Bayesian 2-D deconvolution: Effect of using spatially invariant ultrasoundpoint spread functions

Observed ultrasound images are degraded representations of the true tissue reflectance. The specular reflections at boundaries between regions of diff erent tissue types are blurred, and the diffuse scattering within homogenou s regions causes speckle because of the oscillating nature of the transmitt ed pulse. To reduce both blur and speckle, we have developed algorithms for the restoration of simulated and real ultrasound images based on Markov ra ndom field models and Bayesian statistical methods. The algorithm is summar ized here, although a more detailed description can be found in our compani on paper [1]. Because the point spread function (psf) is unknown, we invest igate the effects of using incorrect frequencies and sizes for the model ps f during the restoration process. First, we degrade the images either with a known simulated psf or a measured psf. Then, we use different psf shapes during restoration to study the robustness of the method. We found that sma ll variations in the parameters characterizing the psf, less than +/-25% ch ange in frequency, width, or length, still yielded satisfactory results. Wh en altering the psf more than this, the restorations were not acceptable. T he restorations were particularly sensitive to large increases in the resto ring psf frequency. Thus, 2-D Bayesian restoration using a fixed psf may yi eld acceptable results as long as the true variant psfs have not varied too much during imaging.