Objective comparisons of statistical methods of image reconstruction

Abstract

In medical image reconstruction, one always has in mind the ultimate goal of obtaining the reconstruction that will yield the most accurate diagnosis given a specified task. The visual appearance of any reconstruction will depend on several factors including the method used to obtain the reconstruction. Statistical methods have been the topic of much recent interest in emission tomography. virtually all statistical methods attempt to find the reconstruction that agrees with the collected data to a degree that does not contradict any prior information assumed about the object. The prior information, and the way that one incorporates this information, is what typically distinguishes one statistical method of reconstruction from another. A plethora of methods exists and because of this, a natural question arises: which method should be used? While many methods have been investigated independent of others, an objective comparison of any of these methods is something that has not yet been sufficiently explored. In this presentation, we make objective comparisons of several statistical methods of image reconstruction. To do this, we define a specific task and use different reconstruction methods on a given class of objects. Based on the specified task, we make objective comparisons of the different methods of reconstruction through psychophysical studies and receiver operating characteristic analysis.