Functional brain network identification and fMRI augmentation using a VAE-GAN framework.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Deep learning models excel at mapping functional brain networks from fMRI data.
• High dimensionality and limited data in fMRI lead to overfitting in deep learning models.
• fMRI data augmentation techniques are underexplored.

Purpose of the Study:

• To develop a novel framework for functional brain network identification and fMRI data augmentation.
• To address overfitting issues in deep learning models for fMRI analysis.
• To improve the performance of brain network modeling and classification tasks.

Main Methods:

• Developed a Variational Auto-Encoder Generative Adversarial Network (VAE-GAN) framework.
• Utilized VAE to model fMRI data distribution for generalized feature extraction.
• Employed GAN's discriminator to enhance the quality of generated fMRI data.

Main Results:

• The VAE-GAN framework effectively models fMRI data distribution, reducing overfitting.
• Generated fMRI data quality surpasses that of standard VAE and GAN methods.
• Demonstrated superior performance in identifying temporal features and functional brain networks on HCP datasets.

Conclusions:

• The VAE-GAN framework offers an effective solution for fMRI data augmentation and brain network analysis.
• Generated fMRI data improves brain network modeling and ADHD classification accuracy on the ADHD-200 dataset.
• The proposed VAE-GAN framework overcomes limitations of VAE and GAN in fMRI data modeling.