Capturing heterogeneous group differences using mixture-of-experts: Application to a study of aging.

Area of Science:

• Neuroimaging and Computational Neuroscience
• Brain Connectivity Analysis
• Aging and Cognitive Function

Background:

• Linear models in MRI studies assume homogeneous disease/aging patterns, limiting the capture of individual differences.
• Kernel-based methods can model non-linear effects but struggle with information extraction and interpretation of heterogeneity.
• Existing methods lack explicit modeling of heterogeneous patterns of change in affected groups relative to controls.

Purpose of the Study:

• To present a novel method for explicitly modeling and capturing heterogeneous patterns of change in aging brain function.
• To identify distinct subgroups within older adults exhibiting different functional connectivity patterns.
• To investigate the relationship between heterogeneous brain aging patterns, cognitive reserve, and cognitive decline.

Main Methods:

• Utilized the Mixture-of-Experts (MOE) framework, combining unsupervised mixture modeling with supervised classification.
• MOE approximates non-linear group boundaries with piecewise linear boundaries to discover multiple patterns of group differences.
• Applied the MOE method to resting-state functional MRI data from the Baltimore Longitudinal Study of Aging (BLSA).

Main Results:

• Identified two distinct subgroups of older adults (>85 years) with similar age distributions but different functional connectivity patterns.
• Both subgroups showed reduced Default Mode Network (DMN) connectivity; one subgroup additionally exhibited increased pre-frontal cortex and insula connectivity.
• The subgroup with increased connectivity showed baseline cognitive similarity to younger adults (<60 years) in some domains and faster decline in others.

Conclusions:

• The MOE framework effectively models heterogeneous brain aging patterns and identifies distinct patient/control subgroups.
• Older adults may exhibit differential recruitment of cognitive reserve, manifesting as varied functional connectivity patterns.
• Heterogeneous aging trajectories highlight the complexity of brain-cognition relationships in later life.