The Medical Image Denoising using Denoising Autoencoders in TensorFlow and Hexagonal Architecture project enhances medical image quality by removing noise using a Denoising Autoencoder (DAE) built with TensorFlow. Designed using Hexagonal Architecture (Ports & Adapters), it ensures modularity, scalability, and flexibility.
Hexagonal Architecture (also called Ports and Adapters) ensures that the core logic of your application remains independent of external frameworks (e.g., TensorFlow, Flask, or specific data sources).
This means:
✅ Core Logic (Domain) contains the ML model, training, and inference logic.
✅ Ports (Interfaces) define how external systems interact with the core.
✅ Adapters implement those interfaces, connecting the model to various data sources, APIs, or frameworks.
Important Folders and Files.
Med-Noise-Cleanse/ │── src/ │ ├── core/ # Core ML Logic (Domain) │ │ ├── model.py # Autoencoder model │ │ ├── training.py # Training logic │ │ ├── inference.py # Inference logic │ ├── ports/ # Interfaces (Ports) │ │ ├── data_loader.py # Data handling interface │ │ ├── trainer.py # Training interface │ │ ├── predictor.py # Inference interface │ ├── adapters/ # Concrete implementations (Adapters) │ │ ├── local_data.py # Implements data_loader (Local Images) │ │ ├── model_trainer.py # Implements trainer (TensorFlow) │ │ ├── predictor_api.py # Implements predictor (Flask API) │── app/ │ ├── api.py # Flask/FastAPI for Model Serving │ ├── frontend.py # Streamlit UI for real-time image denoising │── config/ │ ├── settings.py # Configurations (Paths, Hyperparameters) │ ├── logging_config.py # Centralised Logger │── tests/ │ ├── test_training.py # Unit tests for training │ ├── test_inference.py # Unit tests for inference │── requirements.txt # Dependencies │── README.md
-
Flexibility: You can switch between different storage systems, APIs, or models without modifying core logic.
-
Testability: The domain logic is decoupled, making unit testing easier.
-
Maintainability: The system remains scalable with well-defined boundaries between components.
✅ Autoencoder-based Model for noise reduction
✅ Hexagonal Architecture for modular design
✅ REST API for inference using Flask
✅ Configurable Hyperparameters
✅ Scalable & Extendable
git clone https://github.com/vijayagopalsb/Med-Noise-Cleanse.git
cd Med-Noise-Cleansepython3 -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activatepip install -r requirements.txt-
Download a publicly available Brain MRI Dataset (e.g., from Kaggle).
-
Place images inside dataset_path/ folder.
python src/adapters/model_trainer.pypython app/api.py**Send an image for prediction using cURL
curl -X POST -F "image=@/path/to/mri_image.png" http://127.0.0.1:5000/predict --output output.png| Method | Endpoint | Description |
|---|---|---|
| POST | /predict | Denoise an MRI image |
Run unit tests using
pytest tests/-
Fork the repo
-
Create a new branch
-
Make improvements
-
Submit a Pull Request (PR)
This project is open-source and available under the MIT License.
Email: vijayagopal.sb@gmail.com
GitHub: vijayagopalsb