Differences in GM segmentation

Dear all,

Almost by chance, I found out about the _sct_segment_graymatter routine.
Does it yield better results than the sct_deepseg_gm? Which differences exist between them?

If we want to obtain the WM segmentation, would you advise to run the deepseg routine in both, sc and gm, and then obtain the difference from those segmentations?

I am using a Gradient Echo Medic (GRE-ME) sequence to obtain the gm. As there has been said, the patient movement conditions the results a lot. Nevertheless, when there is none, the GM segmentation is quite decent.

thank you for your time,
aran

Hi @Aran,

Almost by chance, I found out about the _sct_segment_graymatter routine.

Indeed, however the function has prefix “_” to be hidden to the user because it currently doesn’t work (details in this issue)

Does it yield better results than the sct_deepseg_gm ?

In general, sct_deepseg_gm is better, hence the decision to deprecate the former function. More details about the comparison in Prados et al. Neuroimage 2017.

Which differences exist between them?

The former is using a dictionary-based registration while the latter is using deep learning. The reference above details the algorithms. Also, the usage of each function gives you a link to a publication, which describes the algorithms.

If we want to obtain the WM segmentation, would you advise to run the deepseg routine in both, sc and gm , and then obtain the difference from those segmentations?

Absolutely, as done in the example batch_processing.sh.

I am using a Gradient Echo Medic (GRE-ME) sequence to obtain the gm. As there has been said, the patient movement conditions the results a lot. Nevertheless, when there is none, the GM segmentation is quite decent.

Good. In that case sct_deepseg_gm should do a good job.

Best,
Julien