Super-resolution mapping of anisotropic tissue structure with diffusion MRI and deep learning

Area of Science:

• Neuroimaging
• Biophysics
• Medical Physics

Background:

• Diffusion MRI is crucial for detecting CNS diseases by assessing tissue microstructure.
• Quantitative mapping of microstructural parameters like fODF is vital for noninvasive white matter tractography.
• Current diffusion MRI methods face limitations in acquisition time and spatial resolution.

Purpose of the Study:

• To develop a deep-learning-based super-resolution method for diffusion MRI data, specifically enhancing fiber orientation distribution functions (fODFs).
• To evaluate the proposed method's performance against traditional interpolation techniques and assess its accuracy using earth mover's distance.

Main Methods:

• A deep learning approach was developed to increase the spatial resolution of fODFs derived from constrained spherical deconvolution.
• The method was evaluated using high-quality diffusion MRI data from the Human Connectome Project.
• Accuracy was assessed using the earth mover's distance metric, particularly at low signal-to-noise ratios.

Main Results:

• The deep learning method successfully generated upsampled fODFs with higher correspondence to ground truth high-resolution data compared to spline interpolation.
• The super-resolution method provided more accurate fODF estimates than standard methods when using data with 8 times smaller voxel volume (lower SNR).

Conclusions:

• Deep learning offers a powerful tool for enhancing the spatial resolution of diffusion MRI-derived fODFs.
• This technique can improve the accuracy of white matter tractography, especially under challenging low signal-to-noise conditions.
• The developed method represents a significant advancement for quantitative microstructural analysis in neuroimaging.