A simulation-driven supervised learning framework to estimate brain microstructure using diffusion MRI.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biophysics

Background:

• Diffusion magnetic resonance imaging (dMRI) is crucial for non-invasively probing brain microstructure.
• Estimating tissue microstructure parameters from dMRI signals is challenging due to complex signal variations.
• Current methods often require extensive validation and may lack diffusion time independence.

Purpose of the Study:

• To introduce a novel framework for training supervised learning models on synthetic dMRI data.
• To generate a comprehensive synthetic dataset for microstructure estimation.
• To validate the framework's ability to estimate key microstructural parameters with diffusion time independence.

Main Methods:

• Generated synthetic dMRI signals from over 1,000 digital neuronal reconstructions using an optimized simulator.
• Created a large synthetic dataset (1.45 million voxels) with 40 microstructure parameters by combining simulated neuron signals and free diffusion.
• Trained multilayer perceptrons (MLPs) on the synthetic data to estimate volume and area fractions of cellular components.

Main Results:

• Trained MLPs demonstrated satisfactory performance on synthetic test data.
• The framework produced promising in-vivo parameter maps on the MGH Connectome Diffusion Microstructure Dataset (CDMD).
• Estimated volume fractions exhibited low dependence on diffusion time, a key characteristic for quantitative imaging.

Conclusions:

• The proposed framework offers a robust method for microstructure estimation using supervised learning on synthetic dMRI data.
• The generated synthetic dataset and neuron models are valuable resources for validating dMRI microstructure mapping techniques.
• This approach advances quantitative microstructure imaging by providing diffusion time-independent parameter estimates.