Parameter Estimation Error Dependency on the Acquisition Protocol in Diffusion Kurtosis Imaging.

Nima Gilani,Glyn Johnson,Paul N Malcolm

doi:10.1007/s00723-016-0829-x

Abstract

Mono-exponential kurtosis model is routinely fitted on diffusion weighted, magnetic resonance imaging data to describe non-Gaussian diffusion. Here, the purpose was to optimize acquisitions for this model to minimize the errors in estimating diffusion coefficient and kurtosis. Similar to a previous study, covariance matrix calculations were used, and coefficients of variation in estimating each parameter of this model were calculated. The acquisition parameter, b values, varied in discrete grids to find the optimum ones that minimize the coefficient of variation in estimating the two non-Gaussian parameters. Also, the effect of variation of the target values on the optimized values was investigated. Additionally, the results were benchmarked with Monte Carlo noise simulations. Simple correlations were found between the optimized b values and target values of diffusion and kurtosis. For small target values of the two parameters, there is higher chance of having significant errors; this is caused by maximum b value limits imposed by the scanner than the mathematical bounds. The results here, cover a wide range of parameters D and K so that they could be used in many directionally averaged diffusion weighted cases such as head and neck, prostate, etc.

Highlights

Quantitative diffusion MRI has proved useful in characterizing tumours in a number of different cancers [1, 2], by estimating apparent diffusion coefficient of water molecules assuming the diffusion is Gaussian inside the organ
For small target values of the two parameters, there is higher chance of having significant errors; this is caused by maximum b value limits imposed by the scanner than the mathematical bounds
As observed in the table, at least one maximum b value is present in the optimized acquisitions

Summary

Introduction

Quantitative diffusion MRI has proved useful in characterizing tumours in a number of different cancers [1, 2], by estimating apparent diffusion coefficient of water molecules assuming the diffusion is Gaussian inside the organ. The relationship between the signal (S) and acquisition b values is not generally Gaussian because of compartmentalization, hindrance, and restriction effects on diffusion [3, 4], or generally complexity of diffusion. An additional term kurtosis (K) is added to the Gaussian model to describe this non-Gaussian behaviour [3, 4]:. The description of Eq (1) for diffusion has found applications in imaging breast cancer [6], head and neck [7, 8], prostate [9], etc, because of better description of complex non-Gaussian diffusion and better fitting results [10]. Better fits does not mean that the model is better reflective of biophysical changes to a certain disease, because there might be substantial fitting errors associated with its parameters as will be shown in this study

Methods

Results

Discussion

Conclusion