Auto-Regressive Discrete Acquisition Points Transformation for Diffusion Weighted MRI Data.

Emma Metcalfe-Smith,Niloufar Zarinabad,Emma M Meeus,Andrew C Peet,Jan Novak,Hamid Dehghani

doi:10.1109/tbme.2019.2893523

Abstract

A new method for fitting diffusion-weighted magnetic resonance imaging (DW-MRI) data composed of an unknown number of multi-exponential components is presented and evaluated. The auto-regressive discrete acquisition points transformation (ADAPT) method is an adaption of the auto-regressive moving average system, which allows for the modeling of multi-exponential data and enables the estimation of the number of exponential components without prior assumptions. ADAPT was evaluated on simulated DW-MRI data. The optimum ADAPT fit was then applied to human brain DWI data and the correlation between the ADAPT coefficients and the parameters of the commonly used bi-exponential intravoxel incoherent motion (IVIM) method were investigated. The ADAPT method can correctly identify the number of components and model the exponential data. The ADAPT coefficients were found to have strong correlations with the IVIM parameters. ADAPT(1,1)-β0 correlated with IVIM-D: ρ = 0.708, P < 0.001. ADAPT(1,1)-α1 correlated with IVIM-f: ρ = 0.667, P < 0.001. ADAPT(1,1)-β1 correlated with IVIM-D*: ρ = 0.741, P < 0.001). ADAPT provides a method that can identify the number of exponential components in DWI data without prior assumptions, and determine potential complex diffusion biomarkers. ADAPT has the potential to provide a generalized fitting method for discrete multi-exponential data, and determine meaningful coefficients without prior information.

Highlights

M ULTI-EXPONENTIAL fitting is a challenging task for diffusion-weighted magnetic resonance imaging (DWMRI) data, where there are a limited number of data points and the number of components within the diffusion signal is unknown
1) Selection of Optimum Fit: A range of Auto-regressive Discrete Acquisition Points Transformation (ADAPT) orders were fitted to the simulated bi-exponential signal (SNR ࣈ 50) with varying Intravoxel Incoherent Motion (IVIM) parameters (Fig. 1)
The ADAPT method was applied to the bi-exponential signals and the optimum fit was selected by choosing the method with the lowest AICc

Summary

Introduction

M ULTI-EXPONENTIAL fitting is a challenging task for diffusion-weighted magnetic resonance imaging (DWMRI) data, where there are a limited number of data points and the number of components within the diffusion signal is unknown. Both theoretical and experimental studies have suggested that the water diffusion in tissue is characterized by multiexponential behavior [1], [2], [3]. If an exact reverse gradient is subsequently applied, particles that have moved, via diffusion, will experience at net phase shift and the detected signal intensity will attenuate.

Objectives

Methods

Results

Discussion

Conclusion