Building Multivariate Molecular Imaging Brain Atlases Using the NeuroMark PET Independent Component Analysis...

Area of Science:

• Neuroimaging
• Molecular Imaging
• Computational Neuroscience

Background:

• Traditional positron emission tomography (PET) analyses use macro-anatomical regions of interest (ROIs) that may not align with chemo-architectural boundaries.
• Independent component analysis (ICA) has been used to address this, but its data-driven nature complicates cross-study comparisons.

Purpose of the Study:

• Introduce NeuroMark PET, a novel approach using spatially constrained ICA to define overlapping brain regions reflecting molecular architecture.
• Develop a fully automated pipeline for generating replicable amyloid-beta networks (AβNs) from PET data.

Main Methods:

• Generated an ICA template for florbetapir (FBP) PET targeting amyloid-beta (Aβ) using blind ICA on large datasets.
• Defined Aβ networks (AβNs) by selecting components targeting Aβ and omitting others.
• Validated the NeuroMark PET pipeline against a standard neuroanatomical atlas using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Main Results:

• NeuroMark PET captures biologically meaningful, participant-specific features and shows higher sensitivity for detecting age-related changes than traditional ROIs.
• The most age-associated AβN (cognitive control network, CC1) showed a stronger association with age compared to macro-anatomical ROIs.
• The approach successfully differentiated white matter components from AβNs, improving artifact separation.

Conclusions:

• NeuroMark PET provides a fully automated, accurate, and reproducible framework for defining brain AβNs.
• This method enhances the investigation of molecular underpinnings of brain function and pathology.
• NeuroMark PET offers a valuable alternative to traditional ROI-based analyses in molecular imaging.