Decoding brain structure-function dynamics in health and in psychosis via an autoencoder.

Area of Science:

• Neuroscience
• Brain Imaging
• Computational Psychiatry

Background:

• Understanding the brain's structure-function relationship is crucial for neuroscience and psychiatry.
• Diffusion MRI and magnetoencephalography are key techniques for measuring structural and functional connectivity, respectively.

Purpose of the Study:

• To develop and compare a deep-learning model and an analytical model for mapping structural connectivity to functional connectivity.
• To investigate structure-function coupling in healthy individuals and individuals with psychosis.

Main Methods:

• A Graph Multi-Head Attention AutoEncoder (deep-learning) was used to map diffusion MRI-derived structural connectivity to magnetoencephalography-derived functional connectivity.
• An analytical model using shortest-path-length and search-information was employed for comparison.
• Both models were applied to data from healthy participants and individuals with psychosis.

Main Results:

• The deep-learning model achieved higher accuracy (correlation coefficients > 0.8) in predicting functional connectivity in healthy individuals compared to the analytical model (0.45) in alpha and beta bands.
• Distinct structure-function coupling patterns were observed in individuals with psychosis compared to healthy controls.
• Alterations in the structure-function relationship were more pronounced in psychosis than structure- or function-specific changes.

Conclusions:

• Human brain structural and electrophysiological functional connectivity are tightly coupled.
• The deep-learning model offers a more sophisticated approach to understanding brain dynamics, particularly in higher frequency bands.
• The study highlights the utility of advanced computational models in identifying neural alterations in psychiatric conditions like psychosis.