Model hypotheses and task functions investigated in this work. (a) Sphere detection has sphere absent and sphere present, (b) shape discrimination has a sphere and a cylinder, and (c) texture discrimination has a smooth and crenellated cylinder as the two hypotheses, respectively. Task functions adapted from Ref. 17.
(a) Image of the flat-panel detector and the task phantom used to acquire DE images of the various stimuli corresponding to the different imaging tasks. (b) Radiographic projection of various stimuli. Experimental setup and phantom adapted from Ref. 17.
Sphere detection performance as a function of dose allocation in DE imaging (i.e., the fraction of dose allocated to the low-energy projection). Data points represent measurements from human observers, and curves represent model observers. (a–c) Soft-tissue and (d–f) bone image. Results are shown for the three decomposition algorithms (SLS, SSH, and ACNR).
(a–c) Shape and (d–f) texture discrimination performance as a function of dose allocation for the soft-tissue image. Data points represent measurements from human observers, and curves represent model observers. Results are shown for the three decomposition algorithms (SLS, SSH, and ACNR).
Experimental and theoretical observer performance for the shape and texture discrimination tasks evaluated as a function of decomposition parameters in the SSH and ACNR algorithms.
Illustration of basic agreement between Fourier-based descriptions of NEQ, task, and observer model with human observer performance in the discrimination of a textured and smooth disk. (a–c) Plot of NEQ and task function (weighted by the eye filter) for three values of the parameter employed in the SSH decomposition algorithm. Corresponding images of the (d–f) crenellated and (g–i) smooth cylinder. Since the value of determines the zero in the NEQ, the value for which the zero coincides with frequencies of interest for this task causes the discriminate performance to plummets. Such is clearly observed in (b), where the value places the zero of the NEQ near the peak of the task function, and the ability to discriminate images (e) and (h) is lost.
Summary of low- and high-energy convolution filters for three DE decomposition algorithms (SLS, SSH, and ACNR), which are special cases of the general form in Eq. (1). denotes the tissue cancellation parameters, ideally given as the ratio of the effective linear attenuation coefficient of the cancelled material , is the Dirac delta function, denotes the noise cancellation parameters, denotes the tissue cancellation parameters for the complementary image, and and are low- and high-pass convolution filters, respectively.
Nominal values for the decomposition parameters associated with each of the decomposition techniques.
Number of cases in each study for a given observer. Each case was repeated five times (using different projections so that all cases were statistically independent).
Article metrics loading...
Full text loading...