mattragoza / lung-project

Lung biomechanical modeling

Lung biomechanical deep learning project

Summary of procedure

1. Setup conda environment
2. Download and preprocess images
3. Create image segmentation masks
4. Create image registration fields
5. Train deep learning model

Conda environment

Run the following to create the conda environment and register it as jupyter notebook kernel:

mamba env create --file=environment.yml
mamba activate lung-project
python -m ipykernel install --user --name=lung-project

Emory 4DCT image set

Download the images

1. Go to download page
2. Submit access request form
   • Landing page contains dropbox password
3. For each case packet i=1..10,
   • Follow the download link to DropBox
   • Enter password and download Case${i}Pack.zip
   • Move .zip file to lung-project/data/download
4. Unzip case packets into lung-project/data/Emory-4DCT
   • Use the commands below:

cd lung-project
for i in {1..10};
	do unzip data/download/Case${i}Pack.zip -d data/Emory-4DCT;
done

Convert images to NIFTI format

TODO

TotalSegmentator

Install

#pip install TotalSegmentator
TotalSegmentator -i $input_image -o $output_dir --device gpu --preview --statistics -ta total --roi_subset lung_upper_lobe_right lung_upper_lobe_left lung_middle_lobe_right lung_lower_lobe_right lung_lower_lobe_left
TotalSegmentator -i $input_image -o $output_dir --device gpu --preview --statistics -ta lung_vessels
totalseg_combine_masks -i $output_dir -o $output_dir/lung_combined_mask.nii.gz -m lung

Usage

TODO

CorrField registration

Install

git clone [e-mail]/multimodallearning/Lung250M-4B.git
cd Lung250M-4B/corrfield
python corrfield.py -F {fixed_image} -M {moving_image} -m {fixed_mask} -o {output_path}

Usage

TODO

Train deep learning model

TODO