Fast multi-compartment Microstructure Fingerprinting in brain white matter

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• Diffusion MRI is crucial for mapping white matter microstructure.
• Estimating microstructural features, especially in crossing fascicles, is computationally intensive.
• Microstructure Fingerprinting (MF) extends Magnetic Resonance Fingerprinting (MRF) for diffusion MRI but requires significant computation.

Purpose of the Study:

• To accelerate the estimation of microstructural features in white matter, particularly for complex crossing fascicles.
• To improve the efficiency of the multi-dictionary matching problem central to Microstructure Fingerprinting.
• To enable faster quantitative analysis of brain white matter in vivo.

Main Methods:

• Proposed two deep neural network (DNN) based acceleration methods for Microstructure Fingerprinting.
• Method 1: Utilized efficient sparse optimization and a feed-forward DNN to address combinatorial complexity.
• Method 2: Employed a feed-forward DNN using spherical harmonics representation of diffusion-weighted MRI (DW-MRI) signals as input.

Main Results:

• Both DNN methods significantly accelerated the estimation process.
• Method 1 offered high interpretability.
• Method 2 achieved a greater speedup factor, with several orders of magnitude improvement.
• Accurate results were validated on in vivo brain data.

Conclusions:

• The developed DNN-based methods offer substantial speedup for Microstructure Fingerprinting.
• These techniques hold promise for rapid quantitative estimation of white matter microstructural features.
• The findings are particularly relevant for analyzing complex white matter configurations in diffusion MRI studies.