SamRobNODDI:q-space sampling-augmented continuous representation learning for robust and generalized NODDI.

Area of Science:

• Neuroimaging
• Diffusion Magnetic Resonance Imaging (dMRI)
• Computational Neuroscience

Background:

• Neurite orientation dispersion and density imaging (NODDI) is crucial for understanding neurological diseases.
• Current deep learning models for NODDI lack robustness due to strict requirements on diffusion gradient directions.

Purpose of the Study:

• To develop a robust and generalized NODDI estimation method adaptable to varying diffusion gradient directions.
• To address the limitations of existing deep learning models in NODDI parameter estimation.

Main Methods:

• Proposed SamRobNODDI, a q-space sampling augmentation-based continuous representation learning framework.
• Introduced a sampling consistency loss to ensure stable outputs across different sampling schemes.
• Designed a flexible framework applicable to various backbone networks.

Main Results:

• SamRobNODDI demonstrated superior performance, robustness, and generalization compared to seven state-of-the-art methods across 19 q-space sampling schemes.
• Extensive validation confirmed effectiveness under diverse training/testing sampling schemes, rates, and network backbones.

Conclusions:

• SamRobNODDI offers significant improvements in performance and flexibility for NODDI estimation.
• The framework enhances robustness against variations in q-space sampling, crucial for clinical applications.