ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion

arXiv:2604.09450v1 Announce Type: cross Abstract: 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 extbf{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 extbf{64.33%} and extbf{60.58%} respectively, while achieving an extbf{8imes} inference speedup without compromising clinical accuracy.

Related

• Temporal Inversion for Learning Interval Change in Chest X-Rays
• Enhancing the Safety of Medical Vision-Language Models by Synthetic Demonstrations
• Better Eyes, Better Thoughts: Why Vision Chain-of-Thought Fails in Medicine
• VISOR: Agentic Visual Retrieval-Augmented Generation via Iterative Search and Over-horizon Reasoning

Source: arXiv cs.AI | 2026-04-13