hardcoded filtering of short streamlines is a problem

[emanuele]

In tractome.py there is an hardcoded filter to remove short streamlines. First of all it is quite meaningless per se. Moreover it make it pretty difficult to reuse the saved segmentations in other code, because a set of unexpected steps must be done in order to use the correct streamlines IDs, namely:

• when loading a .trk tractography with nibabel it is essential to specify points_space='voxel'
• compute the length of each streamline with dipy.tracking.metrics.length
• keep only streamlines whose length is >= 15.

[emanuele]

Here is a gist to load .trk files as tractome does in this moment. It may be useful for loading segmentations outside tractome: https://gist.github.com/emanuele/b1ffb8edeeaee8b5b8f4