Improved Myelin/Axonal/Extracellular Water Separation Using Adaptive Probabilistic Subspaces and T1W-Translated Spatial Priors

Motivation: Separating myelin, axonal, and extracellular water components from brain gradient-echo imaging data is desirable for characterizing many brain diseases but involves solving a challenging ill-conditioned signal decomposition problem. Goal(s): To develop an effective method to solve the underlying ill-conditioned problem. Approach: We solved the signal decomposition using an adaptive probabilistic subspace model incorporating spatial constraints. This method effectively compensates for practical perturbations by adapting subspace bases to individual imaging dataset and stabilizes decomposition by integrating spatiospectral priors from companion T1W images via deep translation. Results: Simulations and experimental results showed significantly improved maps of myelin/axonal/extracellular water over existing methods. Impact: This method may improve the practical utility of myelin/axonal/extracellular water fraction mapping. The integration of deep translation priors and adaptive spectral priors provides a promising framework for solving other ill-conditioned inverse problems.