Research map / Label-efficient and robust learning / Self-supervised and contrastive
Capturing implicit hierarchical structure in 3D biomedical images with self-supervised hyperbolic representations
Joy Hsu, Jeffrey Gu, Gong Her Wu, Wah Chiu, Serena Yeung
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 |
|---|---|---|---|
| Use hyperbolic instead of Euclidean latent space (base) (ablation) | 3D conv VAE · synthetic Level 1 | Dice: 0.784 -> 0.832 | Improved |
| Use hyperbolic instead of Euclidean latent space (base) (ablation) | 3D conv VAE · synthetic Level 3 | Dice: 0.109 -> 0.135 | Improved |
| Add hierarchical triplet loss in hyperbolic space (ablation) | hyperbolic VAE · synthetic Level 2 | Dice: 0.352 -> 0.534 | Improved |
| Add hierarchical triplet loss in hyperbolic space (ablation) | hyperbolic VAE · synthetic Level 3 | Dice: 0.135 -> 0.222 | Improved |
| Add hierarchical triplet loss in Euclidean space (ablation) | Euclidean VAE · synthetic Level 1 | Dice: 0.784 -> 0.761 | Hurt |
| Add hierarchical triplet loss in Euclidean space (ablation) | Euclidean VAE · synthetic Level 3 | Dice: 0.109 -> 0.153 | Improved |
| Add gyroplane convolutional layer (ablation) | hyperbolic VAE · synthetic Level 1 | Dice: 0.832 -> 0.905 | Improved |
| Add gyroplane convolutional layer (ablation) | hyperbolic VAE · synthetic Level 2 | Dice: 0.352 -> 0.473 | Improved |
| Add gyroplane conv to hyperbolic + triplet model (ablation) | hyperbolic VAE + triplet · synthetic Level 3 | Dice: 0.222 -> 0.216 | No clear effect |
| Add gyroplane conv to hyperbolic + triplet model (ablation) | hyperbolic VAE + triplet · synthetic Level 1 | Dice: 0.945 -> 0.952 | Improved |
| Use hyperbolic unsupervised method instead of best prior unsupervised baseline | Ours · BraTS 2019 WT | Dice: 0.495 -> 0.684 | Improved |
| Use hyperbolic unsupervised method instead of best prior unsupervised baseline | Ours · BraTS 2019 | average Hausdorff distance: 87.704 -> 77.940 | Improved |
Closest papers by meaning
- Learning Generalizable 3D Medical Image Representations from Mask-Guided Self-Supervision (CVPR 2026 · similarity 0.87)
- LaGeM: A Large Geometry Model for 3D Representation Learning and Diffusion (ICLR 2025 · similarity 0.86)
- Learning Structured Representations with Hyperbolic Embeddings (NeurIPS 2024 · similarity 0.85)
- Revisiting MAE Pre-training for 3D Medical Image Segmentation (CVPR 2025 · similarity 0.85)
- Fully Hyperbolic Convolutional Neural Networks for Computer Vision (ICLR 2024 · similarity 0.85)
- Autoregressive Sequence Modeling for 3D Medical Image Representation (AAAI 2025 · 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.
