Evaluation of statistical methods of image reconstruction through ROC analysis (emission tomography)

Abstract

Several statistical methods of image reconstruction are described and objectively compared through the use of receiver-operating-characteristic (ROC) analysis based on a specified detection task performed by a human observer. The simulated imaging system is a multiple-pinhole coded-aperture system for dynamic cardiac imaging, and the objects represent cross sections of the left ventricle at end systole. The task is detection of a profusion representing an akinetic wall segment. Thirteen different reconstruction algorithms are considered. Human observers perform the specified task on this set of reconstructions, and the results are analyzed through the use of ROC analysis. The results show that the methods that utilize the largest amount of (accurate) prior information tend to perform the best.