Axisymmetric diffusion kurtosis imaging with Rician bias correction: A simulation study

Abstract

Purpose: To compare the estimation accuracy of axisymmetric diffusion kurtosis imaging (DKI) and standard DKI in combination with Rician bias correction (RBC) under the influence of noise. Methods: Axisymmetric DKI is more robust against noise-induced variation in the measured signal than standard DKI because of its reduced parameter space. However, its susceptibility to Rician noise bias at low signal-to-noise ratios (SNRs) is unknown. Here, we investigate two main questions: first, does Rician bias correction improve estimation accuracy of axisymmetric DKI?; second, is the estimation accuracy of axisymmetric DKI increased compared to standard DKI? Estimation accuracy was investigated on the five axisymmetric DKI tensor metrics (AxTM): the parallel and perpendicular diffusivity and kurtosis and the mean kurtosis, using a simulation study based on synthetic and in-vivo data. Results: We found that RBC was most effective for increasing accuracy of the parallel AxTM in highly to moderately aligned white matter. For the perpendicular AxTM, axisymmetric DKI without RBC performed slightly better than with RBC. However, the combination of axisymmetric DKI with RBC was the overall best performing algorithm across all five AxTM and the axisymmetric DKI framework itself substantially improved accuracy in tissues with low fiber alignment. Conclusion: The combination of axisymmetric DKI with RBC facilitates accurate DKI parameter estimation at unprecedented low SNRs (approx 15), possibly making it a valuable tool for neuroscience and clinical research studies where scan time is a limited resource. The tools used in this paper are publicly available in the open-source ACID toolbox for SPM.