FOD-Net

FOD-Net enhances the angular resolution of fiber orientation distribution (FOD) images from diffusion magnetic resonance imaging (dMRI) using deep learning to improve tractography and structural connectome reconstruction from clinical-grade acquisitions.

Key Features:

• Angular Super-Resolution: Uses deep learning to perform angular super-resolution of FOD images derived from standard clinical dMRI to approximate the quality of high-resolution research acquisitions that often use multi-shell protocols.
• Training and Validation: Trained on high-quality Human Connectome Project (HCP) data and validated on a local clinical 3.0T scanner and a public multicenter-multiscanner dataset.
• Improved Tractography: Produces FODs that reduce spurious connections and bridge missing connections, yielding more accurate tractography and structural connectome reconstructions.
• Clinical Applicability: Enables generation of high-quality tractography and connectome analyses from existing clinical MRI protocols without requiring advanced dMRI acquisition schemes.

Scientific Applications:

• Neuroscience Research: Supports more precise studies of brain connectivity by enabling higher-quality FODs and tractography from clinical-grade dMRI.
• Clinical Neuroimaging: Facilitates improved structural connectome reconstruction that may inform diagnostic assessments and therapeutic strategy development for neurological conditions.

Methodology:

FOD-Net applies deep learning-based angular super-resolution to FOD images derived from dMRI and was trained on Human Connectome Project data with validation on a local clinical 3.0T scanner and a public multicenter-multiscanner dataset.