We are thrilled to announce the inaugural release of Outlier Detection for Mammograms (ODM). The ODM is an open-source project designed to filter out low-quality and undesired scans from large mammogram collections. It employs a two-stage hybrid algorithm with unsupervised learning, making downstream analysis more efficient and accurate.
Key features of the initial release:
- **Two-Stage Hybrid Algorithm**: The first stage employs a threshold-based 5-bin histogram filtering (5BHIST) method, designed to exclude poor-quality images. The second stage uses a Variational Autoencoder (VAE), an unsupervised machine learning model, to further identify and remove outlier scans.
- **Robust Outlier Detection**: By combining conventional filtering techniques with advanced machine learning, ODM ensures comprehensive outlier detection in mammogram datasets.
- **GPU Support**: ODM supports GPU acceleration for the VAE stage of the pipeline. This significantly enhances the speed of processing.
- **Docker Support**: In addition to the standard installation, ODM also provides a Docker installation option, offering more flexibility and consistency across different systems.
- **Easy to Install and Use**: ODM can be easily installed via pip, making the setup process straightforward. With well-documented installation procedures and usage instructions, getting started with ODM is effortless.
- **Transparent Development**: Based on our published research paper, all the algorithms, strategies, and explanations are provided in the documentation, ensuring transparency in our approach.
The v1.0.0 release can be downloaded from our [GitHub repository](https://github.com/mpsych/ODM). We invite users to provide their feedback, bug reports, and suggestions via the GitHub issues system, which will help us further improve the ODM.
For more details, please refer to the Readme and other documentation included in the repository.
Thank you for your support, and we look forward to hearing from the user community!