tumor-segmentation

Pipeline for the segmentation of spinal tumors

Data

The data is organized according to the BIDS convention, as shown below:

Tumor_BIDS/
└── dataset_description.json
└── participants.tsv
└── sub-Astr144
    └── anat
        └── sub-Astr144_T1w.nii.gz
        └── sub-Astr144_T1w.json
        └── sub-Astr144_T2w.nii.gz
        └── sub-Astr144_T2w.json
└── sub-Hema264
└── sub-Epen393
└── derivatives
    └── labels
        └── sub-Astr144
            └── anat
                └── sub-Astr144_T1w_seg-tumor.nii.gz --> Tumor segmentation

The prefixes 'Astr', 'Epen' and 'Hema' respectively correspond to the tumor types Astrocytoma, Ependymoma and Hemangioblastoma. The subject numbers for 'Astr' subjects go from 144 to 263, for 'Hema' subjects from 264 to 392 and for 'Epen' subjects from 393 to 523.

More information for converting and organizing BIDS data is available here.

Run tumor segmentation model

1. Install SCT software if it's not already on your computer.

2. Download the model. This step should be only be done once:
   sct_deepseg -install-model t2_tumor

3. Apply the model on your data:
   sct_deepseg -i path/to/img.nii.gz -model t2_tumor

To run the model on all images from a BIDS dataset, consult the notebook located in the tutorial folder.

Preprocessing Data

Currently, the preprocessing script crops the images around the spinal cord.

To prepare the data for training, run the following line sct_run_batch parameters/parameters.sh process_data.sh . This will process the BIDS repository specified in the parameters.sh file. Modify the parameters file if needed to select the images to be processed.