Abstract
Not everyone ages the same way. While cognitive decline is common in older adults, a remarkable subset — often called “Super-Agers” or “Positive-Agers” — maintains cognitive performance well above age-matched norms, sometimes rivaling individuals decades younger. At the other extreme, “Cognitive Decliners” show accelerated loss of function that may herald progression toward mild cognitive impairment or Alzheimer’s disease. Understanding what separates these groups at the level of the brain is one of the most consequential questions in aging research: if we can identify the neural signatures of exceptional cognitive resilience, we gain both a screening tool to flag at-risk individuals and a set of biological targets that interventions could aim to preserve.
This line of research, supported by an Alzheimer’s Association Research Grant ($148,300), has been a sustained focus of my program. Using multimodal neuroimaging data from the UK Biobank and ADNI cohorts, we have systematically investigated how different imaging modalities capture the distinction between Positive-Agers, Normal-Agers, and Cognitive Decliners.
Our first major contribution was the development of OLBO (Optimal Labeling using Bayesian Optimization), a hybrid machine learning and optimization algorithm designed to classify extreme cognitive phenotypes from resting-state functional MRI (rsfMRI) data. OLBO addresses a fundamental challenge in this field: because cognitive aging exists on a continuum, the boundaries between groups are not given by nature — they must be learned. OLBO jointly optimizes the labeling of individuals into cognitive classes and the classification model itself, producing more robust and interpretable phenotype assignments than conventional thresholding approaches. Applied to rsfMRI functional connectivity networks from the UK Biobank, OLBO successfully differentiated Positive-Agers from Cognitive Decliners and identified the key neural functional connectivity networks contributing to these differences. This work was published in Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring (2024).
We then extended this investigation to structural MRI and diffusion-weighted MRI, asking whether the same cognitive phenotypes could be distinguished using different aspects of brain anatomy. Using structural MRI, a hybrid feature selection method improved classification accuracy and identified structural brain signatures — patterns of cortical thickness, subcortical volumes, and morphological features — associated with cognitive resilience. This study was published in GeroScience (2024). Using diffusion MRI, we examined white-matter tract integrity and found that specific tracts and demographic factors distinguish potential Super-Agers, with diffusion metrics providing complementary information to what functional connectivity alone can capture. This work appeared in GeroScience (2025). An earlier exploratory study characterized the broader biomarker and neuroimaging profile of super-aging in the UK Biobank (GeroScience, 2023).
An ongoing study brings these threads together by comparing the discriminative power of all three modalities — rsfMRI, sMRI, and dMRI — within a unified analytical framework, asking which modality, or which combination, provides the most reliable identification of cognitive phenotypes. This comparative analysis is in preparation for GeroScience.
Throughout this work, I have mentored two Ph.D. students and multiple undergraduate researchers, producing publications, poster awards at AAIC, IISE, and INFORMS conferences, and an invited podcast for the Alzheimer’s Association’s ISTAART.