Research map / Vision-language and multimodal / CLIP and zero-shot
Ultrasound-CLIP: Semantic-Aware Contrastive Pre-training for Ultrasound Image-Text Understanding
Jiayun Jin, Haolong Chai, Xueying Huang, Xiaoqing Guo, Zengwei Zheng, Zhan Zhou, Junmei Wang, Xinyu Wang, Jie Liu, Binbin Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add UDAF-guided semantic loss and graph encoder to baseline CLIP training (ablation) | Ultrasound-CLIP · US-365K retrieval (I2T) | Recall@10: 0.2542 -> 0.3745 | Improved |
| add UDAF-guided semantic loss to graph-encoder-only model (ablation) | Ultrasound-CLIP · US-365K retrieval (I2T) | Recall@10: 0.2683 -> 0.3745 | Improved |
| add heterogeneous graph encoder to semantic-loss-only model (ablation) | Ultrasound-CLIP · US-365K retrieval (I2T) | Recall@10: 0.3444 -> 0.3745 | Improved |
| add both components versus neither (ablation) | Ultrasound-CLIP · US-365K multi-task classification | average accuracy (%): 50.84 -> 59.61 | Improved |
| add graph encoder to semantic-loss-only model (ablation) | Ultrasound-CLIP · US-365K multi-task classification | average accuracy (%): 49.87 -> 59.61 | Improved |
| add semantic loss to graph-encoder-only model (ablation) | Ultrasound-CLIP · US-365K multi-task classification | average accuracy (%): 48.62 -> 59.61 | Improved |
| pretrain on US-365K with Ultrasound-CLIP instead of BiomedCLIP | BiomedCLIP · US-365K multi-task classification | average accuracy (%): 33.81 -> 59.61 | Improved |
| pretrain with Ultrasound-CLIP versus BiomedCLIP | BiomedCLIP · US-365K diagnosis classification | accuracy (%): 39.40 -> 64.05 | Improved |
| Ultrasound-CLIP versus BiomedCLIP in linear probe | BiomedCLIP · Downstream 4 datasets (linear probe) | average accuracy (%): 74.14 -> 75.40 | Improved |
| Ultrasound-CLIP versus UniMed-CLIP in zero-shot | UniMed-CLIP · Downstream 4 datasets (zero-shot) | average accuracy (%): 49.44 -> 54.42 | Improved |
| Ultrasound-CLIP versus SigLIP on BreastMNIST zero-shot | SigLIP · BreastMNIST (zero-shot) | accuracy (%): 54.30 -> 39.16 | Hurt |
Closest papers by meaning
- U2-BENCH: Benchmarking Large Vision-Language Models on Ultrasound Understanding (ICLR 2026 · similarity 0.90)
- Anatomical Knowledge Mining and Matching for Semi-supervised Medical Multi-structure Detection (AAAI 2025 · similarity 0.87)
- Unsupervised Domain Adaptation for Anatomical Structure Detection in Ultrasound Images (ICML 2024 · similarity 0.87)
- Leveraging Anatomical Consistency for Multi-Object Detection in Ultrasound Images via Source-free Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.87)
- Bringing CLIP to the Clinic: Dynamic Soft Labels and Negation-Aware Learning for Medical Analysis (CVPR 2025 · similarity 0.86)
- FETAL-GAUGE: A BENCHMARK FOR ASSESSING VISION-LANGUAGE MODELS IN FETAL ULTRASOUND (ICLR 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
