Research map / Vision-language and multimodal / CLIP and zero-shot
Comprehensive language–image pre-training for 3D medical image understanding
Tassilo Wald, Ibrahim Ethem Hamamci, Yuan Gao, Sam Bond-Taylor, Harshita Sharma, Maximilian Ilse, Cynthia Lo, Olesya Melnichenko, Anton Schwaighofer, Noel Codella, Maria Teodora Wetscherek, Klaus Maier-Hein and 4 more
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 opposite sentence loss (ablation) | COLIPRI-CRM · Zero-shot short prompts, CT-RATE val | AUROC: 74.09 -> 81.66 | Improved |
| add report generation objective | COLIPRI-C · CT-RATE classification probe | AUPRC: 55.13 -> 57.44 | Improved |
| add MAE objective | COLIPRI-C · CT-RATE classification probe | AUPRC: 55.13 -> 61.52 | Improved |
| add report generation on top of MAE | COLIPRI-CM · CT-RATE retrieval | Recall@5: 34.16 -> 35.10 | Improved |
| add report generation on top of MAE | COLIPRI-CM · CT-RATE zero-shot | AUROC: 80.52 -> 79.81 | No clear effect |
| add MAE objective | COLIPRI-C · LiTS segmentation (37.5k steps) | DSC: 76.77 -> 79.97 | Improved |
| pre-train with COLIPRI-CRM instead of training from scratch | Primus-M · Lung segmentation (250k steps) | DSC: 69.11 -> 67.56 | Hurt |
| add sentence shuffle augmentation (ablation) | default CLIP · CT-RATE val retrieval | Recall@5: 22.64 -> 28.57 | Improved |
| add short sentence augmentation p=0.10 (ablation) | default CLIP · CT-RATE val short zero-shot | AUROC: 66.91 -> 71.15 | Improved |
| train on impressions instead of findings (ablation) | default CLIP · CT-RATE val short zero-shot | AUROC: 66.91 -> 62.40 | Hurt |
| re-translate reports with GPT-4o (ablation) | default CLIP · CT-RATE val short zero-shot | AUROC: 66.91 -> 60.11 | Hurt |
| use BiomedCLIP text encoder instead of CXR-BERT (ablation) | default CLIP · CT-RATE val retrieval | Recall@5: 22.64 -> 13.28 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Merlin — Merlin: Empowering Multimodal LLMs with Foresight Minds (ECCV 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imaging (NeurIPS 2025 · similarity 0.92)
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.90)
- Enhancing vision-language models for medical imaging: bridging the 3D gap with innovative slice selection (NeurIPS 2024 · similarity 0.89)
- VILA-M3: Enhancing Vision-Language Models with Medical Expert Knowledge (CVPR 2025 · similarity 0.88)
- Toward a Vision-Language Foundation Model for Medical Data: Multimodal Dataset and Benchmarks for Vietnamese PET/CT Report Generation (NeurIPS 2025 · similarity 0.88)
- G2D: From Global to Dense Radiography Representation Learning via Vision-Language Pre-training (NeurIPS 2024 · similarity 0.88)
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.
