Brain functional network connectivity interpolation characterizes the neuropsychiatric continuum and heterogeneity

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Schizophrenia (SZ) and autism spectrum disorder (ASD) present heterogeneous symptoms, often viewed as a continuum.
• Conventional diagnostic and neuroimaging methods struggle with individual differences and disease progression.
• Functional network connectivity (FNC) analysis offers insights but requires advanced modeling for continuous representation.

Purpose of the Study:

• To develop and validate a variational autoencoder (VAE) framework for interpolating functional network connectivity (FNC) data.
• To estimate the neuropsychiatric continuum and heterogeneity in SZ and ASD.
• To outperform existing linear and semi-supervised methods in capturing FNC properties.

Main Methods:

• Utilized a VAE framework with static (sFNC) and dynamic (dFNC) data from controls and patients with SZ or ASD.
• Compared VAE performance against linear and semi-supervised baseline models.
• Generated continuous FNC data and analyzed latent space trajectories.

Main Results:

• The VAE framework significantly outperformed baseline methods in interpolating both sFNC and dFNC data.
• Generated FNC data captured representative and generalizable properties of the original datasets.
• Identified specific patterns of altered correlations along the SZ and ASD continua, including within and between brain networks.

Conclusions:

• The proposed VAE framework effectively models the FNC continuum in psychiatric and neurodevelopmental disorders.
• This approach enables data-driven discovery, visualization of individual differences, and estimation of disease stage.
• Offers advantages over traditional methods for understanding heterogeneity and progression in SZ and ASD.