Anatomy Embedding Foundation Model

Installation

• Create a virtual environment conda create -n monai and activate it conda activate monai
• Install PyTroch conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
• git clone https://github.com/DlutMedimgGroup/Anatomy-Embedding-Foundation-Model
• Enter the project folder cd Anatomy-Embedding-Foundation-Model and run pip install -e .

Get Started

Download the model checkpoint and place it at e.g. ./trained_models/abdomen_foundation.pth and ./trained_models/downstream_seg.pth

1. Test the segmentation model on the sample images

   python run_inference.py -c ./config_files/downseg_20.toml

   The results will be stored at ./data/inference/downstream

2. Test the segmentation model on your images

   You need to change the config file to set the test dataset. The method is to change the split_filename in the Inference section of the configuration file downseg_20.toml. If you want to create a new split file, you can use ./tools/split_file_creator.py

Downsteam Segmentation Network Training

Preprocess the training data with ./tools/preprocess.py

Create a new split file with ./tools/split_file_creator.py

Firstly, set the split_filename in the Training section of the configuration file downseg_20.toml.

Run training with:

python run_training.py -c ./config_files/downseg_20.toml