Microscopic Propagator Imaging (MPI) with Diffusion MRI

TL;DR

Microscopic Propagator Imaging (MPI) addresses the limitation that conventional EAP indices are confounded by mesoscopic orientation dispersion. MPI isolates the microscopic propagator indices by exploiting ratios of l-band power spectra that are invariant to the ODF and by learning a mapping from synthetic kernel signals to these indices using a random forest. The study demonstrates: (i) theoretical grounding linking the kernel to observable SH coefficients and their invariants; (ii) a full pipeline from synthetic data to multi-shell dMRI measurements; (iii) empirical validation on synthetic WM/GM/CSF datasets and real Human Connectome Project data, revealing good agreement for most indices and highlighting limitations for $NG_{\parallel}$. The work argues that MPI provides more direct, microstructure-specific metrics with potential diagnostic relevance.

Abstract

We propose Microscopic Propagator Imaging (MPI) as a novel method to retrieve the indices of the microscopic propagator which is the probability density function of water displacements due to diffusion within the nervous tissue microstructures. Unlike the Ensemble Average Propagator indices or the Diffusion Tensor Imaging metrics, MPI indices are independent from the mesoscopic organization of the tissue such as the presence of multiple axonal bundle directions and orientation dispersion. As a consequence, MPI indices are more specific to the volumes, sizes, and types of microstructures, like axons and cells, that are present in the tissue. Thus, changes in MPI indices can be more directly linked to alterations in the presence and integrity of microstructures themselves. The methodology behind MPI is rooted on zonal modeling of spherical harmonics, signal simulation, and machine learning regression, and is demonstrated on both synthetic and Human Diffusion MRI data.

Figures (6)

• Figure 1: The framework implementing the concept of microscopic propagator imaging for what concerns the generation of (signal) features and (kernel) labels to be learned.
• Figure 2: Probed SNR distributions for adding noise on the training data.
• Figure 3: Error distributions and performance metrics across the white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) synthetic testing dataset.
• Figure 4: Mean absolute SHAP values representing the features importance for the selected trained RF regressor. Results were computed by randomly extracting 10000 samples from each of the WM, GM, and CSF testing datasets. A higher value indicates a higher importance of the feature.
• Figure 5: Maps of propagator indices obtained with MAPL implementation of MAP-MRI and with MPI for four subjects of the HCP dataset. MAPL's results were clipped between 0 and the 99$^{\text{th}}$ percentile to remove outliers. The color bars' upper limits were adjusted by considering the highest of values between MAPL and MPI. In the case of NG$_{\parallel}$ two different color bar ranges are used ([0,0.6] for MAPL and [0,0.1] for MPI) to adjust for the markedly different value ranges.