Eigenanatomy improves detection power for longitudinal cortical change.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biostatistics

Background:

• Standard voxel-based morphometry (VBM) analyzes neuroimaging data voxel-by-voxel, often requiring extensive multiple comparisons correction.
• Traditional region-of-interest (ROI) analyses can improve detection power but are often predefined and not data-driven.

Purpose of the Study:

• To introduce eigenanatomy, a novel and interpretable dimensionality reduction strategy specifically designed for neuroimaging data.
• To enhance the detection power in morphometry by utilizing data-driven, anatomically informed regions.

Main Methods:

• Eigenanatomy approximates the eigendecomposition of image sets using sparse, unsigned, and anatomically clustered basis functions (eigenanatomy vectors).
• The strategy reverses the typical VBM workflow by first clustering voxel data and then applying linear regression in a reduced dimensionality space.
• This approach generates localized, data-driven regions of interest.

Main Results:

• Eigenanatomy provides a principled objective function that identifies localized, data-driven regions of interest.
• These eigenanatomy-derived regions significantly improve the ability to detect and quantify cortical changes in neurodegenerative diseases.
• The method demonstrated efficacy in analyzing two distinct forms of neurodegeneration.

Conclusions:

• Eigenanatomy offers a powerful and interpretable alternative to standard VBM, enhancing statistical power in neuroimaging morphometry.
• The data-driven, anatomically clustered regions identified by eigenanatomy are crucial for accurately quantifying biological changes in the brain.
• This method holds significant potential for advancing the study of neurodegenerative diseases.