GAN-MAT: Generative adversarial network-based microstructural profile covariance analysis toolbox.

Area of Science:

• Neuroimaging
• Brain Microstructure Analysis
• Artificial Intelligence in Neuroscience

Background:

• Multimodal magnetic resonance imaging (MRI) offers valuable insights into brain structure and function.
• Estimating in vivo microstructure-sensitive proxies typically requires both T1- and T2-weighted MRI.
• Acquiring multiple MRI modalities poses challenges, particularly for patients with attention deficits.

Purpose of the Study:

• To develop a comprehensive framework for extracting multiple brain microstructural features using only T1-weighted MRI.
• To overcome the limitations of acquiring multiple MRI sequences in clinical settings.
• To enable advanced microstructural analysis in diverse patient populations.

Main Methods:

• Utilized a conditional generative adversarial network to synthesize T2-weighted MRI from T1-weighted MRI.
• Estimated intracortical covariance and moment features from cortical layer-wise microstructural profiles.
• Generated a microstructural gradient as a low-dimensional representation of intracortical microstructure.

Main Results:

• Synthesized T2-weighted MRI closely resembled actual images.
• The framework successfully reproduced key microstructural features.
• Validated the toolbox on independent datasets including healthy controls, episodic migraine, and autism spectrum disorder patients.

Conclusions:

• The developed toolbox offers a new paradigm for multimodal structural MRI analysis in neuroscience.
• This method facilitates the investigation of brain microstructure without requiring multiple MRI acquisitions.
• The openly accessible toolbox promotes wider adoption and research in neuroimaging.