High-Resolution Neural Network-Driven Mapping of Multiple Diffusion Metrics Leveraging Asymmetries in the Balanced SSFP Frequency Profile
Diffusion-weighted imaging (DWI) is widely used for clinical disease assessment as it allows tracking of microstructural diffusion processes and fiber trajectories in white matter (WM) based on the anisotropic motion of water protons. Quantitative measures derived from diffusion tensor imaging (DTI), such as mean, axial, and radial diffusivity (MD, AD, RD) or fractional anisotropy (FA), reflect changes in the tissue microenvironment, facilitating the assessment and monitoring of degenerative diseases such as multiple sclerosis, Parkinson's disease, Alzheimer's disease, or tissue changes after acute ischemic stroke. To assess the complex diffusion processes underlying brain tissue microstructure, it is expected that DWI greatly benefits from imaging at high submillimeter isotropic resolution. However, state-of-the-art diffusion-weighted spin-echo echo-planar-imaging (SE-EPI) is strongly affected by geometric distortions and T2* blurring due to its typically low bandwidth in the phase encoding direction, exacerbated at ultra-high field strengths due to larger B0 field variations causing a loss of the intrinsic spatial resolution. Despite promising correction methods, isotropic high-resolution DWI, which provides volumetric coverage of the whole brain within clinically relevant scan times, remains hardly feasible at ultra-high fields based on conventional methods.
It was reported that information about tissue anisotropies is entangled in the phase-cycled balanced SSFP (bSSFP) signal, manifest by frequency responses with different degrees of asymmetry depending on the underlying tissue type and the corresponding intravoxel frequency distribution. Recent findings suggested significant correlations of the bSSFP asymmetry index (AI) with DTI metrics, such as the component of the principal diffusion eigenvector parallel to B0, providing information about fiber tract orientation, or FA, reflecting the strength of tract directionality. This inherent sensitivity to tissue microstructure, combined with a mixed dependence on both T1 and T2, as well as the ability to enable distortion-free motion-robust volumetric imaging with high SNR efficiency in short scan times make phase-cycled bSSFP an interesting tool for multi-parametric mapping of various MR quantities.
In this work, we propose to utilize the rich information content about tissue microstructure entangled in bSSFP profile asymmetries to simultaneously estimate multiple diffusion measures by means of artificial neural networks (NNs). Concretely, the goal is to quantify the scalar diffusion metrics MD, FA, AD, RD as well as the spherical coordinates Φ (azimuth) and ϴ (inclination) of the principal diffusion eigenvector directly from 12-point 3D phase-cycled bSSFP input data acquired at high (3 T) and ultra-high (9.4 T) field strength in healthy volunteers.