A01-KB111 Magnetic Resonance q-Space Imaging using Generating Function and Bayesian Inference

From Multidisciplinary Computional Anatomy
Jump to: navigation, search

Member

  • Primary Investigator
    Eizou Umezawa (Fujita Health University Graduate School of Health Sciences, Associate Professor)
  • Co-Researcher
    Masutani Yoshitaka (Hiroshima City University Graduate School of Information Sciences, Professor)
    Masayuki Yamada (Fujita Health University Graduate School of Health Sciences, Professor)
    Kazuhiro Murayama (Fujita Health University School of Medicine, Senior Assistant Professor)
    Masato Abe (Fujita Health University Graduate School of Health Sciences, Professor)
    Takayuki Enari (Nihon University College of Science and Technology, Research Associate)

Overview

Q-space imaging (QSI) is a diffusion MRI method to obtain medical images using statistical properties of the Brownian motion of water molecules in living organisms. Diffusional kurtosis imaging (DKI) is a light version of QSI that is specialized to obtain the kurtosis of the diffusion displacement and can be implemented with few MRI data. Because the kurtosis varies depending on microstructures of tissues, DKI is expected to provide a novel tool for diagnoses and biological functional measurements; however, the accuracy of the kurtosis estimation is not sufficient yet. The kurtosis obtained by DKI contains large systematic errors in general. The MR signal intensity of QSI is corresponds to the characteristic function (CF) of the diffusion displacement and the logarithm of the signal is the cumulant generating function (CGF). In DKI analysis, CF or CGF is expressed as a series of b value (a parameter of MRI equipments) and used for fitting of the measured MR signals. In conventional DKI, the series is truncated after the second order term of b value, which contributes to the systematic errors of the kurtosis estimation. In our study we consider the method in which the higher order terms of CF or CGF are taken into account. In this method the overfittings arising from the increased parameters may be encountered. To address the problem we develop a Bayesian approach and realize a robust and precise method for measuring the descriptive statistics values of the diffusion displacement.

Project Design

A01-KB111-E.png