EDAM could diagnose and subtype hepatic echinococcosis (HE) patients at the patient level using plain CT images as input.

You can use EDAM in two ways, slice-level prediction (diagnosing and generating segmentation for a single slice), and patient-level diagnosis (diagnosing a person based on the 3D CT image).

• CT images in .nii format should be placed under data-input/patient-example/. We provide an example of a CT image of a patient diagnosed as hepatic cysts (CE). This is an example for patient-level diagnosis.
• CT slices in .png format should be placed under data-input/slice-example/. We provide an example of a CT slice extracted from the CT image. This is used for slice-level prediction.
• All outputs and temp outputs (liver segmentation, lesion segmentation, and slice attention map) are stored under results/.

• Linux
• NVIDIA GPU (Tested on Nvidia GeForce RTX 3090 x 4)
• Python (3.7.9), Imageio (2.16.1), Matplotlib (3.5.1), Numpy (1.21.2), pencv_python (4.5.5.64), Pillow (9.2.0), Scikit_image (0.19.2), Scikit_learn (1.1.2), Scipy (1.7.3), Seaborn (0.11.2), SimpleITK (2.2.0), Skimage (0.0), Torch (1.11.0), Torchvision (0.12.0), Tqdm (4.63.0).

1. Create a virtual environment and install PyTorch. In the 3rd step, please select the correct Pytorch version that matches your CUDA version from https://pytorch.org/get-started/previous-versions/.

$ conda create -n EDAM python=3.7.9
$ conda activate EDAM
$ pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113

Note: pip install command is required for Pytorch installation.

2. To try out the Python code and set up environment, please activate the EDAM environment first:

$ conda activate pathfinder
$ cd EDAM/

For ease of use, you can just set up the environment and run the following:

$ pip install -r requirements.txt

• 3D CT images, which stored as .nii files under data-input/patient-example/, are first processed by the liver segmentaton model, to generate masks of liver field. The segmentation map and the CT slices within liver field would be generated and saved under results/extracted-liver/filename/. The process can be done by calling:

$ python liver-segmentation.py

• EDAM predicts each CT slice from the above step by slice-level prediction, and generates a label sequence. Then it processes the label sequence and gives a prediction for the patient. Directly calling:

$ python patient-diagnosis.py

• EDAM also generates a slice-attention map for the CT image, which is visualised by using show_slice_attention_map.
• The most salient slice as well as the liver and lesion segmentation are illustrated by using show_salience_segmentation.

• EDAM supports the lesion segmentation and classification of single CT slice, which stored as .png files under data-input/slice-example/, by calling:

$ python single-slice-prediction.py

• We provide simple and user-friendly Jupyter notebook patient-diagnosis.ipynb as a quick diagnosing demo to visualize prediction, segmentaiton, and slice attention map.
• Please click to download pre-trained models, and place the file in module/.
• Before you launch the notebooks, please configure an environment following the instruction in Environment Configuration.
• Then, you can launch the notebook and view the diagnosis from EDAM.