Algorithm #1: sct_propsegΒΆ

The first spinal cord segmentation tool that SCT offers is called sct_propseg, and it works as follows:

1. Centerline detection:

sct_propseg starts by using a machine learning-based method (OptiC) to automatically detect the approximate center of the spinal cord.

https://raw.githubusercontent.com/spinalcordtoolbox/doc-figures/master/spinalcord-segmentation/optic_steps.png

(The centerline detection step is also provided in a standalone script called sct_get_centerline.)

2. Mesh propagation:

Next, a coarse 3D mesh is created by propagating along the spinal cord (PropSeg).

https://raw.githubusercontent.com/spinalcordtoolbox/doc-figures/master/spinalcord-segmentation/mesh_propagation.png
3. Surface refinement:

Finally, the surface of the mesh is refined using small adjustments.

sct_propseg was developed in 2014. In the years following, however, SCT has added a deep learning-based approach called sct_deepseg_sc.