Research map / Detection and segmentation / Open-set and few-shot detection
Neuroverse3D: Developing In-Context Learning Universal Model for Neuroimaging in 3D
Jiesi Hu, Hanyang Peng, Yanwu Yang, Xutao Guo, Yang Shang, Pengcheng Shi, Chenfei Ye, Ting Ma
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 modified cubic smooth-L1 segmentation loss (ablation) | Neuroverse3D · Held-out segmentation avg | Dice: 0.7977 -> 0.8173 | Improved |
| add gradient (intensity difference) loss (ablation) | Neuroverse3D · Held-out generation avg | PSNR: 26.67 -> 28.38 | Improved |
| add gradient loss (not applied to segmentation tasks) (ablation) | Neuroverse3D · Held-out segmentation avg | Dice: 0.8014 -> 0.8173 | Improved |
| add modified smooth-L1 loss (ablation) | Neuroverse3D · Held-out generation avg | PSNR: 28.19 -> 28.38 | No clear effect |
| increase context size (ablation) | Neuroverse3D · Held-out tasks | Dice / PSNR | Improved |
| extend fusion from deepest stage only to stages 5-1 (ablation) | Neuroverse3D · PET segmentation (unseen modality) | Dice: 0.2870 -> 0.5314 | Improved |
| extend fusion from deepest stage only to stages 5-1 (ablation) | Neuroverse3D · Held-out generation | PSNR: 25.82 -> 28.87 | Improved |
| use context image modality mismatched with target (ablation) | Neuroverse3D · 2D-to-3D, T1 target | PSNR: 26.63 -> 12.27 (CT context) | Hurt |
| train without target task (Neuroverse3D-unseen) for segmentation (ablation) | Neuroverse3D · Held-out unseen segmentation tasks | Dice: about 5-point decrease | Hurt |
| use 3D ICL model instead of 2D slice-wise ICL models | ICL models · Hippocampus, thalamus, lateral ventricle, putamen | Dice: over 20 Dice points | Improved |
| use 3D context model instead of 2D models for modality transform and super-resolution | Neuroverse3D · generation tasks | PSNR | No clear effect |
| use APSP with different mini-context sizes (ablation) | Neuroverse3D · inference on V100 | memory / output: 7.35GB at l=1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SegGPT — SegGPT: Towards Segmenting Everything in Context (ICCV 2023) · better on 1 of 1 shared result
- UniverSeg — UniverSeg: Universal Medical Image Segmentation (ICCV 2023) · 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
- Enhancing vision-language models for medical imaging: bridging the 3D gap with innovative slice selection (NeurIPS 2024 · similarity 0.87)
- Revisiting MAE Pre-training for 3D Medical Image Segmentation (CVPR 2025 · similarity 0.86)
- Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imaging (NeurIPS 2025 · similarity 0.86)
- Neuro-3D: Towards 3D Visual Decoding from EEG Signals (CVPR 2025 · similarity 0.86)
- Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction (NeurIPS 2024 · similarity 0.86)
- Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex (NeurIPS 2025 · similarity 0.86)
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.
