mpaquette / tractography Goto Github PK

Currently, we need to create the same assign matrix in compute probability and in build_graph, this should be saved with the probability and loaded by the graph, same for the angle threshold.

Implement random walk from source to a group of targets using the neighbour probability, emulating streamline tractography

The current method for distribution of the fODF weights is biased towards the 6 direct neighbors, we could instead use a fixed size cone for each. I don't expect this to matter much.

Direction pointing toward out-of-mask voxel gets thrown away and everything else gets reweighted, artificially increasing probability for mask edge voxel.

Implement network emulating deterministic tracto

Since the main scripts only accept continuous labels in label_map, add script to subsample label_map and save txtfiles with the info

Handle non continous label in label map, have option to save the existing label or to zeropad

Add an optional flag for target label_map if you want them different from source label_map.

Revamp the deprecated figures scripts into connectome matrices plotting scripts

Make individual source and target nodes for each label. This will allow the removal of the "expensive edges" workaround.

Current code can't handle pairs of labels without paths, i.e. disconnected graph.