Overview
This page provides a technical overview of BCRBench: A breast cancer risk benchmark, including its scope, objectives, and supported modeling paradigms.
🔬 Scope
The framework focuses on image-based breast cancer risk prediction from mammography and supports:
- Single-view and multi-view inputs
- Single-timepoint and longitudinal data
- Breast-wise and patient-wise modeling
🎯 Objectives
The main goals of this benchmark are:
- Standardizing model implementations across methods
- Enabling fair and reproducible comparisons
- Providing a unified research framework
- Accelerating progress in breast cancer risk prediction
📊 Supported Tasks
The framework supports:
- End-to-end model training
- Risk prediction evaluation
- Consistent benchmarking across architectures
🧠 Model Overview
The benchmark includes the following models:
| Model | Venue | Paper | Input | Key Idea |
|---|---|---|---|---|
| LMV-Net | MICCAI 2026 | - | Multi-view, longitudinal | Multi-view longitudinal model with dual-stream attention leveraging both views of one breast. |
| ImgFeatAlign | MICCAI 2025 | Paper | Single-view, longitudinal | Uses image-based deformation (MammoRegNet) applied in feature space for improved longitudinal comparison. |
| VMRA-MaR | MICCAI 2025 | Paper | Multi-view, longitudinal | Extends Mirai to longitudinal mammograms using Spatial Asymmetry Detector and Longitudinal Asymmetry Tracker. |
| OA-BReaCR | MICCAI 2024 | Paper | Single-view, longitudinal | Learns longitudinal changes using feature-based deformation fields for better temporal alignment. |
| Mirai | Sci. Transl. Med. 2021 | Paper | Multi-view, single-timepoint | Learns breast cancer risk from all four views in a single visit |
🏗️ Framework Design
All models in this benchmark share:
- A unified data interface
- A standardized training pipeline
- Consistent evaluation metrics
This ensures fair and reproducible comparison across methods.
🚀 Extensibility
The framework is designed to be easily extended:
- Add new models with minimal changes
- Integrate new datasets
- Extend training and evaluation pipelines
📚 Next Steps
- See Datasets for data preparation details
- See Models for architecture-level documentation
- See Getting Started to run experiments