Research map / Label-efficient and robust learning / Self-supervised and contrastive
An OpenMind for 3D Medical Vision Self-supervised Learning
Tassilo Wald, Constantin Ulrich, Jonathan Suprijadi, Sebastian Ziegler, Michal Nohel, Robin Peretzke, Gregor Kohler, Klaus Maier-Hein
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 MAE pre-training before fine-tuning | ResEnc-L CNN · OpenMind segmentation mean (150 epochs) | mean DSC (%): 68.44 -> 70.91 | Improved |
| add MAE pre-training before fine-tuning | Primus-M Transformer · OpenMind segmentation mean (150 epochs) | mean DSC (%): 63.62 -> 70.42 | Improved |
| add VoCo contrastive pre-training | Primus-M Transformer · OpenMind segmentation mean (150 epochs) | mean DSC (%): 63.62 -> 57.61 | Hurt |
| add VoCo contrastive pre-training | ResEnc-L CNN · OpenMind segmentation mean (150 epochs) | mean DSC (%): 68.44 -> 68.50 | No clear effect |
| add VoCo pre-training | ResEnc-L CNN · OpenMind classification | mean average precision (%): 62.97 -> 68.18 | Improved |
| add MAE pre-training | ResEnc-L CNN · OpenMind classification | mean average precision (%): 62.97 -> 63.57 | No clear effect |
| fine-tune CNN 1000 epochs instead of 150 (ablation) | ResEnc-L CNN MAE · OpenMind segmentation mean | mean DSC (%): 70.91 -> 71.63 | Improved |
| fine-tune CNN 1000 epochs instead of 150 (ablation) | ResEnc-L CNN VoCo · OpenMind segmentation mean | mean DSC (%): 68.50 -> 69.36 | Improved |
| freeze encoder during fine-tuning (vs default) (ablation) | ResEnc-L CNN · OpenMind segmentation, 4 development datasets | average DSC (%): 76.15 -> 55.25 | Hurt |
| freeze encoder during fine-tuning (vs default) (ablation) | Primus-M Transformer · OpenMind segmentation, 4 development datasets | average DSC (%): 68.36 -> 30.38 | Hurt |
| add Sawtooth fine-tuning schedule instead of default (ablation) | ResEnc-L CNN · OpenMind segmentation, 4 development datasets | average DSC (%): 76.15 -> 76.60 | Improved |
| add warm-up to default fine-tuning (ablation) | Primus-M Transformer · OpenMind segmentation, 4 development datasets | average DSC (%): 68.36 -> 68.76 | Improved |
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.
- VoCo — VoCo: A Simple-yet-Effective Volume Contrastive Learning Framework for 3D Medical Image Analysis (CVPR 2024) · better on 2 of 2 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
- Revisiting MAE Pre-training for 3D Medical Image Segmentation (CVPR 2025 · similarity 0.94)
- Autoregressive Sequence Modeling for 3D Medical Image Representation (AAAI 2025 · similarity 0.88)
- Enhancing vision-language models for medical imaging: bridging the 3D gap with innovative slice selection (NeurIPS 2024 · similarity 0.86)
- Multi-modal Vision Pre-training for Medical Image Analysis (CVPR 2025 · similarity 0.86)
- Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imaging (NeurIPS 2025 · similarity 0.85)
- How Well Do Supervised 3D Models Transfer to Medical Imaging Tasks? (ICLR 2024 · 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.
