ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866917504585039872 |
|---|---|
| author | Chen, Lifeng You, Tianqi Liu, Hao Bao, Zhimin Jiao, Jile Han, Xiao Ou, Zhicai Sun, Tao Mou, Xiaofeng Jin, Xiaojie Xu, Yi |
| author_facet | Chen, Lifeng You, Tianqi Liu, Hao Bao, Zhimin Jiao, Jile Han, Xiao Ou, Zhicai Sun, Tao Mou, Xiaofeng Jin, Xiaojie Xu, Yi |
| contents | Chest X-ray report generation (CXR-RG) has the potential to substantially alleviate radiologists' workload. However, conventional autoregressive vision--language models (VLMs) suffer from high inference latency due to sequential token decoding. Diffusion-based models offer a promising alternative through parallel generation, but they still require multiple denoising iterations. Compressing multi-step denoising to a single step could further reduce latency, but often degrades textual coherence due to the mean-field bias introduced by token-factorized denoisers. To address this challenge, we propose \textbf{ECHO}, an efficient diffusion-based VLM (dVLM) for chest X-ray report generation. ECHO enables stable one-step-per-block inference via a novel Direct Conditional Distillation (DCD) framework, which mitigates the mean-field limitation by constructing unfactorized supervision from on-policy diffusion trajectories to encode joint token dependencies. In addition, we introduce a Response-Asymmetric Diffusion (RAD) training strategy that further improves training efficiency while maintaining model effectiveness. Extensive experiments demonstrate that ECHO surpasses state-of-the-art autoregressive methods, improving RaTE and SemScore by \textbf{64.33\%} and \textbf{60.58\%} respectively, while achieving up to \textbf{$8\times$} inference speedup with negligible degradation in clinical accuracy. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_09450 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion Chen, Lifeng You, Tianqi Liu, Hao Bao, Zhimin Jiao, Jile Han, Xiao Ou, Zhicai Sun, Tao Mou, Xiaofeng Jin, Xiaojie Xu, Yi Machine Learning Artificial Intelligence Image and Video Processing Chest X-ray report generation (CXR-RG) has the potential to substantially alleviate radiologists' workload. However, conventional autoregressive vision--language models (VLMs) suffer from high inference latency due to sequential token decoding. Diffusion-based models offer a promising alternative through parallel generation, but they still require multiple denoising iterations. Compressing multi-step denoising to a single step could further reduce latency, but often degrades textual coherence due to the mean-field bias introduced by token-factorized denoisers. To address this challenge, we propose \textbf{ECHO}, an efficient diffusion-based VLM (dVLM) for chest X-ray report generation. ECHO enables stable one-step-per-block inference via a novel Direct Conditional Distillation (DCD) framework, which mitigates the mean-field limitation by constructing unfactorized supervision from on-policy diffusion trajectories to encode joint token dependencies. In addition, we introduce a Response-Asymmetric Diffusion (RAD) training strategy that further improves training efficiency while maintaining model effectiveness. Extensive experiments demonstrate that ECHO surpasses state-of-the-art autoregressive methods, improving RaTE and SemScore by \textbf{64.33\%} and \textbf{60.58\%} respectively, while achieving up to \textbf{$8\times$} inference speedup with negligible degradation in clinical accuracy. |
| title | ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion |
| topic | Machine Learning Artificial Intelligence Image and Video Processing |
| url | https://arxiv.org/abs/2604.09450 |