Learning Sequential Information in Task-based fMRI for Synthetic Data Augmentation

TL;DR

This paper introduces a novel method for generating synthetic task-based fMRI data using an adapted alpha-GAN model, enhancing training datasets for improved autism spectrum disorder classification.

Contribution

It proposes a new approach combining GAN and variational autoencoder techniques to synthesize high-resolution, task-specific fMRI sequences for data augmentation.

Findings

Synthetic fMRI improves ASD classification accuracy

The method produces visually realistic fMRI sequences

Synthetic data enhances training datasets for medical imaging

Abstract

Insufficiency of training data is a persistent issue in medical image analysis, especially for task-based functional magnetic resonance images (fMRI) with spatio-temporal imaging data acquired using specific cognitive tasks. In this paper, we propose an approach for generating synthetic fMRI sequences that can then be used to create augmented training datasets in downstream learning tasks. To synthesize high-resolution task-specific fMRI, we adapt the $\alpha$-GAN structure, leveraging advantages of both GAN and variational autoencoder models, and propose different alternatives in aggregating temporal information. The synthetic images are evaluated from multiple perspectives including visualizations and an autism spectrum disorder (ASD) classification task. The results show that the synthetic task-based fMRI can provide effective data augmentation in learning the ASD classification task.