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

Lifeng Chen; Tianqi You; Hao Liu; Zhimin Bao; Jile Jiao; Xiao Han; Zhicai Ou; Tao Sun

cs.LGApr 10, 2026

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

Authors: Lifeng Chen, Tianqi You, Hao Liu, Zhimin Bao, Jile Jiao, Xiao Han, Zhicai Ou, Tao Sun,

Organizations: AIRC, Midea Group · Beijing Jiaotong University · Dalian University of Technology

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 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 64.33% and 60.58% respectively, while achieving up to \textbf{ $8\times$ } inference speedup with negligible degradation in clinical accuracy.

Figures & tables

Figure 1: Motivation and performance of ECHO. (a) Decoding all tokens simultaneously in one step produces incoherent outputs, as standard diffusion models predict each position independently. Our Direct Conditional Distillation (DCD) distills from a non-factorized target, yielding coherent one-step-per-block outputs. (b) Compared to both autoregressive and diffusion-based baselines, ECHO achieves a favorable trade-off between generation quality (SemScore) and decoding throughput (tokens per forward pass). Here, SDAR* denotes SDAR [ 9 ] implemented under the multimodal large language model setting.

Figure 2: Overview of the ECHO training pipeline. ECHO is built in three successive stages: continued pre-training (Stage 1) produces ECHO ${}_{\text{AR}}$ ; Response-Asymmetric Diffusion adaptation (Stage 2) converts it into the block diffusion model ECHO ${}_{\text{Base}}$ ; and Direct Conditional Distillation (Stage 3) distills ECHO ${}_{\text{Base}}$ into the final one-step-per-block model ECHO . (a) RAD duplicates only the response portion of the training sequence, avoiding the redundant duplication of long vision token contexts required by prior two-stage conversion methods. (b) DCD proceeds in two phases: the teacher’s confidence-heuristic remasking trajectory is collected to form a joint, non-factorized supervision target, which is then used to align the student’s single-step prediction via KL divergence.

Figure 3: (a) Confidence analysis of <eos> tokens vs. content tokens under block sizes blk4 and blk8 in standard multi-step inference. (b) RAD attention FLOPs saved (%) and training speedup ( $\times$ ) as a function of vision token count.

Figure 4: (a) Vanilla block KV cache: after all tokens of a block are committed, a dedicated forward pass is performed to update the KV cache. (b) Fused block KV cache: the KV cache update for the preceding block is fused into the current block’s denoising forward, eliminating the dedicated KV update pass.

Methods	CheXpert-Plus				ReXGradient				MIMIC-CXR				Speed
Methods	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore	TPF	TPS
Proprietary general models
Gemini3-Pro [ 16 ]	26.95	1.53	40.86	29.72	32.10	1.81	33.52	39.02	27.08	1.55	41.42	30.45	$\times$ 1.0	–
Qwen3-Max [ 3 ]	27.19	1.53	41.63	29.24	30.50	1.72	39.12	38.13	26.86	1.51	40.78	27.32	$\times$ 1.0	–
Autoregressive Medical Models
LLaVA-Med [ 25 ]	6.92	0.65	10.11	11.39	7.36	0.72	10.01	17.04	6.69	0.65	10.07	9.22	$\times$ 1.0	36.33

Table 1: Performance comparison with state-of-the-art models. We report average metrics across the MIMIC-CXR, CheXpert-Plus, and ReXGradient test sets. Bold values indicate the best performance. Colored superscript percentages indicate relative quality degradation with respect to ECHO

{}_{\text{AR}}

. Please refer to Appendix. 10 for detailed results on individual datasets.

DCD settings			CheXpert-Plus				ReXGradient				MIMIC-CXR				PPL
SW	CE	RKL	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore	PPL
			52.57	3.61	54.87	46.93	63.28	5.17	61.32	58.36	51.73	3.63	51.19	43.50	23.72
$\checkmark$			52.44	3.53	56.30	46.66	62.90	5.07	61.33	59.90	51.76	3.65	53.14	43.70	21.07
$\checkmark$	$\checkmark$		56.14	4.20	57.40	49.57	66.39	5.54	62.18	66.28	54.12	4.05	53.95	45.57	18.83
$\checkmark$		$\checkmark$	52.51	3.41	55.20	45.32	61.71	4.94	61.26	59.55	50.91	3.48	53.31	42.43	20.23
$\checkmark$	$\checkmark$	$\checkmark$	53.25	3.72	57.25	47.07	63.82	5.21	61.47	63.46	51.48	3.67	53.42	43.86	21.32

Table 2: Ablation on DCD components. SW: step-wise token reweighting; CE: cross-entropy loss on the <eos> token; RKL: reverse KL divergence. Results are reported on CheXpert-Plus, ReXGradient, and MIMIC-CXR.

Figure 5: Effect of training data scale during RAD adaptation. (a) Quality Metric Convergence During RAD. Each colored curve tracks one of four clinical and linguistic metrics across training steps, and dashed horizontal lines mark the corresponding ECHO

{}_{\text{AR}}

baseline values. (b) Decoupled Convergence of Quality and Throughput. The black curve shows the mean quality score and the red curve shows inference throughput (TPF), and dashed horizontal lines mark respective final values.

Block size	Model type	CheXpert-Plus				ReXGradient				MIMIC-CXR
Block size	Model type	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore
Block 4	ECHO ${}_{\text{Base}}$	41.39	2.91	56.39	39.17	53.60	4.32	61.72	57.42	41.50	2.76	53.86	35.41
Block 4	ECHO	56.14 36%↑	4.20 44%↑	57.40 2%↑	49.57 27%↑	66.39 24%↑	5.54 28%↑	62.18 1%↑	66.28 15%↑	54.12 30%↑	4.05 47%↑	53.95 0%↑	45.57 29%↑
Block 8	ECHO ${}_{\text{Base}}$	41.27	2.81	52.12	38.74	53.23	4.19	59.92	54.29	41.08	2.63	51.34	34.18
Block 8	ECHO	55.21 34%↑	4.14 47%↑	56.85 9%↑	51.40 33%↑	65.13 22%↑	5.48 31%↑	60.71 1%↑	64.00 18%↑	53.07 29%↑	4.00 52%↑	52.97 3%↑	44.83 31%↑

Table 3: Comparison of ECHO and ECHO

{}_{\text{Base}}

under native one-step decoding. Green percentages indicate the relative quality improvement of ECHO over ECHO

{}_{\text{Base}}

across three benchmarks. Bold values mark the higher score between the two model types within each block size and metric.

Stage	Data Type	CheXpert-Plus				ReXGradient				MIMIC-CXR
Stage	Data Type	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore	ROUGE-L	CIDEr	RateScore	SemScore
Stage I	Normalized	56.89	4.47	59.18	52.94	67.07	5.68	64.39	66.60	54.55	4.32	56.58	46.70
Stage I	Unnormalized	23.82	1.53	45.59	32.44	25.67	1.49	26.05	40.77	25.06	1.56	42.72	33.54
Stage II	Normalized	56.90	4.25	59.12	52.64	66.46	5.64	63.02	66.68	54.45	4.16	55.85	46.10
Stage II	Unnormalized	23.47	1.51	39.60	30.46	24.97	1.45	23.90	36.97	24.54	1.53	37.76	32.64
Stage III	Normalized	56.14	4.20	57.40	49.57	66.39	5.54	62.18	66.28	54.12	4.05	53.95	45.57

Table 4: Ablation on report data types across all training stages. Each stage contains two settings: normalized and unnormalized reports.

Proprietary general models
Methods	Chinese					English
Methods	ROUGE-L	METEOR	CIDEr	RateScore	SemScore	ROUGE-L	METEOR	CIDEr	RateScore	SemScore
Gemini3-Pro [ 16 ]	26.79	29.40	2.14	41.87	32.21	27.12	25.71	0.92	39.84	27.23
Qwen3-Max [ 3 ]	26.33	30.17	2.12	43.20	29.88	28.04	26.82	0.93	40.06	28.60
Autoregressive Medical Models
LLaVA-Med [ 25 ]	10.58	5.50	1.13	10.03	11.69	3.25	0.56	0.18	10.00	11.09

Table A.1 : Detailed per-metric comparison on CheXpert-Plus across Chinese and English evaluation. Green percentages indicate the relative performance drop of ECHO compared to ECHO

{}_{\text{AR}}

. Bold values denote the best result among all distillation methods.

Proprietary general models
Methods	Chinese					English
Methods	ROUGE-L	METEOR	CIDEr	RateScore	SemScore	ROUGE-L	METEOR	CIDEr	RateScore	SemScore
Gemini3-Pro [ 16 ]	34.10	36.70	2.60	41.78	42.59	30.10	26.34	1.03	25.26	35.46
Qwen3-Max [ 3 ]	31.32	36.62	2.48	45.86	39.13	29.67	29.84	0.97	32.38	37.14
Autoregressive Medical Models
LLaVA-Med [ 25 ]	11.35	5.92	1.23	10.01	18.46	3.37	0.62	0.21	10.00	15.62

Table A.2 : Detailed per-metric comparison on RexGradient across Chinese and English evaluation. Green percentages indicate the relative performance drop of ECHO compared to ECHO

{}_{\text{AR}}

. Bold values denote the best result among all distillation methods.

Proprietary general models
Methods	Chinese					English
Methods	ROUGE-L	METEOR	CIDEr	RateScore	SemScore	ROUGE-L	METEOR	CIDEr	RateScore	SemScore
Gemini3-Pro [ 16 ]	25.86	28.37	2.13	41.46	30.62	28.30	26.28	0.98	41.38	30.28
Qwen3-Max [ 3 ]	25.62	29.10	2.08	41.93	26.95	28.10	27.21	0.94	39.64	27.69
Autoregressive Medical Models
LLaVA-Med [ 25 ]	10.32	5.42	1.12	10.02	10.42	3.05	0.55	0.18	10.00	8.02

Table A.3 : Detailed per-metric comparison on MIMIC across Chinese and English evaluation. Green percentages indicate the relative performance drop of ECHO compared to ECHO

{}_{\text{AR}}

. Bold values denote the best result among all distillation methods.

Figure A.1 : Prompt template for report standardization and normalization. The prompt instructs the model to rewrite a raw radiology report by: (1) separating the content into Findings and Impression sections with standardized clinical terminology, (2) explicitly enumerating all negative findings for clinical completeness, and (3) returning the result in a structured JSON format.

Figure A.2 : Prompt template for translating Chinese radiology reports into English, used to prepare evaluation data for the RaTEScore and SemScore metrics.

Figure A.3 : Qualitative comparison of report generation quality across model variants. ECHO-Base_onestep denotes ECHO ${}_{\text{Base}}$ forced to decode in a single step without distillation.

Figure A.4 : Qualitative examples of positive pathology detection by ECHO . Correctly identified abnormal findings are highlighted in green.

Radiology Report Generation

Radiology report generation uses medical images and clinical context to produce structured, clinically accurate findings and impressions.

See all 159 papers

Discrete Diffusion Language Models for Interactive Radiology Report Drafting
Jul 1, 2026Max Van Puyvelde, Halil Ibrahim Gulluk, Wim Van Criekinge +1Radiology Report Generation Diffusion Language Models
Department of Biomedical Data Science, Stanford University School of Medicine · Department of Mathematical Modelling, Statistics & Bioinformatics, Ghent University · Department of Electrical Engineering, Stanford University
DRRG: A Discrete Diffusion Framework for Radiology Report Generation
Aug 25, 2026Shaoyang Zhoua, Yingshu Li, Yunyi Liu +4Radiology Report Generation Radiology VLMs
School of Electrical and Computer Engineering, The University of Sydney, Sydney, NSW, 2006, Australia · Department of Radiology, The First Affiliated Hospital of Soochow University, Suzhou, Jiangsu, 215006, China · School of Computer Science, The University of Adelaide, Adelaide, SA, 5005, Australia +1
CheXmix: Unified Generative Pretraining for Vision Language Models in Medical Imaging
Apr 24, 2026Ashwin Kumar, Robbie Holland, Corey Barrett +8Radiology Report Generation Chest X-Ray Classification
Stanford AIMI, Stanford University · Oracle Health AI · Department of Radiology, Stanford University

Jul 1, 2026J/K move · Enter open · S save

Discrete Diffusion Language Models for Interactive Radiology Report Drafting

Max Van Puyvelde, Halil Ibrahim Gulluk, Wim Van Criekinge, Olivier Gevaert

Department of Biomedical Data Science, Stanford University School of Medicine · Department of Mathematical Modelling, Statistics & Bioinformatics, Ghent University · Department of Electrical Engineering, Stanford University

Radiology Report Generation Diffusion Language Models Text Generation

Open paper AskPDF arXiv

Diffusion language models, which generate text by denoising a token canvas bidirectionally instead of emitting tokens left to right, have become competitive with autoregressive (AR) generation. Medical foundation models, however, remain almost entirely autoregressive. We adapt a mixture-of-experts diffusion language model, DiffusionGemma-26B, and benchmark it against its same-size AR sibling Gemma-4-26B under an identical LoRA recipe on medical visual question answering datasets, scored by a verbosity-robust LLM judge. Diffusion matches or exceeds AR on all of them, and the finetuned model (3.8B active) is competitive with frontier vision-language models; its decoding is also 3.5-4.4x faster. Beyond this parity, the diffusion model offers a drafting capability AR lacks: any-order infill. Because the canvas is denoised bidirectionally, a radiologist can fix report fragments and have the model fill the text between them, an operation inherent to diffusion but not to autoregression, which is subpar at it. This suits real reports, which are often terse or inconsistent across clinicians and institutions.

Abstract

Radiology Report Generation

Explore similar work

Discrete Diffusion Language Models for Interactive Radiology Report Drafting

DRRG: A Discrete Diffusion Framework for Radiology Report Generation

CheXmix: Unified Generative Pretraining for Vision Language Models in Medical Imaging