Research map / 3D vision / 3D generation
Revisiting 2D Foundation Models for Scalable 3D Medical Image Classification
Han Liu, Bogdan Georgescu, Yanbo Zhang, Youngjin Yoo, Michael Baumgartner, Riqiang Gao, Jianing Wang, Gengyan Zhao, Eli Gibson, Dorin Comaniciu, Sasa Grbic
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 |
|---|---|---|---|
| replace linear probing with AnyMC3D adaptation (ablation) | MII · 10-task average | average AUC: 0.785 -> 0.894 | Improved |
| replace linear probing with AnyMC3D adaptation (ablation) | MedGemma · 10-task average | average AUC: 0.690 -> 0.866 | Improved |
| use DINOv3 backbone instead of MedGemma under AnyMC3D | AnyMC3D · 10-task average | average AUC: 0.866 -> 0.894 | Improved |
| use DINOv3 instead of medical MII under AnyMC3D | AnyMC3D · 10-task average | average AUC: 0.894 -> 0.894 | No clear effect |
| use DINOv3 instead of DINOv2 (ablation) | AnyMC3D · 3D medical classification | AUC | No clear effect |
| add auxiliary pixel-level supervision (ablation) | AnyMC3D (DINOv3) · PDAC early detection T5 | AUC: 0.962 -> 0.973 | Improved |
| use AnyMC3D (DINOv3) instead of PanDx | PDAC detection · T5 | AUC: 0.949 -> 0.962 | Improved |
| use full finetuning instead of linear probing (ablation) | MedicalNet · T4 | AUC: 0.654 -> 0.899 | Improved |
| use full finetuning instead of linear probing (ablation) | VoCo · T4 | AUC: 0.702 -> 0.919 | Improved |
| use LoRA instead of linear probing (ablation) | MedicalNet · T4 | AUC: 0.654 -> 0.889 | Improved |
| use LoRA instead of linear probing (ablation) | VoCo · T4 | AUC: 0.702 -> 0.838 | Improved |
| use AnyMC3D with 20% training data instead of 3D DenseNet | T3 low-data · T3 (20% data) | AUC: 0.741 -> 0.924 | Improved |
Closest papers by meaning
- VISTA3D: A Unified Segmentation Foundation Model For 3D Medical Imaging (CVPR 2025 · similarity 0.88)
- Autoregressive Sequence Modeling for 3D Medical Image Representation (AAAI 2025 · similarity 0.85)
- 3DMedAgent: Unified Perception-to-Understanding for 3D Medical Analysis (ICML 2026 · similarity 0.85)
- Revisiting MAE Pre-training for 3D Medical Image Segmentation (CVPR 2025 · similarity 0.85)
- Time-to-Event Pretraining for 3D Medical Imaging (ICLR 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.
