Automated Melanoma Image Generation and Classification Project

Project Description

This project aims to automatically generate and classify high-quality melanoma images using the Denoising Diffusion Probabilistic Model (DDPM) and ResNet50 to facilitate early diagnosis and improve treatment outcomes. Through efficient and accurate image processing and analysis techniques, this project enables medical professionals to better understand and diagnose melanoma, and thus provide more personalized and effective treatment plans.

Dataset Sources

The dataset for this project comes from the International Skin Imaging Collaboration (ISIC), which provides a large collection of expert-reviewed melanoma images to support dermatology research and education. ISIC:https://challenge.isic-archive.com/data/

Environment requirements

• Python 3.8+
• PyTorch 1.8+
• torchvision
• denoising_diffusion_pytorch (denoising diffusion modeling library)
• See requirements.txt for additional dependencies.

Installation steps

1. Clone the repository locally:

   git clone https://github.com/WuTao1103/Melanoma-Image-Generation-and-Classification.git
   

2. Enter the project directory: ``bash cd Melanoma Image Generation and Classification cd Melanoma Image Generation and Classification
3. Install the dependencies (make sure you have pip installed):

   pip install -r requirements.txt

DDPM Image Generation

Please refer to README.md in the ddpm folder for detailed prediction and training steps.

Prediction Steps

• Generate images directly using pre-trained weights or use your own training weights.
• The generated images will be saved in the results/predict_out directory.

Training step

• Place the expected image files in the datasets folder.
• Run the train.py file for training. The images generated during training can be viewed in the results/train_out folder.

ResNet50 image classification

See README.md in the skin_two_classification folder for detailed steps.

Introduction to the code directory

• args.py: holds the various parameters used for training and testing.
• data_gen.py: implements dataset partitioning, data augmentation and data loading.
• main.py: contains training, evaluation and testing.
• models/Res.py: rewrites ResNet for various types of networks.

Run commands

• Training mode: python main.py --mode=train.
• Test mode: python main.py --mode=test --model_path='path/to/your/model.pth'

Contribution guidelines

We welcome all forms of contributions, including bug fixes, feature additions, or improvements to the project documentation. Please contact us via issue or pull request.

License

This project is under the MIT license. See the LICENSE file for more details.

Acknowledgments

We thank the International Skin Imaging Collaboration (ISIC) for providing valuable datasets, as well as all the individuals and organizations that contributed to this project.