where are the labels generated for the data?

[anoushkrit]

The labels for the inputFeatMatrix.h5 are provided separately in the training data as sf_clusters_label.h5 and there is tract_feat.py in the DeepWMA code which reorients the labels to groups of white matter and grey matter, and also superficial white matter.

So the question remains where are the labels for the training coming from, are they being

• [x] generated by the code itself or

• [x] it is being hand annotated by the near-radiologists or

• [x] they are being registered with the ATLASES available for the specific datasets.

• [x] Are there other ATLASES available for other data that has been sourced in the SupWMA paper.

[anoushkrit]

In wma, white matter analysis, there is automatic tract segmentation from the UKF tractography output VTK, and further they are being processed, There is some speculation that there we will find the method to generate labels for the datasets (especially HCP).

[anoushkrit]

So refer to https://github.com/SlicerDMRI/ORG-Atlases, and the document https://github.com/SlicerDMRI/whitematteranalysis/blob/master/doc/subject-specific-tractography-parcellation.md for understanding how the automatic labelling and fiber clustering is done

[anoushkrit]

https://github.com/SlicerDMRI/whitematteranalysis/blob/73a7948751f49d9fda8ec84fb5caeecaeeb92621/bin/wm_apply_ORG_atlas_to_subject.sh#L1-L40