Research map / Vision-language and multimodal / Multimodal LLMs
RefleXNet: Targeted Self-Reflection for Accurate Chest X-ray Reporting
Xin Mei, Rui Mao, Xiaoyan Cai, Libin Yang, Erik Cambria
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes RefleXNet, a chest X-ray report generation model with targeted self-reflection to better capture subtle abnormalities, as stated in the truncated abstract.
Problem
Automated chest X-ray report generation relies on global visual features and token-level supervision, which limits sensitivity to subtle abnormalities and clinical reliability.
Approach
Introduces targeted self-reflection for CXR reporting as a way to move beyond global visual features and token-level supervision; specific mechanism not stated in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add targeted self-reflection optimization after SFT (ablation) | RefleXNet 3B · MIMIC-CXR | F1-RadGraph: 0.291 -> 0.311 | Improved |
| add targeted self-reflection optimization (ablation) | RefleXNet 3B · MIMIC-CXR | CheXbert micro-F1 (14 obs.): 0.505 -> 0.570 | Improved |
| add targeted self-reflection optimization (ablation) | RefleXNet 3B · MIMIC-CXR | RadCliQ-v0: 3.15 -> 2.96 | Improved |
| add targeted self-reflection optimization (ablation) | RefleXNet 3B · MIMIC-CXR pneumothorax | F1: 0.1696 -> 0.3385 | Improved |
| add targeted self-reflection optimization (ablation) | RefleXNet 3B · MIMIC-CXR cardiomegaly | F1: 0.6433 -> 0.6772 | Improved |
| replace BaML loss with standard BCE (ablation) | MGVE · MIMIC-CXR abnormality prediction | macro-F1: 0.388 -> 0.264 | Hurt |
| remove patch-level feature fusion (region-only) (ablation) | MGVE · MIMIC-CXR abnormality prediction | micro-F1: 0.533 -> 0.476 | Hurt |
| add region-abnormality mapping matrix (ablation) | MGVE · MIMIC-CXR abnormality prediction | macro-F1: 0.357 -> 0.388 | Improved |
| remove mapping matrix (ablation) | MGVE · MIMIC-CXR abnormality prediction | micro-precision: 0.463 -> 0.501 | Mixed |
| add two extra epochs of self-reflection on IU-Xray | RefleXNet · IU-Xray | F1-RadGraph: 0.325 -> 0.384 | Improved |
| add self-reflection to the zero-shot model (ablation) | RefleXNet on IU-Xray · IU-Xray zero-shot | CheXbert vector similarity: 0.590 -> 0.585 | No clear effect |
| continue training Qwen2.5-VL on MIMIC-CXR | Qwen2.5-VL 3B · MIMIC-CXR | F1-RadGraph: 0.094 -> 0.263 | Improved |
Closest papers by meaning
- CXReasonBench: A Benchmark for Evaluating Structured Diagnostic Reasoning in Chest X-rays (NeurIPS 2025 · similarity 0.88)
- Rethinking Visual Attribution for Chest X-ray Reasoning in Large Vision Language Models (NeurIPS 2026 · similarity 0.87)
- AOR: Anatomical Ontology-Guided Reasoning for Medical Large Multimodal Model in Chest X-Ray Interpretation (NeurIPS 2025 · similarity 0.86)
- Enhanced Contrastive Learning with Multi-view Longitudinal Data for Chest X-ray Report Generation (CVPR 2025 · similarity 0.86)
- BenchX: A Unified Benchmark Framework for Medical Vision-Language Pretraining on Chest X-Rays (NeurIPS 2024 · similarity 0.86)
- Reporting Practice Matters: The Impact of Reference Choice on Chest X-ray Report Evaluation (NeurIPS 2026 · 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.
