Information and artifact in second-order statistics from computed tomography (ct) images

Abstract

In conventional computed tomography images, only the average CT number, which is a first-order statistical parameter, is used to characterize the tissues by giving an estimate of tissue density. Second-order statistical parameters such as the signal variance and cross-correlation function have also been used to obtain additional information to discriminate between certain tissues and lesions. However, the contribution of quantum noise to the signal variance and cross-correlation function creates, for the conventional CT patient dose, a background signal often larger than the signal containing the information about tissue structure. The misleading information, called “artifacts, ” in second-order image statistics caused by quantum noise is studied. The distribution of the contribution of quantum noise to second-order statistical parameters is object-dependent, space dependent, and “nonlocal.” The term “nonlocal” refers to the wide-ranging effects of nonlinear artifacts on the distribution of second-order statistical parameters. We present an analysis and examples of some of the artifacts encountered and also comment on how these artifacts affect the potential of these parameters for clinical use, such as tissue characterization. © 1979 SPIE.