Diffusion Kurtosis Imaging ( DKI ) Reconstruction-Linear or Non-Linear ?

Introduction: Diffusion Kurtosis Imaging (DKI) which measures the non-Gaussian behavior in diffusion has shown great potential as being a more sensitive marker in characterizing neural tissues compared to conventional DTI. Processing for DKI originally used nonlinear least squares (NLS) methods which were inefficient and processing time of one hour on a fast computer was not unusual . Later, fast DKI fitting method (fDKI) was proposed which reduced the image reconstruction time to seconds by using an explicit formula to calculate Dapp and Kapp when only 2 non-zero b-values were acquired, with the assumption that diffusion directions remained invariant between the different diffusion sensitivities. Similar to DTI, the DKI model could also be linearized and fit for both the Diffusion Tensor and Kurtosis Tensor directly through linear equations as described by Tabesh et al. 6 with reconstruction times comparable to DTI reconstruction. Although the NLS approach may provide more accurate results, real-time reconstruction is critical in the clinical setting so linear methods are highly preferred. In this study DKI derived maps from both linear and non-linear least squares approaches from a clinically relevant short imaging protocol (using 2 b-values) were compared with ‘gold standard’ dataset that was obtained using five different diffusion sensitivies (5 b-values) to assess their performance.