Accelerated, Physics-Inspired Inference of Skeletal Muscle Microstructure From Diffusion-Weighted MRI.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Biology

Background:

• Skeletal muscle health is vital for overall well-being.
• Current assessment methods overlook crucial muscle microstructural variations.
• Understanding muscle microstructure is key to function and health.

Purpose of the Study:

• To develop a non-invasive framework for estimating skeletal muscle microstructural organization.
• To utilize machine learning and diffusion-weighted MRI (dMRI) for microstructure analysis.
• To provide uncertainty-aware estimations of muscle microstructure.

Main Methods:

• Developed a physics-inspired, machine learning-based framework for dMRI analysis.
• Created a polynomial meta-model to approximate dMRI physics simulations.
• Implemented a Gaussian Process (GP) model for voxel-wise microstructure estimation with confidence intervals.
• Validated a reduced-acquisition GP model and its estimations via histology.

Main Results:

• The GP model accurately estimates microstructural parameters from dMRI data.
• Fiber diameter, intracellular diffusion, and membrane permeability were well-estimated even with noise.
• A reduced dMRI acquisition protocol maintained estimation accuracy.
• Histology validated the GP model's estimation of fiber diameter and volume fraction.

Conclusions:

• The proposed framework offers a promising non-invasive tool for assessing skeletal muscle health.
• Machine learning applied to dMRI can effectively characterize muscle microstructure.
• The uncertainty-aware GP model provides reliable microstructural insights.