![[2602.13930] MamaDino: A Hybrid Vision Model for Breast Cancer 3-Year Risk Prediction](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13930] MamaDino: A Hybrid Vision Model for Breast Cancer 3-Year Risk Prediction
Summary
MamaDino is a novel hybrid vision model that enhances breast cancer risk prediction by utilizing lower-resolution mammograms while maintaining high accuracy through advanced machine learning techniques.
Why It Matters
This research addresses the need for personalized breast cancer screening strategies by demonstrating that effective risk prediction can be achieved with lower-resolution images. This could lead to more accessible screening methods and improved patient outcomes, particularly in resource-limited settings.
Key Takeaways
- MamaDino combines convolutional and transformer-based models for improved risk prediction.
- The model operates effectively on lower-resolution mammograms, reducing resource needs.
- Explicit contralateral asymmetry modeling enhances prediction accuracy.
- MamaDino matches the performance of existing high-resolution models like Mirai.
- The study's findings support the shift towards personalized breast cancer screening.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.13930 (cs) [Submitted on 14 Feb 2026] Title:MamaDino: A Hybrid Vision Model for Breast Cancer 3-Year Risk Prediction Authors:Ruggiero Santeramo, Igor Zubarev, Florian Jug View a PDF of the paper titled MamaDino: A Hybrid Vision Model for Breast Cancer 3-Year Risk Prediction, by Ruggiero Santeramo and 2 other authors View PDF HTML (experimental) Abstract:Breast cancer screening programmes increasingly seek to move from one-size-fits-all interval to risk-adapted and personalized strategies. Deep learning (DL) has enabled image-based risk models with stronger 1- to 5-year prediction than traditional clinical models, but leading systems (e.g., Mirai) typically use convolutional backbones, very high-resolution inputs (>1M pixels) and simple multi-view fusion, with limited explicit modelling of contralateral asymmetry. We hypothesised that combining complementary inductive biases (convolutional and transformer-based) with explicit contralateral asymmetry modelling would allow us to match state-of-the-art 3-year risk prediction performance even when operating on substantially lower-resolution mammograms, indicating that using less detailed images in a more structured way can recover state-of-the-art accuracy. We present MamaDino, a mammography-aware multi-view attentional DINO model. MamaDino fuses frozen self-supervised DINOv3 ViT-S features with a trainable CNN encoder at 512x512 resolution, and aggregates bi...
