Novel diffusion markers for tract-based spatial statistics and bundle-specific analysis – feasibility study in patients with schizophrenia
Stefan Sommer1, Matthias Kirschner2, Oliver Hager2,3, Stefan Kaiser2, Sebastian Kozerke1, Erich Seifritz2, and Philipp Stämpfli2

1Institute for Biomedical Engineering, ETH and University of Zurich, Zurich, Switzerland, 2Department of Psychiatry, Psychotherapy and Psychosomatics Psychiatric Hospital, University of Zurich, Zurich, Switzerland, 3Laboratory for Social and Neural Systems Research, Department of Economics, University of Zurich, Zurich, Switzerland

Synopsis

In the last few years, tract base spatial statistics (TBSS) become prominent tools for analyzing diffusion data in group studies. In this study, we introduce optimized high-resolution tract density (optTD) images based on tractograms. In order to avoid tractography bias, the tractograms are optimized using the convex-optimization modeling for microstructure informed tractography (COMMIT). The optTD were analyzed in a group of patients with schizophrenia and healthy controls using TBSS. Additionally, we have shown significant group differences comparing fiber weights derived from the COMMIT optimization. The introduced measures provide promising tools for investigating white-matter abnormalities in mental disorders.

Purpose

Diffusion imaging provides a useful tool to examine structural white matter changes in health and mental disorders. Tract-based spatial statistics1 (TBSS) is a popular tool for the evaluation of diffusion data in group studies. TBSS typically analyses the fractional anisotropy (FA) to find statistically significant group differences, whereby tractography is not explicitly performed. Calamante proposed a method to create super-resolution track-density (TD) maps2 based on diffusion fiber tractograms. Nevertheless, the resulting super resolution TD images seem unfeasible as a quantitative marker due to biases in the tractograms, introduced by seeding strategies or intrinsic tracking algorithm properties. Daducci introduced a Convex Optimization framework called COMMIT3, whereby a local forward model and a global optimization algorithm improve the fit of the tractogram with regard to the measured diffusion signal. This optimization leads to reduced false positive fibers and lower algorithm specific tracking bias in the tractogram. In this study, we compare TBSS results based on FA values and TD from optimized tractograms to explore the sensitivity and feasibility of these novel markers in patients with schizophrenia and healthy controls. Furthermore, explicit fiber tract weights of inter-cortical connections were compared between these two groups.

Methods

Datasets of 20 adult patients with schizophrenia and 25 healthy controls were acquired on a Philips Achieva 3T system with the following diffusion scan parameters: TR:6.64s, TE:53.6ms, FOV:240x240mm2, 50 slices, slice thickness:2.5mm, acquisition matrix:96x96, SENSE:2. Diffusion-weighted images were acquired along 32 directions distributed uniformly on a half-sphere with a b-value of 1000s/mm2 in addition to a b=0 scan. Additionally, 1mm isotropic T1-weighted structural images were recorded. For each dataset, the diffusion data was corrected for eddy-currents and subject motion using FSL4 and susceptibility induced distortions were reduced by applying the recently proposed INVERSION method using the T1-weighted scan5. Whole brain fiber tracking was performed using the constrained spherical deconvolution approach and the iFOD2 algorithm implemented in the MRtrix package6. Tissue priors were estimated from T1 images in FSL and used for the anatomically constrained tractography with SIFT inspired dynamic seeding. A total of 2.5 million fibers were generated per subject. Afterwards, each tractogram was optimized with the COMMIT framework using an intra-cellular stick model and two isotropic compartments (CSF, gray matter). The derived fiber weights from COMMIT were used to calculate an optimized high-resolution (1.25mm isotropic) track density for each subject. TBSS analysis was performed with the FA images and the optimized track density (optTD) maps. For the optTD maps, the skeleton and transformation parameters resulting from the TBSS analysis of the FA images were applied. Results were corrected for multiple comparison (P<.05). Additionally, a ROI covering the Corpus Callosum was placed in the sagittal midplane to evaluate inter-hemispherical fiber bundles.

Results

The TBSS analysis of both scalar maps (FA and optTD) show clusters of decreased values in patients but no significant regions with increased FA or optTD could be found. Significant FA clusters are depicted in Figure 1. Almost all significant voxels in the optTD analysis contribute to inter-hemispherical white-matter structures (Figure 2). COMMIT fiber weights for all fibers passing through the Corpus Callosum ROI were summed up and compared between the two groups. A statistically significant reduction in sum of Corpus Callosum fiber weights was found in patients compared to healthy controls (P<.05). Additionally, the fit error in the Corpus Callosum ROI did not show any significant difference between the two groups. Figure 3 shows segmented fibers of a single subject colored with the optTD (left) and a comparison of fiber weights in both groups (right).

Discussion

The introduced optTD is partially correlated with the FA, especially in single fiber voxels, but even there, it might help to distinguish between fiber dispersion and connectivity strength which both affect the FA. Additionally, in crossing situations, the COMMIT fiber weights could help to identify the altered fiber bundle due to its fiber specificity. Nevertheless, it is crucial to verify the quality of fiber-derived measures in order to exclude error sources such as tractography bias or model fitting errors. Therefore, the fitting-error resulting from the COMMIT optimization was evaluated in the Corpus Callosum ROI and no significant group difference was found.

Conclusion

We showed a high sensitivity in the newly introduced markers, the optimized TD scalar map and the fiber-specific optimization weights. Furthermore, our results of altered white matter integrity in the corpus callosum in patients with schizophrenia are consistent with data from previous diffusion magnetic imaging studies7. High-resolution maps derived from optimized whole brain tractograms and fiber-weights are promising novel markers investigating white matter abnormalities in mental disorders.

Acknowledgements

No acknowledgement found.

References

1. Smith, S., et al. Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. NeuroImage, 2006:31;1487-505.

2. Calamante F., et al. Track-density imaging (TDI): Super-resolution white matter imaging using whole-brain track-density mapping. NeuroImage, 2010;53:1233-43.

3. Daducci, A., et al. COMMIT: Convex Optimization Modeling for Microstructure Informed Tractography. IEEE, 2015:34;246-57.

4. Jenkinson, M., et al. Improved optimization for the robust and accurate linear registration and motion correction of brain images. Neuroimage, 2002:17;825–41.

5. Bhushan C., et al. Co-registration and distortion correction of diffusion and anatomical images based on inverse contrast normalization. Neuroimage, 2015:115;269-80.

6. Tournier, J-D, et al. https://github.com/MRtrix3/mrtrix3 2012.

7. Wheeler AL, et al. A review of structural neuroimaging in schizophrenia: from connectivity to connectomics. Front Hum Neurosci 2014:8;653

Figures

Figure 1: Two different views of the FA based TBSS results are shown in a) and b). Red voxels represent significantly lower FA in patients compared to healthy controls (P<.05).

Figure 2: TBSS results derived from the optTD maps are shown in a) and b), whereas the views are identically centered as in Figure 1. Red voxels represent significantly lower optTD in patients compared to healthy controls (P<.05).

Figure 3: The left side shows a single subject tractogram filtered with the Corpus Callosum ROI (blue). The color value is given by the underlying optTD map, whereby red is high density and yellow low density respectively. On the right side, the group comparison of the fiber weights is shown.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3054