0166

On the intrahemispheric connectivity of the monkey: a diffusion tractography and tract tracing analysis
Gabriel Girard1,2, Roberto Caminiti3,4, Alexandra Battaglia-Mayer3, Etienne St-Onge5, Karen S. Ambrosen6,7, Simon F. Eskildsen8, Kristine Krug9, Tim B. Dyrby7,10, Maxime Descoteaux5, Giorgio Innocenti11,12, and Jean-Philippe Thiran1,2

1Radiology Department, Centre Hospitalier Universitaire Vaudois and University of Lausanne, Lausanne, Switzerland, 2Signal Processing Lab (LTS5), École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, 3Department of Physiology and Pharmacology, University of Rome SAPIENZA, Rome, Italy, 4Department of Anatomy, Histology, Forensic Medicine, and Orthopedics, University of Rome SAPIENZA, Rome, Italy, 5Sherbrooke Connectivity Imaging Lab, Computer Science Department, Université de Sherbrooke, Sherbrooke, QC, Canada, 6Department of Applied Mathematics and Computer Science, Technical University of Denmark, Kongens Lyngby, Denmark, 7Danish Research Centre for Magnetic Resonance, Center for Functional and Diagnostic Imaging and Research, Copenhagen University Hospital Hvidovre, Hvidovre, Denmark, 8Center of Functionally Integrative Neuroscience, Department of Clinical Medicine, Aarhus University, Aarhus, Denmark, 9Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, United Kingdom, 10Department of Applied Mathematics an Computer Science, Technical University of Denmark, Kongens Lyngby, Denmark, 11Department of Neuroscience, Karolinska Institutet, Stockholm, Sweden, 12Brain and Mind Institute, École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

Synopsis

In this work, we compare diffusion tractography with neuronal retrograde tract tracing of the frontal, cingulate and parietal areas of the monkey. We analyze the agreements between the tractography and the tracing for connected and not connected regions. We report an accuracy of 0.71 across all pairs of regions, with twice the number of true positive than false positive connections. Some regions show accuracy higher than 0.80, while other regions show accuracy lower than 0.6. A further analysis of the location of false positive and false negative connections will help understand the limitations and improve diffusion tractography algorithms.

Introduction

Diffusion-weighted magnetic resonance imaging tractography is a unique tool to probe the white matter (wm) network. However, recent studies have shown that tractography can extract wm fascicles but produces a high incidence of false positive connections1,2. This raises the question on the accuracy tractography network reconstruction. The rich literature in tract tracing studies in the monkey offers the tool to challenge and evaluate the performance of the tractography reconstructions3,4,5,6. In this study, we compared the wm network reconstruction from the state-of-the-art diffusion tractography in the short intrahemispheric connections of frontal, cingulate and parietal areas with neuronal retrograde tracing studies.

Methods

Diffusion data processing. The ex-vivo monkey data was acquired on a 4.7T Agilent scanner (DRCMR, Copenhagen) following the setup described in Dyrby et al.7, on three shells of 180 uniformly distributed gradient directions with $$$b = [1.610, 4.100, 7.700] ms/um^2$$$ ($$$G = [150, 250, 350] mT/m$$$, $$$\delta = 8.4 ms$$$, $$$\Delta = 15 ms$$$, $$$TE = 30ms$$$, 9 b0). The acquisition was done twice and averaged for the computation of the fiber orientation distribution function using the Multi-Shell Multi-Tissue Constrained Spherical Deconvolution algorithm8,9. Partial volume estimates for of the wm, grey matter (gm) and cerebrospinal fluid were obtained from the average b0 image10. The wm-gm and the middle cortical layers surfaces were generated by the Fast Accurate Cortex Extraction algorithm11. Tractography was initiated from the wm-gm surface using the Surface-Enhanced Tractography (SET) algorithm12 and the streamline propagation was performed using the Particle Filtering Tractography (PFT) algorithm13,14.

Brain Parcellation. The average b0 image was used to parcellate the frontal, cingulate and parietal gm of the right hemisphere in 59 regions of interest (ROIs) following Caminiti et al.6 and references therein. A systematic literature review of retrograde neural tracers in the monkey was performed. Pairs of ROIs were labeled connected if any amount tracer was reported in a ROI for an injection site located in the other ROI and labeled not connected otherwise. Pairs of ROIs with conflicting information were excluded from this study, yielding a total 1,686 pairs of ROIs (680 connected).

Connectomic analysis. The tractography connectivity matrix was constructed using the number of streamlines between each pair of ROIs divided by their combined volume to account for ROIs volume differences8,15,16. All pairs of ROIs with a connectivity value higher than a threshold value were considered connected. We report the True Positives ($$$TP$$$) and True Negatives ($$$TN$$$) as pairs of ROIs labeled connected or not connected by both tractography and tract tracing, respectively. False Positives ($$$FP$$$) are reported as pairs of ROIs connected by tractography and not connected by tract tracing, and conversely for False Negatives ($$$FN$$$). Additionally, we report the True Positive Rate ($$$TPR=\frac{TP}{TP+FN}$$$), the False Positive Rate ($$$FPR=\frac{FP}{FP+TN}$$$) and the accuracy ($$$\frac{TP+TN}{TP+TN+FP+FN}$$$). The optimal threshold was selected to minimize $$$\sqrt{(1-TPR)^2+FPR^2}$$$.

Results

Figure 1 shows the wm-gm surface used as input for the SET algorithm (white), after the cortical flow algorithm (green) and a subset of streamlines generated with the PFT algorithm connecting the green surface. From the 1,310,360 generated streamlines, 121,201 streamlines have both endpoints in ROIs, shown in Figure 2. Among those streamlines, 78% have endpoints in pairs of ROIs labeled connected by tracer studies. Figure 3 shows the Receiver Operating Characteristic curve of tractography for a decreasing threshold of the volume weighted streamline count. Diffusion tractography shows an accuracy of 0.61 for a threshold=0 ($$$TPR=0.86,~FPR=0.55$$$) and 0.71 for the optimal threshold=0.026 ($$$TPR=0.69,~FPR=0.28$$$). Figure 4 reports the $$$TP,~TN,~FP,~FN$$$ for both thresholds. Figure 5 shows the spatial distribution of the $$$TPR$$$, $$$FPR$$$ and accuracy across the studied ROIs.

Discussion and Conclusion

Tractography shows a positive outcome in its ability to identify the structural network of the frontal, cingulate and parietal areas. With the optimal threshold, tractography showed an agreement with tracer studies with almost twice the amount of $$$TP$$$ than $$$FP$$$ and over thrice the amount of $$$TN$$$ than $$$FN$$$. It is worth pointing out that we focused solely on the location of the streamline endpoints to identify connections. Thus, streamlines with valid endpoints but erroneous pathways were considered $$$TP$$$, as opposed to $$$FP$$$ in Maier-Hein et al.1. Figure 5 shows that tractography well predicts the structural connectivity of some ROIs with accuracy higher than 0.80, while having an accuracy lower than 0.60 for other ROIs. It is worth mentioning that specific pairs of ROIs might have been out of the scope of tracing studies, and thus, not reported connected. Further investigation of the $$$FP$$$ and $$$FN$$$ connections will help refine the tract tracing connectivity, understand tractography limitations and improve reconstruction algorithms.

Acknowledgements

This work is supported in part by the Center for Biomedical Imaging (CIBM) of the Geneva-Lausanne Universities and the EPFL, as well as the foundations Leenaards and Louis-Jeantet.

References

  1. Maier-Hein, K. H., Neher, P. F., Houde, J.-C., Côté, M.-A., Garyfallidis, E., Zhong, J., … Descoteaux, M. (2017). The challenge of mapping the human connectome based on diffusion tractography. Nature Communications, 8(1), 1349.
  2. Jbabdi, S., Sotiropoulos, S. N., Haber, S. N., Van Essen, D. C., & Behrens, T. E. (2015). Measuring macroscopic brain connections in vivo. Nature Neuroscience, 18(11), 1546–1555.
  3. Markov, N. T., Ercsey-Ravasz, M. M., Ribeiro Gomes, A. R., Lamy, C., Magrou, L., Vezoli, J., Misery, P., Falchier, A., Quilodran, R., Gariel, M. A., Sallet, J., Gamanut, R., Huissoud, C., Clavagnier, S., Giroud, P., Sappey-Marinier, D., Barone, P., Dehay, C.Toroczkai, Z., Knoblauch, K., Van Essen, D. C., Kennedy, H. (2014). A Weighted and Directed Interareal Connectivity Matrix for Macaque Cerebral Cortex. Cerebral Cortex, 24(1), 17--36.
  4. Azadbakht, H., Parkes, L. M., Haroon, H. A., Augath, M., Logothetis, N. K., de Crespigny, A., … Parker, G. J. M. (2015). Validation of High-Resolution Tractography Against In Vivo Tracing in the Macaque Visual Cortex. Cerebral Cortex, 25(11), 4299--4309.
  5. Donahue, C. J., Sotiropoulos, S. N., Jbabdi, S., Hernandez-Fernandez, M., Behrens, T. E., Dyrby, T. B., … Glasser, M. F. (2016). Using Diffusion Tractography to Predict Cortical Connection Strength and Distance: A Quantitative Comparison with Tracers in the Monkey. The Journal of Neuroscience, 36(25), 6758–6770.
  6. Caminiti, R., Borra, E., Visco-Comandini, F., Battaglia-Mayer, A., Averbeck, B. B., & Luppino, G. (2017). Computational Architecture of the Parieto-Frontal Network Underlying Cognitive-Motor Control in Monkeys. Eneuro, 4(1), ENEURO.0306-16.2017.
  7. Dyrby, T. B., Baaré, W. F. C., Alexander, D. C., Jelsing, J., Garde, E., & Søgaard, L. V. (2011). An ex vivo imaging pipeline for producing high-quality and high-resolution diffusion-weighted imaging datasets. Human Brain Mapping, 32(4), 544--563.
  8. Jeurissen, B., Tournier, J.-D., Dhollander, T., Connelly, A., & Sijbers, J. (2014). Multi-tissue constrained spherical deconvolution for improved analysis of multi-shell diffusion MRI data. NeuroImage, 103, 411--426.
  9. Tournier, J.-D., Calamante, F., & Connelly, A. (2012). MRtrix: Diffusion tractography in crossing fiber regions. International Journal of Imaging Systems and Technology, 22(1), 53--66.
  10. Zhang, Y., Brady, M., & Smith, S. (2001). Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Transactions on Medical Imaging, 20(1), 45--57.
  11. Eskildsen, S. F., & Østergaard, L. R. (2006). Active Surface Approach for Extraction of the Human Cerebral Cortex from MRI. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 823--830).
  12. St-Onge, E., Daducci, A., Girard, G., & Descoteaux, M. (2018). Surface-enhanced tractography (SET). NeuroImage, 169, 524--539.
  13. Girard, G., Whittingstall, K., Deriche, R., & Descoteaux, M. (2014). Towards quantitative connectivity analysis: reducing tractography biases. NeuroImage, 98, 266--278.
  14. Garyfallidis, E., Brett, M., Amirbekian, B., Rokem, A., Van Der Walt, S., Descoteaux, M., & Nimmo-Smith, I. (2014). Dipy, a library for the analysis of diffusion MRI data. Frontiers in Neuroinformatics, 8.
  15. Yo, T.-S., Anwander, A., Descoteaux, M., Fillard, P., Poupon, C., & Knösche, T. R. (2009). Quantifying brain connectivity: a comparative tractography study. International Conference on Medical Image Computing and Computer-Assisted Intervention, 12(Pt 1), 886--893.
  16. Sotiropoulos, S. N., & Zalesky, A. (2017). Building connectomes using diffusion MRI: why, how and but. NMR in Biomedicine, e3752.

Figures

Figure 1. Diffusion Tractography. The wm-gm surface used as input for the SET algorithm is shown in white. The surface after the cortical flow algorithm is shown in green. A subset of streamlines generated with the PFT algorithm connecting the green surface is shown using the direction-encoded color (red: left-right, green: anterior-posterior, blue: inferior-superior).

Figure 2. Regions of interest (ROIs) of the frontal, cingulate and parietal gm of the right hemisphere of the studied monkey. The gm was manually parcellated in 59 ROIs following literature report of retrograde neural tracers.


Figure 3. Receiver Operating Characteristic curve of diffusion tractography for a decreasing threshold of the volume weighted streamline count. For a threshold=0, diffusion tractography shows a $$$TPR$$$ of 0.86, a $$$FPR$$$ of 0.55 and an accuracy of 0.61. Using the optimal threshold=0.026, $$$TPR$$$ and $$$FPR$$$ decrease to 0.69 and 0.28, respectively, resulting an accuracy increase to 0.71. The optimal threshold value is selected to minimize $$$\sqrt{(1-TPR)^2+FPR^2}$$$.


Figure 4. Confusion matrices for threshold=0 (top) and optimal threshold=0.026 (bottom). Tractography shows an agreement with tracer studies with almost twice the amount of $$$TP$$$ than $$$FP$$$ and over thrice the amount of $$$TN$$$ than $$$FN$$$ for the optimal threshold. Moreover, the number of missing connections ($$$FN$$$) from diffusion tractography is similar to the number of erroneous connections ($$$FP$$$).


Figure 5. Spatial distribution of the $$$TPR$$$ (top), $$$FPR$$$ (middle) and accuracy (bottom) over the 59 ROIs for the optimal threshold, shown on the middle cortical layers surface. The structural connectivity of some ROIs is well predicted with tractography with accuracy higher than 0.80 (yellow and white) while tractography underperformed for other ROIs with accuracy lower than 0.60 (black and dark red).

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
0166