Estimating Tissue Microstructure with Undersampled Diffusion Data via Graph Convolutional Neural Networks.

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• Advanced diffusion models are crucial for studying brain disorders using diffusion MRI (dMRI).
• Current models require densely sampled q-space data, which is often not feasible in clinical environments.
• Existing deep learning methods often overlook the inherent structure within dMRI q-space data.

Purpose of the Study:

• To develop a deep learning framework that accurately estimates tissue microstructure from sparsely sampled dMRI data.
• To address the limitations of existing methods by incorporating q-space data structure.
• To improve the clinical applicability of diffusion MRI-based microstructure analysis.

Main Methods:

• Representing dMRI data as graphs to leverage structural information.
• Utilizing graph convolutional neural networks (GCNs) for microstructure estimation.
• Exploiting q-space angular neighboring information within the graph structure.

Main Results:

• The proposed graph-based GCN method significantly improves the accuracy of diffusion microstructural index estimation.
• Qualitative and quantitative experimental results demonstrate superior performance compared to state-of-the-art methods.
• The method shows promise for applications using the Baby Connectome Project dataset.

Conclusions:

• Graph representation and GCNs offer a powerful approach for analyzing dMRI data with sparse q-space sampling.
• This method enhances the estimation of tissue microstructure, potentially advancing the study of brain disorders.
• The findings suggest a more clinically viable pathway for advanced dMRI analysis.