Computationally efficient 4D spectral-spatial EPR imaging.

Four-dimensional spectral-spatial imaging (4D SSI) enables noninvasive mapping of spin probes and their microenvironments. Despite its demonstrated utility, 4D SSI remains constrained by substantial computational demands, including large data volumes, the iterative nature of reconstruction algorithms, and significant requirements for memory and computational resources. These resource demands scale cubically with the size of the imaged object. To address these limitations, a set of computational strategies has been developed to improve reconstruction efficiency without compromising image fidelity. These include the use of filtered back projection (FBP) to generate an initial spin concentration map, which serves both as an initial guess for further iterations and as a mask to exclude non-signal voxels. Eliminating these empty voxels significantly reduces the problem size, thereby lowering memory usage and computation time. Additional acceleration is achieved by transforming the 4D reconstruction into a reduced 2D problem, minimizing redundant computation through precomputed values, and employing a compact look-up table for spectral fitting. The resulting workflow, implemented in MATLAB with performance-critical routines compiled as C-based MEX functions, achieves iteration times as low as one minute. Numerical phantom simulations and experimental data from physical phantoms confirm that convergence is substantially improved by excluding non-signal voxels. Among all evaluated approaches, the FBP-based masking of non-signal voxels and the use of a lookup table proved most effective in accelerating algorithm convergence. These improvements enable scalable and computationally efficient 4D SSI suitable for high-resolution, larger-animal preclinical studies and future clinical imaging applications.