Application of Hamiltonian Monte Carlo Method to Metal Artifact Quantitative Evaluation in Computed Tomography (CT).

Ryosuke Kasai, Kenji Yamada

Nihon Hoshasen Gijutsu Gakkai zasshi 2017

filter terms:

In this study, we propose an evaluation method for Bayesian estimation of Gumbel distribution parameters by the Hamiltonian Monte Carlo method (HMC method), with changing the pixel size of the CT image to investigate streak artifacts, without using a significant difference test. Placed a titanium endcap in the center of the CT dose index (CTDI) measurement phantom and got the CT image by changing the display-field of view (D-FOV) to S, M, L, LL. We compared Gumbel distribution parameters with conventional estimation method and Bayesian estimation method using HMC method. In addition, we evaluated streak artifacts by Bayesian statistical analysis. The difference in streak artifact between D-FOV was more than 90% except between D-FOV M and L. The effect of streak artifacts is small as the pixel size was small. By using the HMC method, we can estimate the Gumbel distribution parameters accurately and objectively, and quantitatively evaluated that the streak artifacts differ in pixel size using Bayesian statistical analysis.

Citation

Ryosuke Kasai, Kenji Yamada. Application of Hamiltonian Monte Carlo Method to Metal Artifact Quantitative Evaluation in Computed Tomography (CT). Nihon Hoshasen Gijutsu Gakkai zasshi. 2017;73(8):654-663