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

TL;DR

ECHO is a diffusion-based model for chest X-ray report generation that achieves one-step inference, significantly faster speeds, and better performance than previous autoregressive methods.

Contribution

The paper introduces ECHO, a novel diffusion-based approach with a direct conditional distillation framework enabling single-step report generation.

Findings

ECHO surpasses state-of-the-art autoregressive methods in RaTE and SemScore.

ECHO achieves up to 8x inference speedup with negligible accuracy loss.

ECHO demonstrates improved clinical report quality in experiments.

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 \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.