Lesion segmentation in spinal cord injury (SCI)

SCT provides a deep learning model called SCIseg for segmenting lesions in spinal cord injury (SCI) patients. The model is available in SCT v6.2 and higher via sct_deepseg -task seg_sc_lesion_t2w_sci. In SCT v6.4, the model was updated to SCIsegV2, the command remains the same.

The model was trained on raw T2-weighted images of SCI patients comprising traumatic (acute preoperative, intermediate, chronic) and non-traumatic (ischemic SCI and degenerative cervical myelopathy, DCM) SCI lesions.

The data included images with heterogeneous resolutions (axial/sagittal/isotropic) and scanner strengths (1T/1.5T/3T).

Given an input image, the model segments both the lesion and the spinal cord.

https://raw.githubusercontent.com/spinalcordtoolbox/doc-figures/master/lesion-analysis/sciseg.png

Run the following command to segment the lesion using SCIseg from the input image:

sct_deepseg -i t2.nii.gz -task seg_sc_lesion_t2w_sci -qc ~/qc_singleSubj
Input arguments:

  • -i : Input T2w image with fake lesion

  • -task : Task to perform. In our case, we use the SCIseg model via the seg_sc_lesion_t2w_sci task

  • -qc : Directory for Quality Control reporting. QC reports allow us to evaluate the segmentation slice-by-slice

Output files/folders:

  • t2_sc_seg.nii.gz : 3D binary mask of the segmented spinal cord

  • t2_lesion_seg.nii.gz : 3D binary mask of the segmented lesion

  • t2_lesion_seg.json : JSON file containing details about the segmentation model

Details: