Skip to content

Latest commit

 

History

History
45 lines (30 loc) · 2.32 KB

README.md

File metadata and controls

45 lines (30 loc) · 2.32 KB
Raw

How to use the model for MS lesion segmentation

Step 1: Cloning the Repository

Open a terminal and clone the repository using the following command:

git clone https://github.com/ivadomed/canproco.git

Step 2: Setting up the Environment

The following commands show how to set up the environment. Note that the documentation assumes that the user has conda installed on their system. Instructions on installing conda can be found here.

  1. Create a conda environment with the following command:
conda create -n venv_nnunet python=3.9
  1. Activate the environment with the following command:
conda activate venv_nnunet
  1. Install the required packages with the following command:
cd canproco
pip install -r packaging/requirements.txt
  1. Download the model (model_ms_seg_sc-lesion_regionBased_2d.zip or model_ms_seg_sc-lesion_regionBased_3d_fullres.zip) from the repository's latest release, which can be found here, and unzip it.

Step 3: Getting the Predictions

To segment a single image using the trained model, run the following command from the terminal. This assumes that the model has been downloaded and is available locally. The release contains two models, the 2D nnUNet as well as the 3D nnUNet. Our experiments showed that both worked similarly.

However, because the model was trained on inverted PSIR (multiplied by -1) and STIR images, for performance purposes, we recommend inverting your PSIR images before performing inference. That can be done using sct_maths -i <INPUT> -mul -1 -o <OUTPUT>

python packaging/run_inference_single_subject.py --path-image /path/to/image --path-out /path/to/output/directory --path-model /path/to/model

The output contains the spinal cord segmentation (with value 1) and the MS lesion segmentation (with value 2). It uses a region-based approach, meaning that lesions are always located within the spinal cord segmentation.

ℹ️ The script also supports getting segmentations on a GPU. To do so, simply add the flag --use-gpu at the end of the above commands. By default, the inference is run on the CPU. It is useful to note that obtaining the predictions from the GPU is significantly faster than the CPU.