Quantitative Comparisons of Choices of Prior Information in Image Reconstruction

Abstract

Medical image reconstruction is fraught with problems that are a result of noisy and incomplete data. The incomplete data give rise to null functions that are associated with the imaging operator, thus yielding an infinite number of solutions that fit the data equally well. Noise in the data often will lead to very rough reconstructions, which can be inconsistent with previous experience. The use of prior information can sometimes be introduced to help alleviate the aforementioned problems. If one knows that an object (or class of objects) possesses certain characteristics, then the reconstructions should possess the same characteristics.