Motivation: The state-of-the-art diffusion MRI harmonization technique, LinearRISH, is based on signal representation on spherical harmonics, where the sensitivity to higher harmonic order terms can vary across scanners and protocols. Goal(s): To explore if we can achieve better harmonization by exploiting the fact that DKI well describes the signal at lower b-values. Approach: We represent the diffusion and kurtosis tensors, instead of the MRI signal, on a spherical harmonics basis and do scaling akin to LinearRISH. We dubbed this method LinearRICE and compared it with LinearRISH in a multi-vendor traveling subject dataset. Results: LinearRICE reduced the inter-site variabilities better than LinearRISH. Impact: We proposed a new method for retrospective harmonization of DKI-type diffusion MRI data. The proposed method outperformed the current state-of-the-art technique in a multi-vendor traveling subject dataset and can be useful for multi-institutional studies.
Kamiya et al. (Tue,) studied this question.