Research map / Vision-language and multimodal / Multimodal LLMs
DiffVP:Differential Visual Semantic Prompting for LLM-Based CT Report Generation
Yuhe Tian, Kun Zhang, Haoran Ma, Rui Yan, Yingtai Li, Rongsheng Wang, S Kevin Zhou
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 diagnostic anchor E (predicted disease prompts) (ablation) | baseline · RadGenome-ChestCT | CE F1: 0.159 -> 0.256 | Improved |
| add global delta operator (ablation) | baseline with E · RadGenome-ChestCT | CE F1: 0.256 -> 0.411 | Improved |
| add local delta operator (ablation) | baseline with E · RadGenome-ChestCT | CE F1: 0.256 -> 0.406 | Improved |
| combine global and local delta (ablation) | DiffVP · RadGenome-ChestCT | BLEU-4: 33.20 / 33.89 -> 34.28 | Improved |
| add local delta on top of global delta, without E (ablation) | DiffVP without E · RadGenome-ChestCT | CE F1: 0.188 -> 0.238 F1; ROUGE-L 42.50 -> 32.83 | Mixed |
| increase visual prefix length beyond 16 (ablation) | DiffVP · RadGenome-ChestCT | BLEU-4: 34.28 (p=16) -> 33.32 (p=24) -> 31.53 (p=32) | Hurt |
| decrease prefix length to 4 (ablation) | DiffVP · RadGenome-ChestCT | BLEU-4: 34.28 -> 29.72 | Hurt |
| compute discrepancy in latent token space instead of pixel space (ablation) | DiffVP · RadGenome-ChestCT | BLEU-1 and CE-F1 | Improved |
| DiffVP vs Reg2RG on RadGenome-ChestCT | DiffVP · RadGenome-ChestCT | BLEU-4: 24.87 -> 34.28 | Improved |
| DiffVP vs Reg2RG on ROUGE-L | DiffVP · RadGenome-ChestCT | ROUGE-L: 36.65 -> 32.32 | Hurt |
| DiffVP vs Reg2RG on clinical precision | DiffVP · RadGenome-ChestCT | CE Precision: 0.423 -> 0.366 | Hurt |
| joint end-to-end training of classifier and generator (ablation) | DiffVP · RadGenome-ChestCT | report quality | Hurt |
Closest papers by meaning
- BenchX: A Unified Benchmark Framework for Medical Vision-Language Pretraining on Chest X-Rays (NeurIPS 2024 · similarity 0.87)
- Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imaging (NeurIPS 2025 · similarity 0.86)
- S2D-Align: Shallow-to-Deep Auxiliary Learning for Anatomically-Grounded Radiology Report Generation (AAAI 2026 · similarity 0.86)
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.86)
- LLM-CXR: Instruction-Finetuned LLM for CXR Image Understanding and Generation (ICLR 2024 · similarity 0.86)
- HC-LLM: Historical-Constrained Large Language Models for Radiology Report Generation (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.
