Numerous studies have used diffusion tensor imaging (DTI) to investigate schizophrenia-related white matter microstructural deficits¹. DTI assumes a Gaussian distribution for the water molecule displacement. However, this assumption may not be valid in the complex biological tissues such as white matter. The ensemble average diffusion propagator (EAP) gives the probability of the average displacement of water molecules in a given voxel. Without making simplistic assumptions about the tissue, EAP can provide more information about diffusion properties of the tissue such as generalized kurtosis (GK, a measure of non-Gaussianity of water diffusion) and return-to-axis-probability (RTAP, related to cross-sectional area of the pore space in tissue). The generalized kurtosis measure we used in this work is derived directly from the EAP and is obtained by estimating the EAP from all the b-values and measurements. This is in contrast to standard mean kurtosis model used in the literature², where only information from b-values less than 2500 can be used. In this study, by estimating EAP using directional radial basis functions (RBF) we sought to investigate the white matter differences in scalar parameters derived from EAP between persons with schizophrenia and healthy controls.

Our dataset consisted of a young group (age<39) of healthy individuals (n=21; mean age=27.2; F/M:9/12) and schizophrenia patients (n=18; mean age=29.7; F/M:8/10) who underwent multi-shell diffusion MRI. Diffusion-weighted images were acquired using a 3 Tesla GE Discovery MR750 system equipped with an 8-channel head coil. A single shot dual-spin echo planar imaging sequence was used with three different b-values (1000, 3000, and 4500 s/mm²) along 30 noncollinear gradient directions with 5 interspersed b₀ images for each diffusion-weighted shell. The acquisition parameters were: TE/TR=108/12000 ms, resolution of 2×2×2 mm³, and 82 slices. Diffusion data preprocessing (eddy current correction and brain extraction) and diffusion tensor model fitting (only performed in b=1000 images) were conducted using FMRIB's Diffusion Toolbox (part of FSL; http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FDT). For each participant, EAP was estimated voxel-by-voxel using directional RBF, as implemented in RBF-Propagator MATLAB toolbox (https://github.com/LipengNing/RBF-Propagator)³. For the current study, GK and RTAP maps from the RBF-Propagator outputs were used for further analysis.

Fractional anisotropy (FA) maps were submitted to the tract-based spatial statistics (TBSS) pipeline⁴. Briefly, FA images were nonlinearly registered to a study-specific template (the most representative FA image) and then transformed to the MNI space. Average FA image across subjects was skeletonized. To generate skeletonized FA images for each subject, local FA maxima were projected to the average skeleton. Skeletonized images for other tensor model (mean diffusivity [MD] and radial diffusivity [RD]) and RBF model (GK and RTAP) parametric maps were created using the tbss_non_FA script. For each modality, permutation-based voxel-wise comparison was conducted between persons with schizophrenia and healthy participants using randomise with 20,000 permutations, while controlling for age and sex. To control for multiple comparisons using Bonferroni method, significance threshold was set to family-wise error corrected [FWE]-P <0.01 (0.05/5).

Persons with schizophrenia showed significantly lower (FWE-P<0.01) GK throughout the deep white matter (corpus callosum, corona radiata, internal capsule, external capsule) and superficial white matter structures (mainly involving frontotemporal areas as well as medial occipital lobe and precuneus; Figure 1; Figure 2). Schizophrenia patients tended to also have a lower RTAP in the posterior corpus callosum and frontotemporal white matter (peak FWE-P=0.013). Widespread differences were also observed in white matter RD (higher in schizophrenia) and FA (lower in schizophrenia), while MD was higher mainly in the frontal lobes and anterior corpus callosum among schizophrenia patients. Regions showing significant difference (FWE-P<0.01) exclusively in GK are also illustrated in Figure 1 (bottom row). At varying FWE-P thresholds, greater percentage of voxels demonstrated effects of diagnosis on GK than on other diffusion metrics (Figure 3).Our results suggest that white matter GK is more sensitive to differences between persons with schizophrenia and healthy controls than diffusion tensor model parameters (particularly in frontotemporal superficial white matter areas). These are in line with a recent study demonstrating similar deficits in mean kurtosis among people with schizophrenia⁵. Taken together, GK seems to be a robust marker for schizophrenia. Future studies are required to further examine the relationship of GK variations to clinical/cognitive deficit in schizophrenia. Finally, in light of recent advances in tissue clearing techniques and high-throughput histological imaging methods it may be possible to understand the neuropathological basis of GK deficits in schizophrenia.