3392

The spatial vs. angular resolution trade-off in diffusion MRI explored ex vivo at 9.4T
Giorgia Grisot1,2, Joseph M. Mandeville2, and Anastasia Yendiki2

1Harvard-MIT Health Sciences and Technology, Massachusetts Institute of Technology, Cambridge, MA, United States, 2Athinoula A. Martinos Center for Biomedical Imaging, MGH, Charlestown, MA, United States

Synopsis

The time constraints of in vivo diffusion MRI require a compromise to be made between spatial and angular resolution. Given the lack of ground truth on the configuration of human brain connections, determining the optimal operating point along this trade-off is still an open problem. We use high-SNR ex vivo human data at microscopic resolution to study the effect of spatial and angular resolution on dMRI tractography accuracy. Our findings show that voxel size has a much more dramatic effect on tractography reconstruction of challenging white matter configurations than the number of gradient directions.

Purpose

The time constraints of in vivo diffusion MRI (dMRI) require a compromise to be made between spatial resolution and angular resolution. The optimal operating point along this trade-off is not known, and it is likely dependent on the specific white-matter (WM) architecture of interest. The lack of ground truth on the configuration of WM bundles in the human brain makes it challenging to determine optimal acquisition parameters for reconstructing these bundles with dMRI tractography. Here, we take advantage of the high signal-to-noise ratio (SNR) that is feasible only ex-vivo with very long scans at ultra-high field strengths to acquire dMRI data on human brain samples at microscopic resolutions, allowing us to investigate the limits of dMRI performance. We present results from a human WM area with multiple crossing and splitting bundles that we have previously found to cause errors in tractography when compared to chemical tracing in non-human primates1,2. We image it at 9.4T and show the effects that reducing the spatial or angular resolution has on tractography accuracy.

Methods

We identify a critical area with crossing and splitting pathways (Figure 1), which we have found to cause errors in prior studies comparing dMRI tractography and chemical tracing in macaques2. We image a 3x2x2cm block of this area from an ex vivo human brain (Figure 1) in a small-bore 9.4T MRI system with maximum gradient=480mT/m. We collect a dMRI dataset using a 2-shot EPI sequence with δ=15ms, Δ=19ms, 514 directions, 0.25mm resolution and b­max=40000s/mm2. From this ultra-high resolution dataset, we generate datasets with varying number of gradient directions (122, 256, and 514) and spatial resolutions (0.25mm, 0.5mm, 1mm, and 2mm). We fit the generalized q-sampling image (GQI) model3 to each dataset and perform deterministic tractography4 with 100000 subvoxel seeds in DSIstudio. Using the two ROIs (ROI1 and ROI2) shown in Figure 1, we select the streamlines we use in our analysis (Figure 3). Then, using additional inclusion and exclusion masks, we isolate, from that subset of streamlines, six unique bundles (Figure 2) in the original dataset and assess how well they can be reconstructed as we reduce the spatial and angular resolution.

Results

We evaluate tractography accuracy on each dataset by computing the fraction of incorrect tractography streamlines. A streamline is deemed “incorrect” if a) it goes through the inclusion masks for one of the six bundles shown in Fig. 2, but also diverges outside the corresponding bundle, or b) is not neuroanatomically sound1 (e.g. streamlines cutting through the putamen instead of splitting into either extreme capsule or internal capsule). Figure 3 shows the portion of streamlines we used for this analysis and Figure 4 shows which fraction of those streamlines is deemed incorrect. We assess tractography accuracy by calculating the ratio of the number of incorrect streamlines over the total number of streamlines that go through ROIs 1 and 2 of Figure 1; we perform this analysis across datasets with varying spatial resolution and angular resolution (Figure 5).

Discussion

Our findings show that spatial resolution has a larger effect on tractography reconstruction than angular resolution. Across all the datasets with each different number of diffusion directions, an increase in voxel size resulted in an increase in incorrect streamlines, with a dramatic accuracy loss at the 2mm resolution that is commonly used in vivo (Fig. 5). Also, the number of errors generally decreased when adding more directions, and this gain was substantial at the 2mm resolution but much less so at higher spatial resolutions.

Conclusion

We used ex-vivo human data acquired at an ultra-high field strength to study the effect of spatial and angular resolution on dMRI tractography. Our results indicate that voxel size has a much more dramatic effect than the number of gradient directions on the accuracy of the reconstruction of a challenging WM area in the human brain. These results are specific to the crossing and splitting pathways dorsal to the striatum and the internal capsule, and the analyses methods used here. Our future work will analyze tractography accuracy for pathways from additional areas, as well as under varying SNR levels. Ultimately, our goal is to determine whether there is a trade-off between SNR, spatial and angular resolution that is feasible in vivo and that yields satisfactory tractography performance.

Acknowledgements

This research was carried out in whole or in part at the Athinoula A. Martinos Center for Biomedical Imaging at the Massachusetts General Hospital, and was funded by NIH P50 MH106435-01, R01 EB021265, R01 5RO1MH04557325, NBIB P41EB015896 and the NIH Blueprint for Neuroscience Research (T90DA022759/R90DA023427). This work also involved the use of instrumentation supported by the NIH Shared Instrumentation Grant Program and/or High-End Instrumentation Grant Program; specifically, grant number(s) S10RR016811, S10RR023401, S10RR019307, S10RR019254, S10RR023043.

References

1. Schmahmann, J. D. & Pandya, D. N. Fiber Pathways of the Brain. New York Oxford University Press 1, (2006).

2. Grisot, G. et al. Optimization of acquisition parameters for diffusion MRI using chemical tracing. ISMRM proceedings, (2016).

3. Yeh, F. C., Wedeen, V. J. & Tseng, W. Y. I. Generalized q-sampling imaging. IEEE Trans. Med. Imaging 29, 1626–1635 (2010).

4. Yeh, F. C., Verstynen, T. D., Wang, Y., Fernández-Miranda, J. C. & Tseng, W. Y. I. Deterministic diffusion fiber tracking improved by quantitative anisotropy. PLoS One 8, (2013).

Figures

Figure 1: Block of human tissue imaged ex-vivo at ultra-high resolution (0.25mm). A: Human tissue sample used for this analysis. The yellow-dashed box shows the block of tissue we focused on. B: b=0 image (top) from the 0.25mm volume of the tissue block. Whole block tractography (middle) and the isolated six bundles we focus on (bottom). The ROIs used to select the subset of streamlines we use in our analysis are shown in the bottom box. We then investigate which of these streamlines correctly travels through the six selected bundles and which ones do not.

Figure 2: We investigate tractography accuracy in reconstructing six bundles: the Muratoff bundle (green), two branches of the superior longitudinal fasciculus (orange and light blue), the extreme capsule (purple), the internal capsule (blue) and the corpus callosum (red). In prior studies comparing dMRI tractography and chemical tracing, we found that the intricate organization of this area (where pathways split and cross at multiple locations) often causes tractography reconstruction errors. Thus, this sample is a great test-bed for tractography.

Figure 3: Streamlines used for our analysis. Each image represents the streamlines going through either of the two ROIs shown in Figure 1 in all the datasets. Spatial resolution varies along rows, angular resolution along columns. The comparison of streamlines from each dataset already reveals a general loss of bundle definition and addition of spurious pathways as resolution worsens.

Figure 4: The number of incorrect pathways increases greatly as resolution decreases. The figure shows the fraction of the streamlines shown in Fig. 3 that are deemed incorrect. A streamline is deemed “incorrect” if it goes through the inclusion masks for one of the six bundles shown in Fig. 2, but also diverges outside the corresponding bundle or is not neuroanatomically sound (e.g. streamlines cutting through the putamen instead of splitting). Spatial resolution varies along rows, angular resolution along columns.

Figure 5: Analysis of incorrect streamlines across all datasets. Across all the datasets with each different number of diffusion directions, an increase in voxel size resulted in an increase in incorrect streamlines, with a dramatic accuracy loss at the 2mm. Also, the number of errors generally decreased when adding more directions.

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)
3392