Tractography Filtering
Re-establishing quantitativeness in dMRI-based tracrography.
Among the many problematic issues of tractography-based structural connectivity estimation, the non-quantitative nature of tractography, the methodological limitations of dMRI-based tractography and the presence of many false positive connections within connectomes are the ones that inspired some of the most recent methodological advances in the field of non-invasive tractography. With the aim of re-establishing the quantitative properties that one expects from tractography, i.e. the strength of the structural connection being associated to quantitative aspects of the fiber population represented by each streamline, several evolution of the standard tracking paradigm have been developed. One such evolution is the so-called global tractography approach, which makes use of a simulated annealing process and has been proven to mitigate the potential bias induced by the geometry of the streamlines to be tracked (bending, fanning, crossing, kissing, …). Another technique called AxTract simultaneously tracks the streamlines and associates an axonal diameter to the tracked bundles. The global and AxTract methods are generative (a.k.a. bottom-up) techniques, as they build streamlines according to criteria that are more restrictive and better informed than the ones of ordinary tractography algorithms.
In contrast with the bottom-up approach, a series of top-down techniques have been proposed, namely the Spherical-deconvolution Informed Filtering of Tractograms (SIFT) technique and its evolution SIFT2, the Convex Optimization Modelling for Microstructure Informed Tractography (COMMIT) and its evolution COMMIT2 and the Linear Fascicle Evaluation (LiFE) model. These techniques go under the name of Tractography Filtering Techniques (TFTs). They take a predefined tractogram and associate to each streamline a single weight that represents the amount of signal explained by or the connectivity strength associated to the streamline depending on the used technique. In my PhD thesis we reviewed these methods, proposed their unification under a more general formulation and presented the Functionally Informed TFT, which is a novel filtering technique that extends the previous ones by integrating functional information in the model.
In (Frigo et al., 2020) we showed that TFTs change the topology of structural connectomes estimated with dMRI.