I’m interested in performing within subject analysis of some tracts of interest. In particular, i would like to quantitatively compare homologous connections between both hemispheres.

For this purpose, I have read the preprint “Quantitative streamlines tractography: methods and inter-subject normalisation” and I have a few questions about it.

I’m wondering if I perform a whole brain tractography and filter it with SIFT algorithm (instead of SIFT2), Can I assume that the streamline number remaining between two regions represents the Fibre Bundle Capacity (FBC) of the white matter pathway? In other words, I would like to know if the SIFT algorithm is still valid to obtain the FBC between two regions.

In case that the above assumption was true, the number of streamlines in the connectome matrix would represent the FBC between any pair of regions, isn’t it?.

Could I directly compare the FBC of two different tracts within the same patient or is necessary to apply some kind of normalization to the connectome matrix before doing that?

Finally, I briefly describe my dataset and processing pipeline in case it were useful to answer my questions. Also, feel free to comment and make any suggestions about it.

I have single shell data (one b0 and 55 b1 volumens) with low b-value (b1=1000). I know that my dataset is far to be optimum, but I tried to do the “best” with it.

Pre-processing: denoising, unringing, preproc.

Response function estimation (Dhollander algorithm)

Estimation FOD (SS3T-CSD algorithm)

Intensity Normalization (mtnormalise algorithm)

Whole Brain Tractography using ACT, iFOD2 and seed dynamic strategy. I generate 10M streamlines

Apply SIFT to filter the tractogram from 10M to 2.5M of streamlines. The cost function reach the value of 12%. I choosed to stop in 2.5M of streamlines because with lowers values the warning “quantisation error” appears.

I’m wondering if I perform a whole brain tractography and filter it with SIFT algorithm (instead of SIFT2) … In other words, I would like to know if the SIFT algorithm is still valid to obtain the FBC between two regions.

The relationship to the original SIFT method is mentioned only briefly in the cited preprint (for which I’m sure I will one day find the time to respond to reviewers… ) . Let’s say that in the process of running SIFT, over and above the removal of streamlines, the proportionality coefficient also increases by a factor of 10. This outcome is equivalent to the SIFT2 formulation whereby the proportionality coefficient remains fixed and the streamline weights are 0 for all removed streamlines and 10 for all retained streamlines. Therefore the same consequential logic applies.

I would however question why you are pursuing the use of SIFT rather than SIFT2. There are situations where that may be preferable, but you would want such to be justified.

the streamline number remaining between two regions represents the Fibre Bundle Capacity (FBC) of the white matter pathway

I am hesitant to agree with this statement fully. FBC as defined there specifically includes multiplication by the proportionality coefficient; so directly equating streamline number and FBC is technically incorrect. That scaling is however inconsequential if you are only interested in the relative connection densities of different pathways within an individual, as opposed to either absolute connection density or relative connection densities between individuals.

Could I directly compare the FBC of two different tracts within the same patient or is necessary to apply some kind of normalization to the connectome matrix before doing that?

It is precisely the biases present in quantification of such a ratio that SIFT and related methods are designed to address. It’s obviously still highly imperfect, but it is nevertheless their raison d’être.