Data augmentation using generative adversarial neural networks on brain structural connectivity in multiple...

Area of Science:

• Neuroimaging
• Machine Learning
• Biomedical Data Science

Background:

• Machine learning requires large datasets, which are often limited and imbalanced in biomedical research, especially for conditions like Multiple Sclerosis (MS).
• Generative models offer a solution by creating synthetic data to augment existing datasets.

Purpose of the Study:

• To propose a framework using generative adversarial networks (GANs) to synthesize structural brain networks in MS patients.
• To evaluate the quality and utility of these synthetic brain networks.

Main Methods:

• A GAN-based framework was developed to generate synthetic structural brain networks from T1 and diffusion tensor imaging (DTI) data of MS patients.
• The generated networks were assessed for structural properties and their impact on classification performance.

Main Results:

• Generated synthetic brain networks showed no significant quantitative or qualitative differences compared to real data.
• Augmenting the dataset with synthetic samples improved classification performance from 66% to 81% F1-score.

Conclusions:

• The proposed GAN framework provides a novel tool for data augmentation in connectome-based biomedical research.
• This approach offers a viable alternative to traditional image-based data augmentation methods.