span (SLS) and Symbols test that are combined into a MTD composite. Participants also completed in‐person neuropsychological assessment (global‐z; Mayo Preclinical Alzheimer’s disease Cognitive Composite [Mayo‐PACC]) and fasting venipuncture. Plasma was assayed via HD‐X simoa platform for amyloid‐beta (Aß)‐42 and Aß‐40 to create an Aß42/40 ratio, glial fibrillary acidic protein (GFAP), and neurofilament light chain (NfL) using the N4PE Quanterix kit as well as phosphorylated‐tau protein 181 (p‐tau181). P‐tau181, GFAP, and NfL were log10‐transformed. Multiple regression analyses adjusting for age, sex, and education examined associations between plasma biomarkers and composite and subtest scores from remote and in‐person measures. Davidson‐Mackinnon J‐test compared R‐squared values of models with remote versus in‐person composites. Result Lower performances on MTD, global‐z, and Mayo‐PACC (Table 1) were associated with higher GFAP (p’s = .02), NfL (p’s = .003), and p‐tau181 (p’s = .04). For memory, worse SLS and AVLT sum of trials were associated with higher GFAP (p’s = .01) and NfL (p’s = .03). For processing speed/executive function, worse Symbols and Trails B were associated with higher p‐tau181 (p’s = .02). Worse Trails B and Digit Symbol Coding were associated with higher NfL (p’s = .02), and worse Digit Symbol Coding was associated with lower Aß42/40 (p = 0.03; Table 2). A multivariable model with MTD explained significantly more variance (p = .04) for NfL (adjusted R‐squared = 0.461) than a multivariable model with Mayo‐PACC (adjusted R‐squared = 0.447); remote and in‐person composite measures were comparable for other associations (p’s>.05). Conclusion Remote and in‐person composite measures showed comparable associations with AD‐related plasma biomarkers, supporting MTD’s criterion validity.

AbstractBackgroundWe have previously shown that 3D T1‐weighted and 3D T2‐FLAIR brain MRI can potentially be re‐identified using face recognition, but this has not been tested for other MRI sequences commonly used in imaging studies of aging and dementia. Here we compare match rates of automated face recognition to re‐identify 3D T1‐weighted, 3D T2‐FLAIR, 3D T2‐weighted, 2D Axial T2‐FLAIR, multi‐echo GRE (T2*), diffusion MRI, functional MRI, ASL, and T2* GRE images.MethodWe used a dataset of 182 Mayo Clinic participants each with MRI and face photographs. MRIs used Siemens and GE vendor‐product sequences similar to ADNI3 and ADNI4 protocols. We created face reconstructions (“basic”, with minimal preprocessing, and “advanced” that automatically filled missing face regions using a template) from each MRI sequence, and we trained a Microsoft Azure PersonGroup face recognition classifier to recognize these MRI‐based faces. We then queried the classifier with five face photographs for each participant, and we recorded how often the classifier chose the correct MRI as the top‐ranked match. For image types with match rates > = 10%, we then repeated the experiment after de‐facing each MRI with mri_reface, our leading software that replaces faces in each image (“re‐faces”) with an average face.ResultResults are presented in Table 1. 3D T2‐FLAIR, 3D T1‐weighted, and 3D T2‐weighted images were matched at 96‐98%, and mri_reface reduced all to 8%. Axial T2‐FLAIR and ME‐GRE were matched at 32‐45%, and mri_reface reduced both to 1%. T2*‐GRE matched at 10%, and mri_reface reduced this to 1%. ASL matched at 8%, while dMRI and fMRI matched at only 0‐2%.ConclusionAmong current‐generation vendor‐product sequences, we recommend that de‐facing be applied before research data sharing of T1‐weighted, T2‐weighted, T2‐FLAIR, T2*, and ME‐GRE images. Minimal match rates for current‐generation product fMRI, ASL, and fMRI images suggest that they present low risk of re‐identification and can be shared without de‐facing, but this conclusion should be re‐evaluated if they are acquired without fat‐suppression, with a larger field of view, or if newer developments reduce the current levels of artifacts and distortion around the face.

AbstractBackgroundAutomated face recognition can potentially re‐identify de‐identified research brain MRI, PET, and CT, and direct differences of brain measurements from original and de‐faced images are statistically significant but very minor, i.e. below scan‐rescan differences. Their effects on analyses correlating brain imaging biomarkers with clinical variables remain unknown.MethodWe sampled 3 age‐and‐sex‐matched groups (CU, MCI, and clinical AD) from Mayo Clinic aging studies, each with 61 participants with same‐visit 3D T1‐weighted and 3D T2‐FLAIR MRI, PiB amyloid PET, and Flortaucipir tau PET (total n = 183). We automatically measured hippocampal volume, PIB global SUVR, and tau temporal meta‐ROI SUVR using a) our in‐house pipeline with SPM12, MCALT, and ANTs, and b) FreeSurfer 7.3.2 (PETSurfer), and we measured voxels associated with clinical group separation using SPM12. We also measured WMH from FLAIR using a) our in‐house pipeline, b) Lesion Segmentation Tool (LST), and c) FreeSurfer’s Samseg. For each measurement, we calculated pair‐wise group separation AUC values and Spearman’s rho with age and CDR‐SOB. We then de‐faced all PET and MRI using mri_reface, our top‐performing automated de‐facing software that replaces imaged faces with an average face (“re‐facing”), and we repeated all measurements and analyses with de‐faced images. We then compared the magnitudes of clinical correlations with original images and those of de‐faced images.ResultAnalyses of original and de‐faced images had very high agreement. Only 2/55 paired comparisons were significant (Tables 1‐2): FreeSurfer‐measured global PIB SUVR from de‐faced images was more correlated with age (rho 0.003 vs ‐0.227, p = 0.046), and Mayo‐measured WMH volume from original images was more correlated with CDR‐SOB (rho 0.321 vs 0.308, p = 0.020). Another comparison‐pair tested p = 0.041 for an AUC change +0.008, which we consider not practically meaningful. All others had p>0.05. In voxel‐based analyses of CU vs AD participants, all comparisons were qualitatively similar between original and de‐faced variants (Figure 1).ConclusionDe‐facing PET and MRI using mri_reface did not significantly affect 53/55 correlations between imaging and clinical variables; of the two affected, one became stronger with re‐faced images. Stronger correlations with clinical variables may be explained by de‐facing’s standardizing face voxels and artifacts that irrelevantly affect brain segmentations.