Deep learning-based free-water correction for single-shell diffusion MRI | JoVE Visualize

Area of Science:

Neuroimaging
Biomedical Engineering
Computational Neuroscience

Background:

Free-water elimination (FWE) modeling in diffusion MRI is vital for accurate diffusion property estimation.
Partial volume effects from free water, especially at white matter-cerebrospinal fluid interfaces, bias diffusion measurements.
Existing FWE models are often ill-posed for common single-shell dMRI data.

Purpose of the Study:

To develop and evaluate a deep-learning framework for mapping and correcting free-water partial volume contamination in diffusion MRI.
To enable accurate FWE modeling using data-driven techniques across various dMRI acquisition schemes, including single-shell data.
To improve the reliability and clarity of diffusion MRI analyses in diverse datasets.

Main Methods:

A novel deep-learning framework was proposed to infer and correct free-water volumes in diffusion-weighted imaging (DWI).
The method was applied to single-shell and multi-shell dMRI datasets, including the Human Connectome Project (HCP) Young Adults and Aging datasets, and Brain Tumor Connectomics (BTC) data.
Model generalizability was assessed through fine-tuning and b-value re-mapping for new datasets.

Main Results:

The deep-learning approach yielded more plausible free-water estimations compared to existing single-shell methods.
The method demonstrated improved consistency in diffusion property estimation between scan/rescan sessions.
Enhanced accuracy in neural pathway identification and clearer visualization of white matter tracts were observed.

Conclusions:

The proposed deep-learning framework effectively corrects free-water partial volume effects in dMRI, even with single-shell acquisitions.
The method offers a generalizable solution for improving the accuracy and robustness of diffusion MRI analyses across different datasets and acquisition parameters.
This approach holds significant potential for advancing neuroimaging research and clinical applications by providing more reliable brain microstructure information.