Disentangling Alzheimer's disease neurodegeneration from typical brain aging using MRI and machine learning. (December 2021)

• Record Type: Journal Article
• Authors: Hwang, Gyujoon; Abdulkadir, Ahmed; Erus, Guray; Habes, Mohamad; Pomponio, Raymond; Shou, Haochang; Doshi, Jimit; Mamourian, Elizabeth; Rashid, Tanweer; Bilgel, Murat; Fan, Yong; Sotiras, Aristeidis; Srinivasan, Dhivya; Morris, John C.; Marcus, Daniel S.; Albert, Marilyn S.; Bryan, Nick; Resnick, Susan M.; Nasrallah, Ilya M.; Davatzikos, Christos; Wolk, David A.
• Is Part Of: Alzheimer's & dementia. Volume 17(2021)Supplement 4
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.051532
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Neuroimaging biomarkers that discriminate between healthy brain aging (BA) and Alzheimer's disease (AD) are valuable in assessing the heterogeneity of neurodegeneration. Prior work has demonstrated that machine learning can detect patterns of brain change related to the two processes on an individual level (including the SPARE [Spatial Patterns of Atrophy for REcognition]‐AD and SPARE‐BA indices investigated herein). However, the substantial overlap between brain regions affected in AD and BA confounds our ability to measure them independently. We present a methodology, and associated results, toward disentangling the two. Method: T1‐weighted MR images of 4, 078 participants (48‐95 years) with AD, mild cognitive impairment (MCI), or cognitively normal (CN) diagnoses from the iSTAGING (Imaging‐based coordinate SysTem for AGIng and NeurodeGenerative diseases) consortium were analyzed. First, a subset of 711 patients with probable AD and 711 age‐ and sex‐matched probable CN adults were selected based purely on clinical diagnosis to train multivariate models of brain age (SPARE‐BA1; regression of age using CN) and of an AD index (SPARE‐AD1; classification of CN versus AD). Second, analogous groups a+AD and a‐CN were selected based on clinical and molecular markers (a+AD: 711 amyloid‐positive subjects with AD/MCI, as well as amyloid‐ and tau‐positive CN adults; a‐CN: 711 amyloid‐negative CN adults). Finally, the combined group of a‐CN and a+AD was used to train SPARE‐BA3 model, with the notion that it would avoid AD‐related changes. Result: Several subcortical brain regions became more weighted in the SPARE‐BA3 model than in SPARE‐BA1 and SPARE‐BA2, while regions in the temporal and occipital lobes became more weighted in SPARE‐AD2. The correlation between the brain age gap (SPARE‐BA minus chronological age) and SPARE‐AD in the a‐CN group was reduced to insignificant levels ( r =0.29 to ‐0.06). While both SPARE‐AD scores well separated AD from CN (area‐under‐the‐curve [AUC]>0.89), the brain age gap with SPARE‐BA3 (AUC=0.63) performed worse compared to SPARE‐BA1 (AUC=0.82). SPARE‐BA3 was generally less correlated with cognition and amyloid biomarkers in patients with AD than SPARE‐BA1, while SPARE‐AD2 demonstrated similar correlations compared to SPARE‐AD1. Conclusion: By employing conservative molecular diagnoses and decoupling SPARE‐BA from SPARE‐AD, we achieved more dissociable neuroanatomical biomarkers of healthy brain aging and AD.