CORTEX Student Club by Yujie Hou

It is increasingly believed that understanding brain function will require resolving its connectivity. In the human brain this can only be achieved with non-invasive techniques such as diffusion MRI (dMRI) based tractography. Using tract tracing we can establish a weighted and directed matrix of the cortical connectome. Tractography by definition does not provide information on direction. In addition, there is still a huge debate about the anatomical accuracy of the connectivity revealed by tractography. In this study, we set out to compare tractography with ground-truth tract tracing, in order to have an objective estimation of the accuracy of tractography. 

(1)   Comparing weighted measures of tractography and tract tracing (with 51 injections) revealed large numbers of false positives and a Spearman correlation of 0.56 (non-symmetrized) and 0.64 (symmetrized). Meaning that there is only a modest relation between tractography and tract tracing weight, which is penalized by the absence of directionality information in the dMRI tractography data. 

(2)   Our findings show that the correlation between tract tracing and tractography is mainly driven by the short distance connections and there is no correlation between the long distance tractography and tract tracing. Meaning that the correlation reported in (1) above is only reflecting very short distance connectivity.

(3)   Our tract tracing study shows that the variability of the connection weight across animal is around one order of magnitude in terms of fraction of labeled neurons (FLN). Hence, we carried out tract tracing and tractography in the same brain. This gave only a minimal improvement of the correlation. Meaning that the low correlation is not the consequence of inter-individual variability. 

(4)   Previously we have published that tract-tracing shows an exponential decrease of connectivity with distance leading to the formulation of the Exponential Distance Rule (EDR). In addition the team showed that the EDR allows prediction of many empirical findings concerning the geometry of connectivity. We examined the effect of distance on weighted connectivity extracted from tractography. This showed that tractography connectivity followed a much stricter exponential decline with distance than did tract tracing.  These findings suggest that distance effects might be contributing to the correlation effects between tractography and tract tracing. We have therefore carried out partial correlations which reduced the correlation from 0.6 to 0.3.  Meaning that when the effect of distance is removed the capacity of tractography to capture ground truth tract tracing is poor. 

(5)   Interestingly, we found that there is an important inter-area variability of tractography tract-tracing correlation. Meaning that some areas are better characterized by tractography than others. 

Conclusion. Globally our findings show that present-day dMRI tractography is not going to deliver an accurate and detailed understanding the connectivity of the human brain.  Some improvement of dMRI tractography could perhaps be achieved for instance by using anatomical priors (see point 5 above). One can expect that understanding the connectivity of the human brain will require continued efforts with indirect methods including functional connectivity and electrophysiology, as well as extrapolation from findings in Non-Human Primate (NHP).