Research map / Label-efficient and robust learning / Self-supervised and contrastive
Beyond Instance-Level Self-Supervision in 3D Multi-Modal Medical Imaging
Tan Pan, Shuhao Mei, Yixuan Sun, Kaiyu Guo, Chen Jiang, Zhaorui Tan, Mengzhu Li, LIMEI HAN, Xiang Zou, Yuan Cheng, Mahsa Baktashmotlagh
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 Lintra and Linter topology-consistency losses to reconstruction pre-training (ablation) | Luni baseline · BraTS-MET | avg Dice (%): 61.63 -> 68.17 | Improved |
| add Lintra and Linter losses to reconstruction pre-training (ablation) | Luni baseline · ADHD-200 | F1 (%): 63.06 -> 63.82 | Improved |
| add Lintra and Linter losses to reconstruction pre-training (ablation) | Luni baseline · ADHD-200 | ACC (%): 64.94 -> 67.53 | Improved |
| add only one of Lintra or Linter (ablation) | Luni baseline · ADHD-200 | F1 (%): 63.06 -> 57.61 (first row), 63.06 -> 61.44 (second row) | Mixed |
| switch from Swin UNETR-based BrainMVP to TACO pre-training | segmentation pre-training · four segmentation datasets average | Dice (%): 76.36 -> 77.47 | Improved |
| pre-train with TACO instead of best baselines on classification | Swin-B on ADNI and ADHD · average F1 | F1 (%): 72.76 -> 78.7 | Improved |
| TACO pre-training on ADHD-200 accuracy | Swin-B · ADHD-200 | ACC (%): 69.26 -> 67.53 | Hurt |
| use TACO pre-training for missing-modality segmentation | BraTS18 segmentation · whole tumor, all modality configurations | mean Dice (%): 87.63 -> 89.55 | Improved |
| use TACO pre-training for missing-modality segmentation | BraTS18 segmentation · tumor core, all modality configurations | mean Dice (%): 79.78 -> 79.68 | No clear effect |
| increase neighbourhood size Top-omega (ablation) | TACO · BraTS-MET and ADHD | downstream performance | Mixed |
| increase triplet margin delta (ablation) | TACO · BraTS-MET and ADHD | downstream performance | Mixed |
| TACO pre-training with fewer labeled samples (5-10%) | BraTS-MET segmentation and ADHD classification · label efficiency | performance | 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 1 of 1 shared result
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
- Multi-modal Vision Pre-training for Medical Image Analysis (CVPR 2025 · similarity 0.88)
- Towards a Universal 3D Medical Multi-modality Generalization via Learning Personalized Invariant Representation (ICCV 2025 · similarity 0.88)
- Learning Generalizable 3D Medical Image Representations from Mask-Guided Self-Supervision (CVPR 2026 · similarity 0.88)
- Representing Part-Whole Hierarchies in Foundation Models by Learning Localizability Composability and Decomposability from Anatomy via Self Supervision (CVPR 2024 · similarity 0.87)
- Modality-Agnostic Structural Image Representation Learning for Deformable Multi-Modality Medical Image Registration (CVPR 2024 · similarity 0.86)
- Knowledge Transfer Scaling Laws for 3D Medical Imaging (NeurIPS 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.
