This repository contains the code for deep learning-based segmentation of the spinal cord and intramedually lesions in spinal cord injury (SCI). The code is based on the nnUNetv2 framework.

The model was trained on raw T2-weighted images of SCI patients from seven sites comprising traumatic (acute preoperative, intermediate, chronic) and non-traumatic (ischemic SCI and degenerative cervical myelopathy, DCM) SCI lesions. The data included images with heterogenous resolutions (both axial and sagittal) and scanner strengths (1T/1.5T/3T). To ensure uniformity across sites, all images were initially re-oriented to RPI. Given an input image, the model is able to segment both the lesion and the spinal cord.

• We have released SCIsegV2: A Universal Model for Intramedually Lesion Segmentation in SCI. The new model is trained on a larger cohort covering both traumatic and non-traumatic SCI lesions. The new model is available for download using the same installation instructions as below.
• The computation of midsagittal tissue bridges is now fully-automatied and powered by SCIsegV2.
• We have moved away from ANIMA metrics and have started to use MetricsReloaded instead. This wrapper script is used to compute metrics and an internal fork of the package is maintained here.

• Spinal Cord Toolbox (SCT) v6.2 or higher -- follow the installation instructions here
• conda
• Python (v3.9)

Once the dependencies are installed, download the latest SCIseg model:

sct_deepseg -install-task seg_sc_lesion_t2w_sci

To segment a single image, run the following command:

sct_deepseg -i <INPUT> -task seg_sc_lesion_t2w_sci

For example:

sct_deepseg -i sub-001_T2w.nii.gz -task seg_sc_lesion_t2w_sci

The outputs will be saved in the same directory as the input image, with the suffix _lesion_seg.nii.gz for the lesion and _sc_seg.nii.gz for the spinal cord.

This new functionality is available via SCT's sct_analyze_lesion. The function computes the midsagittal tissue bridges and outputs the ventral and dorsal tissue bridges.

sct_analyze_lesion -m <SUBJECT>_lesion_seg.nii.gz -s <SUBJECT>_sc_seg.nii.gz

If you find this work and/or code useful for your research, please cite our paper:

@article {Naga Karthik2024.01.03.24300794,
	author = {Enamundram Naga Karthik* and Jan Valosek* and Andrew C. Smith and Dario Pfyffer and Simon Schading-Sassenhausen and Lynn Farner and Kenneth A. Weber II and Patrick Freund and Julien Cohen-Adad},
	title = {SCIseg: Automatic Segmentation of T2-weighted Intramedullary Lesions in Spinal Cord Injury},
	elocation-id = {2024.01.03.24300794},
	year = {2024},
	doi = {10.1101/2024.01.03.24300794},
	publisher = {Cold Spring Harbor Laboratory Press},
	URL = {https://www.medrxiv.org/content/early/2024/04/21/2024.01.03.24300794},
	eprint = {https://www.medrxiv.org/content/early/2024/04/21/2024.01.03.24300794.full.pdf},
	journal = {medRxiv},
	note = {*Shared first authorship}
}