sapjunior / pytorch-liver-segment-sample

PyTorch Semantic Segmentation Sample Code for Liver (Medical Segmentation Decathlon Dataset)

PyTorch semantic segmentation example code for Liver segmentation (Medical Decalthon Dataset)

Environment installation

Download and install mambaforge (faster miniforge/mniniconda alternative) from below link

https://github.com/conda-forge/miniforge#mambaforge

mamba create -n liver-segmentation python=3.9 -c conda-forge
conda activate liver-segmentation
mamba install numpy matplotlib jupyterlab tqdm qudida scikit-image scipy pyyaml scikie-learn pywavelets tifffile imageio networkx threadpoolctl joblib dicom2nifti  -c conda-forge -y
mamba install pytorch torchvision torchaudio cudatoolkit=11.6 -c pytorch -c conda-forge -y
pip install opencv-python-headless albumentations
pip install git+https://github.com/qubvel/segmentation_models.pytorch.git

Dataset Preparation

Download Task03_Liver.tar to dataset/ from http://medicaldecathlon.com and extract by using the following command

tar -xvf dataset/Task03_Liver.tar

Then run dataPrepation.py to convert niff file to png format

PS. If you used your lab server, the prepared dataset is located at "/mnt/datasets/liver"

Training

See detail and adjust parameters to suited your need in train.py

python train.py

Training output (weight/output_mask) will be placed in

outputs/{expName}/