MCAT-Net is a deep learning model for pulmonary nodule segmentation. This repository provides the model implementation, trained weights, and test paths.
The pretrained best_model and testdata can be downloaded from the following link:
To set up a new environment, ensure you have Conda installed, then run the following commands to create a new environment and install the dependencies:
conda create -n MCAT python==3.11
conda activate MCAT
pip install -r requirements.txt
Before running the model, update the following paths in the global.py file to match your directory structure:
bbox_img_path = "/home/hutianjiao/Project/Test_data/testdatas/image/" # Path to the test images
bbox_msk_path = "/home/hutianjiao/Project/Test_data/testdatas/label/" # Path to the test labels
best_dice_path = "/home/hutianjiao/Project/best_dice.pth" # Path to the best model file
output_path = "/home/hutianjiao/MCAT_NET/output/" # Path for saving outputs.
After configuring the paths, you can run the test script. By default, the script will use GPU number '1'. If needed, you can modify the GPU index in the script.
(i.e.: (DEVICE = torch.device("cuda:1" if torch.cuda.is_available() else "cpu")) in test.py)
python test.py