SIFT2 weights & streamlines lengths


#1

Another question regarding streamline lengths is whether or not tracks weights should be used when available, e.g. as in:

tck2connectome tracks.tck nodes.mif connectome.csv \
    -scale_length -stat_edge mean -tck_weights_in sift2.txt

I suppose this relates to other posts regarding the normalisation of SIFT2 weights. Is it recommended to use them, or perhaps another output of tcksift2 like -out_mu or -out_coeffs?


Streamline length
#2

Hi Jonathan,

I’ve moved this to a separate topic as it was only very loosely related to the thread in which it was posted. Generally if a question is a bit of a sidestep from another topic, it’s better to post a new thread and provide a link to the related one :+1:

It’s entirely appropriate to incorporate SIFT2 weights into tck2connectome when quantifying the mean length of each edge. In this case a weighted mean is performed: e.g. if there are 2 streamlines, one with length 50mm and weight 0.2 and the other with length 150mm and weight 1.8, the weighted mean length will be 140mm, as opposed to 100mm if the streamline weights were not provided.

Scaling by the proportionality coefficient would have no effect here: global scaling of streamline weights would disappear in the process of calculation of the weighted mean length, as the sum of streamline weights appears in the denominator.

Using the weighting coefficients rather than the weighting factors would be inappropriate here. Within the SIFT2 algorithm it is the coefficients that are optimised, but the weighting factors are the quantities with physical meaning; hence why the latter are the compulsory output from the command.

Rob