Anatomy-guided visual prompt tuning for cross-modal breast cancer understanding - Takeaways - MDSpire

Anatomy-guided visual prompt tuning for cross-modal breast cancer understanding

  • By

  • Shaorong Zhao

  • Qingxiang Meng

  • Yang He

  • Xiaotong Xu

  • Jiayao Zhu

  • Jiawen Qiu

  • Chao Wu

  • Yamei Han

  • Jinhai Deng

  • Teng Pan

  • Jingjing Liu

  • February 13, 2026

  • 0 min

Share

  • 1

    A-VPT integrates anatomical structures into the prompt space of a frozen Vision Transformer for improved breast cancer detection.

  • 2

    The framework generates tissue-aware prompts based on glandular, fatty, and ductal region embeddings for enhanced model performance.

  • 3

    A cross-modal contrastive alignment strategy harmonizes anatomical semantics across mammography, ultrasound, and MRI.

  • 4

    A-VPT achieves state-of-the-art performance in lesion classification and segmentation with less than 2% of the parameters needed for full fine-tuning.

  • 5

    Embedding anatomical priors into prompt tuning enhances efficiency, generalization, and provides interpretable connections to human anatomical reasoning.

Original Source(s)

Anatomy-guided visual prompt tuning for cross-modal breast cancer understanding

  • by Shaorong Zhao, Qingxiang Meng, Yang He, Xiaotong Xu, Jiayao Zhu, Jiawen Qiu, Chao Wu, Yamei Han, Jinhai Deng, Teng Pan, Jingjing Liu

  • February 13, 2026

Related Content

New Mammography Screening Guidance: Clinical Considerations for Risk-Based Care

New guidance from the American College of Physicians (ACP) recommends biennial mammography screening for average-risk women ages 50 to 74.

  • by

  • May 26, 2026

  • 5 min
Comparison of the diagnostic power of serum circulating miR-21, VEGF, and CA15-3 in breast cancer

Peripheral blood monocytes could be associated with brain metastasis and affect patient prognosis in breast cancer: a retrospective study