Providing context: Extracting non-linear and dynamic temporal motifs from brain activity.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Psychiatric Disorders

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for studying brain dynamics.
• Time-resolved functional connectivity (tr-FC) is a common analysis method, but often relies on linear approaches.
• Existing methods may not fully capture the complex temporal dynamics of brain activity.

Purpose of the Study:

• To develop a non-linear deep learning model for analyzing rs-fMRI dynamics.
• To investigate if the model can differentiate between individuals with schizophrenia and healthy controls.
• To explore the relationship between the model's output and clinical characteristics like age and symptom severity.

Main Methods:

• Utilized a disentangled variational autoencoder (DSVAE), a generative non-linear deep learning model.
• DSVAE factorizes window-specific (context) and timestep-specific (local) information to capture multi-scale temporal differences.
• Compared the model's performance against baseline models and standard linear tr-FC approaches.

Main Results:

• The DSVAE model's context embeddings more accurately separated schizophrenia patients from controls in a low-dimensional space compared to baseline and tr-FC methods.
• The model's embeddings showed significant correlations with age and symptom severity in individuals with schizophrenia.
• Identified three distinct patient-specific brain activity clusters with varying functional network connectivity (FNC) patterns.

Conclusions:

• The proposed DSVAE model offers a novel, non-linear approach to analyzing rs-fMRI temporal dynamics.
• This method enhances the ability to distinguish patient groups and identify clinically relevant features.
• The DSVAE model has potential for improving psychiatric disorder research by providing more sensitive neuroimaging analysis tools.