Deblurring and Noise Suppression in Spatial EPR Imaging

Electron paramagnetic resonance (EPR) imaging has been used to map in vivo spatial distributions of endogenous and introduced free radicals. Because of the appreciable spectral line widths of the radicals and the finite magnitude of the magnetic field gradients that encode spatial information, it is necessary to perform deblurring and noise suppression to achieve images with high spatial resolution. Recently, a regularized inverse-filtering technique for suppressing noise and blurring was described and evaluated for use with 3D single photon emission computed tomography. This filtering technique uses an a priori random image field (RIF) to specify the expected signal region and degree of smoothness in the image. Here, we generalize this technique for application to spatial EPR imaging, including a realistic noise model and experimentally measured blurring function. A set of experimental spectra for a 2D spatial image was acquired. Images were reconstructed from the original spectra and again after the implementation of a previously described deconvolution technique or the generalized RIF technique. This was repeated after addition of Gaussian distributed Brownian noise. The resulting images and profiles through the signal regions were quantitatively compared. Without the addition of noise spatial resolution is markedly improved by both filters. With added noise, application of the deconvolution technique resulted in an image with suppressed noise, but compromised spatial resolution. The RIF filtering technique suppressed image noise, to a lesser degree, and showed improvement in spatial resolution.