This project implements a 3D Fast Anomaly GAN (f-AnoGAN) for detecting anomalies in prostate CT scans. The implementation is inspired by the original f-AnoGAN paper [1] and extends it to work with 3D medical imaging data.
- 3D Wasserstein GAN with Gradient Penalty (WGAN-GP) implementation
- Spectral normalization for training stability
- Adaptive gradient penalty weighting
- Comprehensive anomaly scoring system
- Separate reconstruction and feature-based error metrics
- Multiple thresholding methods for anomaly detection
- Visualization tools for analysis
The current implementation includes:
- A Generator network with 3D convolutional layers
- A Discriminator with feature extraction capabilities
- An Encoder network for mapping to latent space
- Advanced stabilization techniques for GAN training
- Batch processing for memory efficiency
- GPU acceleration support
- Configurable model parameters
- Multiple evaluation metrics
- Enhanced architecture with residual connections
- Self-attention mechanisms
- Progressive growing capabilities
- Additional visualization tools
- Performance optimization
- Extended documentation
[1] Schlegl, T., Seeböck, P., Waldstein, S.M., Langs, G., & Schmidt-Erfurth, U. (2019). f-AnoGAN: Fast unsupervised anomaly detection with generative adversarial networks. Medical Image Analysis, 54, 30-44.
Last Updated: January 2025