Statistical image analysis of longitudinal RAVENS images

Area of Science:

• Neuroimaging
• Biostatistics
• Medical Image Analysis

Background:

• Regional Analysis of Volumes Examined in Normalized Space (RAVENS) images are crucial for brain morphometry studies.
• Standard longitudinal voxel-based morphometry (VBM) analyses struggle to differentiate registration errors from true biological changes.
• This limitation is particularly problematic in studies of normal aging and chronic diseases where longitudinal changes are subtle.

Purpose of the Study:

• To introduce and validate longitudinal functional principal component analysis (LFPCA) as a superior method for analyzing RAVENS images.
• To demonstrate LFPCA's ability to disentangle registration errors from genuine longitudinal brain alterations.
• To apply LFPCA to identify neuroimaging markers associated with multiple sclerosis (MS) progression.

Main Methods:

• Analysis of RAVENS images using a longitudinal VBM approach.
• Application of LFPCA for high-dimensional longitudinal image analysis.
• Decomposition of RAVENS images into subject-specific random intercepts, slopes, and visit-specific deviations to isolate signals.

Main Results:

• LFPCA effectively separates registration errors from baseline and longitudinal signals in RAVENS images.
• The method decomposes variability into cross-sectional, irreversible longitudinal changes, and visit-specific deviations.
• Analysis revealed associations between regional brain atrophy, ventricular enlargement, and multiple sclerosis disease progression.

Conclusions:

• LFPCA offers a robust framework for analyzing longitudinal neuroimaging data, overcoming limitations of traditional VBM.
• The technique enhances the accuracy of detecting subtle brain changes in aging and disease states.
• LFPCA provides valuable insights into the neurobiological underpinnings of multiple sclerosis progression.