🫁 Lung Nodule Detection with MONAI Deploy

This project implements a full pipeline for working with medical images in DICOM format, using MONAI Deploy and a pretrained model for lung nodule detection on CT scans.

---

📦 Project Setup

To use this repository, you need to set up a Python virtual environment with MONAI and all required libraries.

1. Activate the MONAI Environment

In your terminal:

.\your_folder_name\Scripts\python.exe

Then activate the environment:

• On Command Prompt or PowerShell:

.\your_folder_name\Scripts\activate

• On Git Bash:

source your_folder_name/Scripts/activate

2. Install Dependencies

Make sure you're inside the activated environment. Then run:

pip install -r requirements.txt

The requirements.txt file should be located at the root of your project.

---

🗃 Dataset: LUNA16

This project uses the LUNA16 dataset for lung CT scans. You must download it to access .mhd image files and convert them to DICOM for further processing.

🔗 Download link:
https://zenodo.org/records/3723295

After downloading, convert the .mhd files to DICOM format by running the following command:

python convert_image.py

🗂️ The converted DICOM files will be saved in the /dicom folder.

---

🧠 Pretrained Model: Lung Nodule Detection

We use MONAI Model Zoo’s pretrained model for lung nodule detection.

Clone the Repository

git clone https://github.com/Project-MONAI/model-zoo.git

Navigate to the Model Directory

cd model-zoo/models/lung_nodule_ct_detection

Copy the Pretrained Model to Your Project

cp model.pt /path/to/your/project

Make sure model.pt is located at the root of your project directory.

---

🚀 Build and Start Containers

Before executing the full pipeline, you need to build and start the required Docker containers.

Create Docker Network

First, create a Docker network so all containers can communicate:

docker network create lung_net

1. Lung Nodule Container

The Lung Nodule container is built from the Dockerfile provided in the project.

• Build the container:

docker build -t lung_nodule -f Dockerfile .

• Start the container in the created network:

docker run -d --name lung_nodule --network lung_net lung_nodule

2. Orthanc Container

The Orthanc container is built from the pre-built image jodogne/orthanc.

• Pull the Orthanc image:

docker pull jodogne/orthanc

• Run the Orthanc container in the created network:

docker run -d --name orthanc --network lung_net -p 8042:8042 jodogne/orthanc

3. MONAI Deploy Container

The MONAI Deploy container is based on the default MONAI Deploy image.

• Pull the MONAI Deploy image:

docker pull monai/monai-deploy

• Run the MONAI Deploy container in the created network:

docker run -d --name monai_deploy --network lung_net monai/monai-deploy

---

🏃‍♂️ Run Full Pipeline

Once the containers are up and running, follow these steps to execute the full pipeline.

1. Upload Images to Docker Container

Before running the full pipeline, you need to upload images to the lung_nodule container. This can be done using curl.exe:

curl -X POST -F "file=@uploads/image.dcm" http://localhost:5000/upload

🗂️ The images to be uploaded should be stored in the /uploads folder.

2. Run app.py in Docker Lung Nodule

Run the app.py script inside the Lung Nodule Docker container to start the image processing pipeline.

docker exec -it lung_nodule python app.py

3. Execute orthanc_client.py to Upload Images to Orthanc

Run orthanc_client.py inside the Lung Nodule container to upload processed images to Orthanc:

docker exec -it lung_nodule python orthanc_client.py

4. Run the Full Pipeline

Once the images are uploaded to Orthanc and processed by MONAI, you can execute the full pipeline:

docker exec -it lung_nodule python pipeline.py

🗂️ The images downloaded from Orthanc will be saved in the /downloads folder.
🗂️ The images processed by MONAI will be saved in the /infered folder.

---

🔍 View Processed Images

After executing the full pipeline, you can view the processed images in Orthanc by navigating to the following URL in your browser:

http://localhost:8042/instances

---

License

This project is licensed under the MIT License - see the LICENSE file for details.