Precision Recall Controllable Radiology Report Generation via Hybrid Natural Language and Clinical Reward Learning

Dufan Wu; Hanliang Chen; Jun Luo; Ling Chen; Luciano Prevedello; Quirin Strotzer; Rongkai Yan; Ruinan Jin; Yuan Xue

arxiv: 2606.21447 · v3 · pith:XJWCX4X5 · submitted 2026-06-19 · cs.CL · cs.CV

Precision Recall Controllable Radiology Report Generation via Hybrid Natural Language and Clinical Reward Learning

Ling Chen , Ruinan Jin , Jun Luo , Hanliang Chen , Quirin Strotzer , Rongkai Yan , Yuan Xue , Luciano Prevedello

show 1 more author

Dufan Wu

This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-07-03 23:15 UTCgrok-4.3pith:XJWCX4X5 record.json open to challenge →

classification cs.CL cs.CV

keywords radiology report generationreinforcement learningprecision recall controlclinical rewardMIMIC-CXRnatural language generationclinical efficacy

0 comments

The pith

A reinforcement learning framework generates radiology reports with a tunable control parameter that adjusts the clinical precision-recall trade-off while adding a clinical reward to improve efficacy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a reinforcement learning method for automated radiology report generation that incorporates a control parameter to explicitly balance clinical precision and recall at inference time. It combines this with a clinical reward term in the training objective and a group-relative normalization strategy to stabilize learning. Experiments on the MIMIC-CXR dataset show gains over prior methods in both language fluency metrics and clinical efficacy scores. A sympathetic reader would care because current systems often produce fluent but clinically inflexible reports that cannot be adjusted to different diagnostic priorities. The design aims to make generated reports more usable across varying clinical contexts without separate models for each priority.

Core claim

The authors claim that a hybrid reinforcement learning objective combining natural language generation rewards with a clinical reward, modulated by an explicit control parameter, enables reliable adjustment of the precision-recall trade-off in generated radiology reports. This produces outputs that exceed state-of-the-art performance on both NLG and clinical efficacy metrics on the MIMIC-CXR dataset while maintaining training stability through group-relative reward normalization.

What carries the argument

The control parameter that scales the clinical precision-recall trade-off inside the reinforcement learning policy, paired with the clinical reward function.

If this is right

Reports can be generated on demand with higher emphasis on precision or on recall depending on the clinical scenario.
Clinical efficacy scores rise in tandem with natural language generation metrics rather than trading one for the other.
Training stability improves because rewards are normalized within each batch group.
A single trained model can serve multiple clinical priorities instead of requiring separate models for each precision-recall target.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Radiologists could adjust the control parameter in real time to generate an initial high-recall screening report and then a high-precision follow-up version for the same images.
The approach might extend to other medical text tasks such as discharge summary generation where similar precision-recall tensions exist.
Integration with hospital workflows could allow the parameter to be set automatically based on patient risk profiles stored in the electronic record.
If the clinical reward can be further decomposed by disease category, the same framework might offer targeted control over sensitivity for specific conditions.

Load-bearing premise

The clinical reward function, when scaled by the control parameter, produces reports whose clinical efficacy improves or holds steady as precision and recall are traded off without creating new clinical errors.

What would settle it

On the MIMIC-CXR test set, sweeping the control parameter across its range produces no statistically detectable change in measured clinical precision-recall balance while clinical efficacy scores remain at or above baseline levels.

Figures

Figures reproduced from arXiv: 2606.21447 by Dufan Wu, Hanliang Chen, Jun Luo, Ling Chen, Luciano Prevedello, Quirin Strotzer, Rongkai Yan, Ruinan Jin, Yuan Xue.

**Figure 1.** Figure 1: Overall framework of the proposed precision recall controllable RRG model. (a) Overall architecture. (b) Reinforcement learning process. (c) Transformer decoder. The overall framework and decoder are shown in gray dashed boxes with details omitted. Red dotted lines denote gradient back-propagation during training. Blue dotted lines indicate the precision recall control parameter λ applied at both the repre… view at source ↗

**Figure 2.** Figure 2: Effect of the precision recall control parameter λ on clinical precision, recall, and F1-score. Performance Comparison We compared our method with several stateof-the-art methods on the MIMIC-CXR dataset, including R2Gen [2], METrans [22], R2GenGPT [23], BoostRRG [24], Diff-RRG [25], MLRG [26], and MedGemma 1.5 4B [27]. Results of the methods except MedGemma were reported as provided in their original p… view at source ↗

**Figure 3.** Figure 3: Qualitative comparison of radiology reports generated by the baseline method and our method with ground truth references. Main improvements are highlighted in blue. diac silhouette whereas the baseline model described the size as "borderline". These examples demonstrate that the reports generated by our framework are more consistent with the ground truth and better reflect clinically meaningful findings. 5… view at source ↗

read the original abstract

Automated radiology report generation (RRG) has gained increasing attention because it can reduce the heavy workload of clinical report writing. However, most existing methods mainly optimize for natural language generation (NLG) metrics that focus on language fluency, while providing little control over clinically important factors such as precision and recall. As consequence, generated reports may be fluent but not well aligned with different clinical needs. To address this challenge, we propose a reinforcement learning framework for precision recall controllable RRG, where a control parameter explicitly adjusts the trade-off between clinical precision and recall during inference. This design allows the model to flexibly generate reports according to different clinical requirements. To ensure clinical correctness, we introduce a clinical reward into the training objective, which helps improve clinical efficacy (CE) beyond standard language-based optimization. In addition, we apply a group-relative training strategy that normalizes rewards within each training group, reducing reward variance and improving training stability. Extensive experiments on the MIMIC-CXR dataset show that our method consistently outperforms state-of-the-art approaches in both NLG and CE evaluation metrics, while providing reliable control over the CE precision recall trade-off.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This adds an inference-time control knob for clinical precision-recall trade-offs in radiology reports via RL with a hybrid reward, but the abstract gives almost no implementation or result details to judge whether it works.

read the letter

The main thing to know is that the authors put a single control parameter into an RL setup so that at inference you can shift the balance between clinical precision and recall in generated reports, while training includes a clinical reward term and group-relative normalization to keep training stable.

What is new is the explicit inference-time control combined with that hybrid reward and the normalization step, applied to radiology report generation on MIMIC-CXR. Prior work had controllable generation and reward learning, but this specific mix for trading off clinical metrics is not already described. The paper does a straightforward job naming the real problem that fluent NLG-optimized reports often fail to match varying clinical priorities.

The soft spots are mostly about missing substance. The abstract mentions extensive experiments and consistent outperformance on NLG and CE metrics plus reliable control, yet supplies no equations, reward definitions, ablation tables, or even sample numbers. Without those it is impossible to check whether the control parameter actually produces the claimed trade-off or whether the clinical reward creates new errors. The group-relative strategy is mentioned as reducing variance, but again no evidence is shown here. These gaps are not fatal on their own, but they make the central claims hard to evaluate from the provided text.

This is for researchers already working on medical report generation or RL-based controllable text who need a concrete example of adding clinical knobs. A reader looking for new paradigms will not find one; someone wanting to see how existing RL ideas get adapted to a clinical constraint might pick up a useful setup.

I would send it to peer review. The problem is legitimate and the proposed direction is reasonable, so referees can check the missing details and decide whether the control and reward actually deliver.

Referee Report

0 major / 0 minor

Summary. The manuscript proposes a reinforcement learning framework for radiology report generation that introduces a control parameter to explicitly adjust the clinical precision-recall trade-off at inference time. It augments the training objective with a clinical reward function and applies group-relative training to normalize rewards and stabilize learning. Experiments on the MIMIC-CXR dataset are reported to show consistent outperformance over state-of-the-art methods on both NLG and clinical efficacy (CE) metrics while enabling reliable control over the precision-recall balance.

Significance. If the central claims hold after verification of the reward definition, control mechanism, and statistical results, the work would address a practical gap between fluency-focused NLG optimization and clinically tunable report generation. The hybrid reward and controllable inference design could support deployment scenarios with varying diagnostic priorities.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their summary of our work and for recognizing its potential to address the gap between fluency-focused optimization and clinically tunable report generation. We note that the report lists no specific major comments for us to address point by point.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The abstract and visible description outline a reinforcement learning setup with an explicit control parameter for precision-recall trade-off plus a clinical reward term added to the objective, plus a group-relative normalization step. No equations, definitions, or self-citations are shown that would make any reported prediction or CE metric reduce by construction to a fitted input or prior self-referential result. The central claims rest on external MIMIC-CXR experiments and standard RL training, which remain independent of the method's own definitions. This is the normal self-contained case.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

Abstract-only view limits visibility into exact parameter counts or background assumptions; the control parameter and clinical reward appear as the main additions beyond standard RL.

free parameters (1)

control parameter
Explicit scalar that trades off clinical precision versus recall at inference time.

axioms (1)

domain assumption A scalar clinical reward can be defined that meaningfully improves clinical efficacy when added to the RL objective.
Invoked to justify the hybrid training objective.

invented entities (1)

clinical reward no independent evidence
purpose: To optimize for clinical correctness beyond standard NLG metrics.
Introduced as an additional term in the training objective.

pith-pipeline@v0.9.1-grok · 5755 in / 1235 out tokens · 34798 ms · 2026-07-03T23:15:18.725307+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages · 7 internal anchors

[1]

On the automatic generation of medical imaging reports

Baoyu Jing, Pengtao Xie, and Eric Xing. On the automatic generation of medical imaging reports. InProceedings of the 56th annual meeting of the association for computational linguistics (volume 1: long papers), pages 2577–2586, 2018

work page 2018
[2]

Generating radiology reports via memory-driven transformer.arXiv preprint arXiv:2010.16056, 2020

Zhihong Chen, Yan Song, Tsung-Hui Chang, and Xiang Wan. Generating radiology reports via memory-driven transformer.arXiv preprint arXiv:2010.16056, 2020

work page arXiv 2010
[3]

Evaluating progress in automatic chest x-ray radiology report generation.Patterns, 4(9), 2023

Feiyang Yu, Mark Endo, Rayan Krishnan, Ian Pan, Andy Tsai, Eduardo Pontes Reis, Eduardo Kaiser Ururahy Nunes Fonseca, Henrique Min Ho Lee, Zahra Shak- eri Hossein Abad, Andrew Y Ng, et al. Evaluating progress in automatic chest x-ray radiology report generation.Patterns, 4(9), 2023

work page 2023
[4]

Auto- mated radiology report generation: A review of recent advances.IEEE Reviews in Biomedical Engineering, 18:368–387, 2024

Phillip Sloan, Philip Clatworthy, Edwin Simpson, and Majid Mirmehdi. Auto- mated radiology report generation: A review of recent advances.IEEE Reviews in Biomedical Engineering, 18:368–387, 2024

work page 2024
[5]

Optimizing the factual correctness of a summary: A study of summarizing radiology reports

Yuhao Zhang, Derek Merck, Emily Tsai, Christopher D Manning, and Curtis Lan- glotz. Optimizing the factual correctness of a summary: A study of summarizing radiology reports. InProceedings of the 58th annual meeting of the association for computational linguistics, pages 5108–5120, 2020

work page 2020
[6]

Hybrid retrieval-generation reinforced agent for medical image report generation.Advances in neural informa- tion processing systems, 31, 2018

Yuan Li, Xiaodan Liang, Zhiting Hu, and Eric P Xing. Hybrid retrieval-generation reinforced agent for medical image report generation.Advances in neural informa- tion processing systems, 31, 2018

work page 2018
[7]

Lhr- rfl: Linear hybrid-reward-based reinforced focal learning for automatic radiology report generation.IEEE transactions on medical imaging, 44(3):1494–1504, 2025

Xiulong Yi, You Fu, Jianzhi Yu, Ruiqing Liu, Hao Zhang, and Rong Hua. Lhr- rfl: Linear hybrid-reward-based reinforced focal learning for automatic radiology report generation.IEEE transactions on medical imaging, 44(3):1494–1504, 2025

work page 2025
[8]

Reinforced cross-modal alignment for radiology report generation

Han Qin and Yan Song. Reinforced cross-modal alignment for radiology report generation. InFindings of the Association for Computational Linguistics: ACL 2022, pages 448–458, 2022

work page 2022
[9]

Deep Communicating Agents for Abstractive Summarization

Asli Celikyilmaz, Antoine Bosselut, Xiaodong He, and Yejin Choi. Deep commu- nicating agents for abstractive summarization.arXiv preprint arXiv:1803.10357, 2018. 10 L. Chen et al

work page internal anchor Pith review Pith/arXiv arXiv 2018
[10]

Bleu: a method for automatic evaluation of machine translation

Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. InProceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318, 2002

work page 2002
[11]

Meteor: An automatic metric for mt evalua- tion with improved correlation with human judgments

Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evalua- tion with improved correlation with human judgments. InProceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65–72, 2005

work page 2005
[12]

Rouge: A package for automatic evaluation of summaries

Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText summarization branches out, pages 74–81, 2004

work page 2004
[13]

High-performance medicine: the convergence of human and artificial intelligence.Nature medicine, 25(1):44–56, 2019

Eric J Topol. High-performance medicine: the convergence of human and artificial intelligence.Nature medicine, 25(1):44–56, 2019

work page 2019
[14]

A Deep Reinforced Model for Abstractive Summarization

Romain Paulus, Caiming Xiong, and Richard Socher. A deep reinforced model for abstractive summarization.arXiv preprint arXiv:1705.04304, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[15]

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024
[16]

MIMIC-CXR-JPG, a large publicly available database of labeled chest radiographs

Alistair EW Johnson, Tom J Pollard, Nathaniel R Greenbaum, Matthew P Lun- gren, Chih-ying Deng, Yifan Peng, Zhiyong Lu, Roger G Mark, Seth J Berkowitz, and Steven Horng. Mimic-cxr-jpg, a large publicly available database of labeled chest radiographs.arXiv preprint arXiv:1901.07042, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1901
[17]

Film: Visual reasoning with a general conditioning layer

Ethan Perez, Florian Strub, Harm De Vries, Vincent Dumoulin, and Aaron Courville. Film: Visual reasoning with a general conditioning layer. InProceedings of the AAAI conference on artificial intelligence, volume 32, 2018

work page 2018
[18]

Simple statistical gradient-following algorithms for connec- tionist reinforcement learning.Machine learning, 8(3):229–256, 1992

Ronald J Williams. Simple statistical gradient-following algorithms for connec- tionist reinforcement learning.Machine learning, 8(3):229–256, 1992

work page 1992
[19]

Rrg- dpo: Direct preference optimization for clinically accurate radiology report gener- ation

Hong Liu, Dong Wei, Zhe Xu, Xian Wu, Yefeng Zheng, and Liansheng Wang. Rrg- dpo: Direct preference optimization for clinically accurate radiology report gener- ation. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 552–562. Springer, 2025

work page 2025
[20]

Chexbert: combining automatic labelers and expert an- notations for accurate radiology report labeling using bert.arXiv preprint arXiv:2004.09167, 2020

Akshay Smit, Saahil Jain, Pranav Rajpurkar, Anuj Pareek, Andrew Y Ng, and Matthew P Lungren. Chexbert: combining automatic labelers and expert an- notations for accurate radiology report labeling using bert.arXiv preprint arXiv:2004.09167, 2020

work page arXiv 2004
[21]

Ohio supercomputer center, 1987

Ohio Supercomputer Center. Ohio supercomputer center, 1987

work page 1987
[22]

Metransformer: Radi- ology report generation by transformer with multiple learnable expert tokens

Zhanyu Wang, Lingqiao Liu, Lei Wang, and Luping Zhou. Metransformer: Radi- ology report generation by transformer with multiple learnable expert tokens. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 11558–11567, 2023

work page 2023
[23]

R2gengpt: Radiology report generation with frozen llms.Meta-Radiology, 1(3):100033, 2023

Zhanyu Wang, Lingqiao Liu, Lei Wang, and Luping Zhou. R2gengpt: Radiology report generation with frozen llms.Meta-Radiology, 1(3):100033, 2023

work page 2023
[24]

Contrastive knowledge- guided large language models for medical report generation

Yuyang Sha, Hongxin Pan, Weiyu Meng, and Kefeng Li. Contrastive knowledge- guided large language models for medical report generation. InInternational Con- ference on Medical Image Computing and Computer-Assisted Intervention, pages 111–120. Springer, 2025

work page 2025
[25]

Diff-rrg: Lon- gitudinal disease-wise patch difference as guidance for llm-based radiology re- port generation

Hannah Yun, Junyeong Maeng, Eunsong Kang, and Heung-Il Suk. Diff-rrg: Lon- gitudinal disease-wise patch difference as guidance for llm-based radiology re- port generation. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 152–161. Springer, 2025. Precision-Recall Controllable Radiology Report Generation 11

work page 2025
[26]

Enhanced contrastive learning with multi-view longitudinal data for chest x-ray report generation

Kang Liu, Zhuoqi Ma, Xiaolu Kang, Yunan Li, Kun Xie, Zhicheng Jiao, and Qiguang Miao. Enhanced contrastive learning with multi-view longitudinal data for chest x-ray report generation. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 10348–10359, 2025

work page 2025
[27]

MedGemma Technical Report

Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, Atilla Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, Cían Hughes, Charles Lau, et al. Medgemma technical report.arXiv preprint arXiv:2507.05201, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[28]

GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization

Shih-Yang Liu, Xin Dong, Ximing Lu, Shizhe Diao, Peter Belcak, Mingjie Liu, Min-Hung Chen, Hongxu Yin, Yu-Chiang Frank Wang, Kwang-Ting Cheng, et al. Gdpo:Groupreward-decouplednormalizationpolicyoptimizationformulti-reward rl optimization.arXiv preprint arXiv:2601.05242, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026
[29]

CURE: Curriculum-guided Multi-task Training for Reliable Anatomy Grounded Report Generation

Pablo Messina, Andrés Villa, Juan León Alcázar, Karen Sánchez, Carlos Hinojosa, Denis Parra, Álvaro Soto, and Bernard Ghanem. Cure: Curriculum-guided multi- task training for reliable anatomy grounded report generation.arXiv preprint arXiv:2601.15408, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026

[1] [1]

On the automatic generation of medical imaging reports

Baoyu Jing, Pengtao Xie, and Eric Xing. On the automatic generation of medical imaging reports. InProceedings of the 56th annual meeting of the association for computational linguistics (volume 1: long papers), pages 2577–2586, 2018

work page 2018

[2] [2]

Generating radiology reports via memory-driven transformer.arXiv preprint arXiv:2010.16056, 2020

Zhihong Chen, Yan Song, Tsung-Hui Chang, and Xiang Wan. Generating radiology reports via memory-driven transformer.arXiv preprint arXiv:2010.16056, 2020

work page arXiv 2010

[3] [3]

Evaluating progress in automatic chest x-ray radiology report generation.Patterns, 4(9), 2023

Feiyang Yu, Mark Endo, Rayan Krishnan, Ian Pan, Andy Tsai, Eduardo Pontes Reis, Eduardo Kaiser Ururahy Nunes Fonseca, Henrique Min Ho Lee, Zahra Shak- eri Hossein Abad, Andrew Y Ng, et al. Evaluating progress in automatic chest x-ray radiology report generation.Patterns, 4(9), 2023

work page 2023

[4] [4]

Auto- mated radiology report generation: A review of recent advances.IEEE Reviews in Biomedical Engineering, 18:368–387, 2024

Phillip Sloan, Philip Clatworthy, Edwin Simpson, and Majid Mirmehdi. Auto- mated radiology report generation: A review of recent advances.IEEE Reviews in Biomedical Engineering, 18:368–387, 2024

work page 2024

[5] [5]

Optimizing the factual correctness of a summary: A study of summarizing radiology reports

Yuhao Zhang, Derek Merck, Emily Tsai, Christopher D Manning, and Curtis Lan- glotz. Optimizing the factual correctness of a summary: A study of summarizing radiology reports. InProceedings of the 58th annual meeting of the association for computational linguistics, pages 5108–5120, 2020

work page 2020

[6] [6]

Hybrid retrieval-generation reinforced agent for medical image report generation.Advances in neural informa- tion processing systems, 31, 2018

Yuan Li, Xiaodan Liang, Zhiting Hu, and Eric P Xing. Hybrid retrieval-generation reinforced agent for medical image report generation.Advances in neural informa- tion processing systems, 31, 2018

work page 2018

[7] [7]

Lhr- rfl: Linear hybrid-reward-based reinforced focal learning for automatic radiology report generation.IEEE transactions on medical imaging, 44(3):1494–1504, 2025

Xiulong Yi, You Fu, Jianzhi Yu, Ruiqing Liu, Hao Zhang, and Rong Hua. Lhr- rfl: Linear hybrid-reward-based reinforced focal learning for automatic radiology report generation.IEEE transactions on medical imaging, 44(3):1494–1504, 2025

work page 2025

[8] [8]

Reinforced cross-modal alignment for radiology report generation

Han Qin and Yan Song. Reinforced cross-modal alignment for radiology report generation. InFindings of the Association for Computational Linguistics: ACL 2022, pages 448–458, 2022

work page 2022

[9] [9]

Deep Communicating Agents for Abstractive Summarization

Asli Celikyilmaz, Antoine Bosselut, Xiaodong He, and Yejin Choi. Deep commu- nicating agents for abstractive summarization.arXiv preprint arXiv:1803.10357, 2018. 10 L. Chen et al

work page internal anchor Pith review Pith/arXiv arXiv 2018

[10] [10]

Bleu: a method for automatic evaluation of machine translation

Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. InProceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318, 2002

work page 2002

[11] [11]

Meteor: An automatic metric for mt evalua- tion with improved correlation with human judgments

Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evalua- tion with improved correlation with human judgments. InProceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65–72, 2005

work page 2005

[12] [12]

Rouge: A package for automatic evaluation of summaries

Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText summarization branches out, pages 74–81, 2004

work page 2004

[13] [13]

High-performance medicine: the convergence of human and artificial intelligence.Nature medicine, 25(1):44–56, 2019

Eric J Topol. High-performance medicine: the convergence of human and artificial intelligence.Nature medicine, 25(1):44–56, 2019

work page 2019

[14] [14]

A Deep Reinforced Model for Abstractive Summarization

Romain Paulus, Caiming Xiong, and Richard Socher. A deep reinforced model for abstractive summarization.arXiv preprint arXiv:1705.04304, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017

[15] [15]

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024

[16] [16]

MIMIC-CXR-JPG, a large publicly available database of labeled chest radiographs

Alistair EW Johnson, Tom J Pollard, Nathaniel R Greenbaum, Matthew P Lun- gren, Chih-ying Deng, Yifan Peng, Zhiyong Lu, Roger G Mark, Seth J Berkowitz, and Steven Horng. Mimic-cxr-jpg, a large publicly available database of labeled chest radiographs.arXiv preprint arXiv:1901.07042, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1901

[17] [17]

Film: Visual reasoning with a general conditioning layer

Ethan Perez, Florian Strub, Harm De Vries, Vincent Dumoulin, and Aaron Courville. Film: Visual reasoning with a general conditioning layer. InProceedings of the AAAI conference on artificial intelligence, volume 32, 2018

work page 2018

[18] [18]

Simple statistical gradient-following algorithms for connec- tionist reinforcement learning.Machine learning, 8(3):229–256, 1992

Ronald J Williams. Simple statistical gradient-following algorithms for connec- tionist reinforcement learning.Machine learning, 8(3):229–256, 1992

work page 1992

[19] [19]

Rrg- dpo: Direct preference optimization for clinically accurate radiology report gener- ation

Hong Liu, Dong Wei, Zhe Xu, Xian Wu, Yefeng Zheng, and Liansheng Wang. Rrg- dpo: Direct preference optimization for clinically accurate radiology report gener- ation. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 552–562. Springer, 2025

work page 2025

[20] [20]

Chexbert: combining automatic labelers and expert an- notations for accurate radiology report labeling using bert.arXiv preprint arXiv:2004.09167, 2020

Akshay Smit, Saahil Jain, Pranav Rajpurkar, Anuj Pareek, Andrew Y Ng, and Matthew P Lungren. Chexbert: combining automatic labelers and expert an- notations for accurate radiology report labeling using bert.arXiv preprint arXiv:2004.09167, 2020

work page arXiv 2004

[21] [21]

Ohio supercomputer center, 1987

Ohio Supercomputer Center. Ohio supercomputer center, 1987

work page 1987

[22] [22]

Metransformer: Radi- ology report generation by transformer with multiple learnable expert tokens

Zhanyu Wang, Lingqiao Liu, Lei Wang, and Luping Zhou. Metransformer: Radi- ology report generation by transformer with multiple learnable expert tokens. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 11558–11567, 2023

work page 2023

[23] [23]

R2gengpt: Radiology report generation with frozen llms.Meta-Radiology, 1(3):100033, 2023

Zhanyu Wang, Lingqiao Liu, Lei Wang, and Luping Zhou. R2gengpt: Radiology report generation with frozen llms.Meta-Radiology, 1(3):100033, 2023

work page 2023

[24] [24]

Contrastive knowledge- guided large language models for medical report generation

Yuyang Sha, Hongxin Pan, Weiyu Meng, and Kefeng Li. Contrastive knowledge- guided large language models for medical report generation. InInternational Con- ference on Medical Image Computing and Computer-Assisted Intervention, pages 111–120. Springer, 2025

work page 2025

[25] [25]

Diff-rrg: Lon- gitudinal disease-wise patch difference as guidance for llm-based radiology re- port generation

Hannah Yun, Junyeong Maeng, Eunsong Kang, and Heung-Il Suk. Diff-rrg: Lon- gitudinal disease-wise patch difference as guidance for llm-based radiology re- port generation. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 152–161. Springer, 2025. Precision-Recall Controllable Radiology Report Generation 11

work page 2025

[26] [26]

Enhanced contrastive learning with multi-view longitudinal data for chest x-ray report generation

Kang Liu, Zhuoqi Ma, Xiaolu Kang, Yunan Li, Kun Xie, Zhicheng Jiao, and Qiguang Miao. Enhanced contrastive learning with multi-view longitudinal data for chest x-ray report generation. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 10348–10359, 2025

work page 2025

[27] [27]

MedGemma Technical Report

Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, Atilla Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, Cían Hughes, Charles Lau, et al. Medgemma technical report.arXiv preprint arXiv:2507.05201, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025

[28] [28]

GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization

Shih-Yang Liu, Xin Dong, Ximing Lu, Shizhe Diao, Peter Belcak, Mingjie Liu, Min-Hung Chen, Hongxu Yin, Yu-Chiang Frank Wang, Kwang-Ting Cheng, et al. Gdpo:Groupreward-decouplednormalizationpolicyoptimizationformulti-reward rl optimization.arXiv preprint arXiv:2601.05242, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026

[29] [29]

CURE: Curriculum-guided Multi-task Training for Reliable Anatomy Grounded Report Generation

Pablo Messina, Andrés Villa, Juan León Alcázar, Karen Sánchez, Carlos Hinojosa, Denis Parra, Álvaro Soto, and Bernard Ghanem. Cure: Curriculum-guided multi- task training for reliable anatomy grounded report generation.arXiv preprint arXiv:2601.15408, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026