Research map / Vision-language and multimodal / Visual question answering
Uncertainty-Aware Medical Diagnostic Phrase Identification and Grounding
Ke Zou, Yang Bai, Bo Liu, Yidi Chen, Zhihao Chen, Yang Zhou, Xuedong Yuan, Meng Wang, Xiaojing Shen, Xiaochun Cao, Yih Chung Tham, Huazhu Fu
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 uncertainty-aware prediction model (multi-hypothesis head) (ablation) | MedGround · MRG-MS-CXR | mIOU: 46.04 -> 47.65 | Improved |
| add uncertainty-aware prediction model (ablation) | MedGround · MRG-ChestX-ray8 | AP50: 31.82 -> 38.38 | Improved |
| add SAM decoder fine-tuning with frozen encoder to LLaVA box regression (ablation) | LLaVA+LoRA · MRG-MS-CXR | AP30: 29.34 -> 71.26 | Improved |
| fine-tune SAM encoder with LoRA instead of keeping it frozen (ablation) | uMedGround · MRG-MS-CXR | mIOU: 47.65 -> 45.31 | Hurt |
| fully fine-tune SAM encoder and decoder instead of decoder only (ablation) | uMedGround · MRG-MS-CXR | mIOU: 47.65 -> 41.53 | Hurt |
| use LoRA-fine-tuned LLaMA2 phrases instead of zero-shot GPT-4 phrases | TransVG · MRG-MS-CXR | mIOU: 26.28 -> 42.60 | Improved |
| replace two-stage phrase extraction with end-to-end uMedGround | TransVG with LLaMA2-FT phrases · MRG-MS-CXR | mIOU: 42.60 -> 47.65 | Improved |
| add uncertainty-aware prediction under Gaussian noise 0.1 (ablation) | MedGround · MRG-ChestX-ray8 (noisy) | mIOU: 32.52 -> 35.16 | Improved |
| add Gaussian noise to input images | uMedGround · MRG-MS-CXR | mIOU: 47.65 -> 47.01 | Hurt |
| cross-dataset transfer: add uncertainty-aware prediction (ablation) | MedGround · MRG-ChestX-ray8 (trained on MRG-MS-CXR) | AP50: 21.72 -> 23.74 | Mixed |
| use VQA pairs instead of class descriptions as text input | uMedGround · MRG-MIMIC | mIOU: 46.61 -> 47.39 | Improved |
| use original report instead of VQA pairs as text input | uMedGround · MRG-MIMIC | mIOU: 47.39 -> 47.65 | Improved |
Closest papers by meaning
- MedSG-Bench: A Benchmark for Medical Image Sequences Grounding (NeurIPS 2025 · similarity 0.89)
- How Do Medical MLLMs Fail? A Study on Visual Grounding in Medical Images (ICLR 2026 · similarity 0.88)
- MedReasoner: Reinforcement Learning Drives Reasoning Grounding from Clinical Thought to Pixel-Level Precision (AAAI 2026 · similarity 0.87)
- M3Grounder: Mask-Based Multi-Span and Multi-Granular Grounding for Document QA (CVPR 2026 · similarity 0.85)
- Falcon Medical Visual Question Answering (AAAI 2025 · similarity 0.85)
- MIMO: A Medical Vision Language Model with Visual Referring Multimodal Input and Pixel Grounding Multimodal Output (CVPR 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.
