When Vision-Language Models Meet Fetal Cardiac Ultrasound: Dual-Level Contrastive Learning for Out-of-Distribution Detection

Liu, Ziyi; Song, Weihu; Ma, Yupeng; Liu, Tianrui; Gu, Yifan; Zhu, Haogang

doi:10.24963/ijcai.2026/756

When Vision-Language Models Meet Fetal Cardiac Ultrasound: Dual-Level Contrastive Learning for Out-of-Distribution Detection

Ziyi Liu, Weihu Song, Yupeng Ma, Tianrui Liu, Yifan Gu, Haogang Zhu

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. AI and Health. Pages 6797-6806.

https://doi.org/10.24963/ijcai.2026/756

Keywords

Multimodal data: Multimodal data
Medical imaging: Medical imaging

Abstract

Recent advances in vision-language models (VLMs) have shown remarkable performance in medical image classification tasks. However, applying VLMs to fetal cardiac ultrasound (FCU) remains challenging due to compound distribution shifts, including covariate shifts caused by cross-center heterogeneity and semantic shifts arising from clinically non-standard views. To address this issue, we propose Dual-Level Contrastive Learning (DLCL), the first prompt-based VLM framework for out-of-distribution (OOD) detection in FCU, to the best of our knowledge. DLCL explicitly shapes the vision-language representation space through complementary local-level and global-level contrastive objectives. Specifically, local contrastive learning aligns instance-level features to mitigate covariate shifts, whereas global contrastive learning regularizes global prototypes to address semantic shifts. We conduct extensive experiments on a private multi-center FCU dataset and the public ISIC-OOD dataset to validate the proposed approach. On the challenging FCU task, DLCL achieves an AUROC of 89.61% and a harmonic mean of 80.35%, significantly outperforming state-of-the-art methods.

Ziyi Liu, Weihu Song, Yupeng Ma, Tianrui Liu, Yifan Gu, and Haogang Zhu. 2026. When Vision-Language Models Meet Fetal Cardiac Ultrasound: Dual-Level Contrastive Learning for Out-of-Distribution Detection. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 6797–6806. https://doi.org/10.24963/ijcai.2026/756

@inproceedings{ijcai2026p756,
  title     = {When Vision-Language Models Meet Fetal Cardiac Ultrasound: Dual-Level Contrastive Learning for Out-of-Distribution Detection},
  author    = {Liu, Ziyi and Song, Weihu and Ma, Yupeng and Liu, Tianrui and Gu, Yifan and Zhu, Haogang},
  booktitle = {Proceedings of the Thirty-Fifth International Joint Conference on
               Artificial Intelligence, {IJCAI-26}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Diego Calvanese},
  pages     = {6797--6806},
  year      = {2026},
  month     = {8},
  note      = {AI and Health},
  doi       = {10.24963/ijcai.2026/756},
  url       = {https://doi.org/10.24963/ijcai.2026/756},
}

TY  - CPAPER
TI  - When Vision-Language Models Meet Fetal Cardiac Ultrasound: Dual-Level Contrastive Learning for Out-of-Distribution Detection
AU  - Liu, Ziyi
AU  - Song, Weihu
AU  - Ma, Yupeng
AU  - Liu, Tianrui
AU  - Gu, Yifan
AU  - Zhu, Haogang
T2  - Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2026
SP  - 6797
EP  - 6806
DO  - 10.24963/ijcai.2026/756
UR  - https://doi.org/10.24963/ijcai.2026/756
ER  -