Research map / Architectures and efficiency / Attention and transformers
Mimicking Radiologists: A Coarse-to-Fine Framework with Structural Sparse Tokens for Dual-LLM Computed Tomography Report Generation
Hong Liu, Dong Wei, Yefeng Zheng, Xian Wu, Liansheng Wang
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 mask-guided sparse negative-entropy loss (ablation) | baseline CTRG · CT-RATE (test) | CE F1: 0.356 -> 0.377 | Improved |
| add mask-guided sparse negative-entropy loss (ablation) | baseline CTRG · CT-RATE (test) | BLEU-1: 0.379 -> 0.375 | Hurt |
| add abnormality-prefixed SFT (ablation) | baseline + L_sparse · CT-RATE (test) | CE F1: 0.377 -> 0.394 | Improved |
| add abnormality-prefixed SFT (ablation) | baseline + L_sparse · CT-RATE (test) | BLEU-4: 0.192 -> 0.282 | Improved |
| add AP-LTF (K=200 filter) without GRPO (ablation) | baseline + L_sparse + AP-SFT · CT-RATE (test) | CE F1: 0.394 -> 0.391 | Hurt |
| add AP-LTF (K=200 filter) without GRPO (ablation) | baseline + L_sparse + AP-SFT · CT-RATE (test) | CE recall: 0.397 -> 0.418 | Improved |
| add GRPO collaborative refinement (ablation) | baseline + L_sparse + AP-SFT + AP-LTF · CT-RATE (test) | CE F1: 0.391 -> 0.415 | Improved |
| add GRPO collaborative refinement (ablation) | baseline + L_sparse + AP-SFT + AP-LTF · CT-RATE (test) | BLEU-4: 0.289 -> 0.305 | Improved |
| use two specialized LLMs instead of single multitask LLM (ablation) | Ours · CT-RATE (test) | CE F1: 0.381 -> 0.415 | Improved |
| use two specialized LLMs instead of single multitask LLM (ablation) | Ours · CT-RATE (test) | BLEU-4: 0.302 -> 0.305 | Hurt |
| use composite F1+BLEU-4 reward instead of BLEU-4 alone (ablation) | GRPO refinement · CT-RATE (test) | CE F1: 0.401 -> 0.415 | Improved |
| use composite F1+BLEU-4 reward instead of F1 alone (ablation) | GRPO refinement · CT-RATE (test) | CE F1: 0.408 -> 0.415 | Improved |
Closest papers by meaning
- S2D-Align: Shallow-to-Deep Auxiliary Learning for Anatomically-Grounded Radiology Report Generation (AAAI 2026 · similarity 0.88)
- Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imaging (NeurIPS 2025 · similarity 0.87)
- LLM-RG4: Flexible and Factual Radiology Report Generation Across Diverse Input Contexts (AAAI 2025 · similarity 0.86)
- DART: Disease-aware Image-Text Alignment and Self-correcting Re-alignment for Trustworthy Radiology Report Generation (CVPR 2025 · similarity 0.86)
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.86)
- DAMPER: A Dual-Stage Medical Report Generation Framework with Coarse-Grained MeSH Alignment and Fine-Grained Hypergraph Matching (AAAI 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.
