MamaDino: A Hybrid Vision Model for Breast Cancer 3-Year Risk Prediction
Fondazione Human Technopole(人类技术极地基金会)
Ruggiero Santeramo, Igor Zubarev, Florian Jug
备注:
20 pages, 3 figures. Extended version of the peer-reviewed MICCAI 2026 conference paper, including additional architectural details, cohort characteristics, subgroup evaluations, and tumour-specific analyses
发表信息:
Medical Image Computing and Computer Assisted Intervention (MICCAI 2026), Lecture Notes in Computer Science, vol. 16887, pp. 418-427, Springer, 2026
Breast cancer screening programmes increasingly seek to move from one-size-fits-all intervals to risk-adapted and personalized strategies. Deep learning (DL) has enabled image-based risk models with stronger 1- to 5-year prediction than clinical models, but leading systems (e.g., Mirai) typically use convolutional backbones, 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 on lower-resolution mammograms, suggesting that using less detail 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 bilateral breast information via a BilateralMixer to output a 3-year breast cancer risk score. We train on 53,883 women from OPTIMAM (UK) and evaluate on matched 3-year case-control cohorts: an in-distribution test set from four screening sites and an external out-of-distribution cohort from an unseen site. At breast level, MamaDino achieves discrimination comparable to Mirai on internal and external tests while using ~13x fewer input pixels. Adding the BilateralMixer improves patient-level discrimination to AUC 0.736 (vs Mirai 0.713) in-distribution and 0.677 (vs Mirai 0.676) out-of-distribution, with overlapping confidence intervals. Performance was evaluated across age, ethnicity, scanner manufacturer, tumour type and grade. These findings suggest that complementary inductive biases and explicit contralateral modelling can achieve Mirai-comparable discrimination at lower resolution.
Towards Early Detection: AI-Based Five-Year Forecasting of Breast Cancer Risk Using Digital Breast Tomosynthesis Imaging
Athinoula A. Martinos Center for Biomedical Imaging, Mass General Brigham and Harvard Medical School(Athinoula A. Martinos 生物医学成像中心,马萨诸塞总医院与哈佛医学院);École Polytechnique Fédérale de Lausanne (EPFL)(洛桑联邦理工学院);Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt-Universität zu Berlin(柏林夏里特医学院,自由大学和洪堡大学的成员);Department of Radiology, University of Wisconsin School of Medicine and Public Health(威斯康星大学医学与公共卫生学院放射科)
Robust breast cancer detection in mammography and digital breast tomosynthesis using annotation-efficient deep learning approach
DeepHealth, Inc.(深健公司);Rhode Island Hospital(罗德岛医院);Brown University(布朗大学);Henan Provincial People’s Hospital(河南省人民医院);Medford Radiology Group(梅德福放射学集团);University of Massachusetts Medical School(马萨诸塞大学医学院)
Breast Cancer Image Classification Method Based on Deep Transfer Learning
The Hong Kong University of Science and Technology(香港科技大学);University of Southampton(南安普顿大学);Trine University(特莱恩大学);Southwest Jiaotong University(西南交通大学)
Cancerous Nuclei Detection and Scoring in Breast Cancer Histopathological Images
Department of Electrical and Electronics Engineering, Shiraz University of Technology(设拉子理工大学电气与电子工程系);Department of Pathology, School of Medicine, Shiraz University of Medical Sciences(设拉子医科大学医学院病理学系)
Pretrained Vision Models for Predicting High-Risk Breast Cancer Stage
McGill University(麦吉尔大学);Mila Quebec AI Institute(米拉魁北克人工智能研究所);Lelapa AI(莱拉帕人工智能);Masakhane Research Foundation(马萨卡内研究基金会);Drexel University(德雷塞尔大学);University Of Maryland Baltimore County(马里兰大学巴尔的摩县分校)
癌症日益成为全球性的健康问题。仅次于心血管疾病,癌症是世界上第二大死亡原因,每年有数百万人死于癌症。根据世界卫生组织(WHO)的报告,截至2020年底,超过780万女性被诊断患有乳腺癌,使其成为世界上最流行的癌症。在本文中,使用数字病理学(乳腺活检)图像的Nightingale Open Science数据集,我们利用预先训练的计算机视觉模型的功能进行乳腺癌分期预测任务。虽然个别模型实现体面的表现,我们发现,一个合奏模型的预测更有效,并提供了一个成功的解决方案\footnote{https://www.nightingalescience.org/updates/hbc1-results}。我们还提供了结果的分析,并探索更好的可解释性和泛化的途径。我们的代码在\url{https://github.com/bonaventuredossou/nightingale_winning_solution}上是开源的
A Lightweight CNN Integrated Compact Convolutional Transformer for Multi-Scale Feature Learning and reducing computational complexity for breast cancer mammography image detection and classification
Utilizing Automated Breast Cancer Detection to Identify Spatial Distributions of Tumor Infiltrating Lymphocytes in Invasive Breast Cancer
Stony Brook University(石溪大学);Stony Brook Medicine(石溪医学院);Stony Brook University Hospital(石溪大学医院);University of Michigan Medical School(密歇根大学医学院);National Cancer Institute(美国国家癌症研究所);National Institutes of Health(美国国立卫生研究院);Emory University(埃默里大学)
Deep Learning in Breast Cancer Imaging: A Decade of Progress and Future Directions
The Hong Kong University of Science and Technology(香港科技大学);Stanford University School of Medicine(斯坦福大学医学院);The Chinese University of Hong Kong(香港中文大学);The University of Hong Kong(香港大学);HKUST Shenzhen-Hong Kong Collaborative Innovation Research Institute(香港科技大学深圳-香港协同创新研究院)
Deep learning-based assessment of tumor-associated stroma for diagnosing breast cancer in histopathology images
Radboud University Medical Center(拉德堡德大学医学中心);Beth Israel Deaconess Medical Center(贝斯以色列女执事医疗中心);Harvard Medical School(哈佛医学院);National Cancer Institute(美国国家癌症研究所);The Harker School(哈克学校);Mayo Clinic(梅奥诊所)
Axial Attention Transformer Networks: A New Frontier in Breast Cancer Detection
University of California, Los Angeles(加州大学洛杉矶分校);Washington University in St. Louis(华盛顿大学圣路易斯分校);Johns Hopkins University(约翰斯·霍普金斯大学);Boston University(波士顿大学);Northeastern University(东北大学);Tulane University(杜兰大学)