Pith. sign in

REVIEW 4 major objections 5 minor 34 references

Online Iterative Self-Alignment for Radiology Report Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A lightweight radiology report model can become its own teacher: after supervised training, it generates candidate reports, ranks them with automatic clinical metrics, and aligns itself iteratively through multi-objective preference…

desk verdict A promising self-iteration loop for radiology report generation, but the headline SOTA claim rests on per-metric best rows that no single checkpoint realizes. read the letter →

arxiv 2505.11983 v2 pith:LRQQ2YNS submitted 2025-05-17 cs.CV cs.AI

Ting Xiao , Lei Shi , Yang Zhang , HaoFeng Yang , Zhe Wang , Chenjia Bai This is my paper · ORCID
classification cs.CVcs.AI
keywords radiologyreportgenerationself-alignmentmulti-objectivepreferenceoptimizationdirectautomaticclinicalmetricsiterativetrainingParetofrontchestX-ray
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that a lightweight radiology report generator can improve itself without new radiologist annotations. After supervised training, the model writes many candidate reports for each image under different objective conditions, ranks them with automatic clinical metrics, builds preference pairs from the top and bottom of the rankings, and optimizes itself with multi-objective direct preference optimization. Repeating this loop expands the data distribution beyond the fixed training set, and the paper reports steady metric gains across three iterations on two chest X-ray benchmarks. If true, strong report quality no longer requires a large foundation model or an expensive preference-labeling step.

What carries the argument

The engine is a one-hot weight vector $\hat{w}$ that conditions the model on a single preference objective, together with a multi-objective direct preference optimization (MODPO) loss. For each objective $k$, an automatic metric $M_k$ ranks self-generated reports inside disease-stratified groups; the top-ranked reports become chosen responses and randomly selected lower-ranked reports become rejected responses, forming datasets $\mathcal{D}_k$. The MODPO loss then tunes the policy with a margin term built from the other objectives' reward models, so the weight vector $w$ interpolates between objectives. Self-iteration then sets the reference policy to the updated policy and repeats, and the paper's Theorem 1 bounds the policy suboptimality gap by a coverage term $\lVert \mathbb{E}_{x\sim\rho}[\phi(x,\pi^\star(x))]\rVert_{(\Sigma_{\mathcal{D}_k}+\lambda I)^{-1}}$, which decreases as each new round's preference data better covers the target policy's outputs.

What would settle it

A reader could settle the claim by having radiologists blind-compare third-iteration reports with the supervised baseline on the same MIMIC-CXR images, counting missed findings and hallucinated findings. If clinical quality does not improve, or declines, while the automatic metrics keep rising, the assumption that those metrics are faithful preferences fails.

Watch

Extended reading notes

Core claim

The paper claims that a lightweight supervised radiology report generation model can bootstrap itself into a stronger model through an online loop that uses only its own outputs and automatic metrics. In each round, the model, conditioned on a one-hot weight vector that selects a preference objective, generates diverse candidate reports for chest X-rays; automatic radiology metrics (RadCliQ, RadGraphF1, GREEN) rank those candidates; top-ranked reports become chosen responses and randomly sampled lower-ranked reports become rejected responses; and the model is updated with multi-objective direct preference optimization so that a user-supplied weight vector steers the report toward fluency, clinical accuracy, or a balance of the two. The updated model then replaces the reference model and the loop repeats. The paper reports that this loop improves all evaluated metrics across three iterations and outperforms prior methods on MIMIC-CXR and IU-Xray, including larger vision-language models, and it proves a bound showing that the suboptimality gap shrinks as the new preference data covers the target policy's output distribution.

Load-bearing premise

The automatic radiology metrics used to rank self-generated reports (RadCliQ, RadGraphF1, GREEN) faithfully capture what clinicians value in a report; if they reward superficial fluency or miss serious findings, iteration amplifies those metric biases.

Editorial extensions

If this is right

  • A lightweight RRG model (about 230M parameters) can reach or surpass much larger vision-language baselines on report metrics without any new radiologist-written preference labels.
  • Because each iteration starts from the previous round's policy, the preference data itself gets higher-scoring over rounds: RadGraphF1 and GREEN rise and RadCliQ falls, and the final model improves on every metric compared with the SFT baseline.
  • A single trained policy can serve different clinical needs by changing the preference weight vector, producing a Pareto front between objectives instead of one fixed report style.
  • The theoretical bound implies that adding rounds of self-generated preference data is not just regularization: it reduces the suboptimality gap through better coverage of the target policy's output distribution.
  • The per-iteration training cost is small relative to supervised fine-tuning because each preference dataset contains 10,000 pairs rather than the full 227,835-image training set.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same loop should transfer to other reporting tasks with trustworthy automatic graders, but its ceiling is set by those graders; where the metric disagrees with clinician judgment, the loop will amplify the metric's blind spots.
  • A testable extension is to inject a small clinician-ranked seed set at each iteration to check whether the metric-based chosen reports agree with clinician picks; disagreement rates would predict where self-alignment degrades.
  • The reliance on BERTScore during deduplication may bias early-round preference data toward formulaic language, so gains on NLG-style metrics do not by themselves imply that clinically important findings are reported more completely.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes Online Iterative Self-Alignment (OISA), a four-stage pipeline (self-generation, self-evaluation, self-alignment, self-iteration) for radiology report generation (RRG). The method conditions a lightweight RRG policy on a preference weight vector, generates diverse self-reports, ranks them with three automatic radiology metrics (RadCliQ, RadGraphF1, GREEN) to build multi-objective preference data, and optimizes the policy via Multi-Objective Direct Preference Optimization (MODPO). The authors claim that this iterative process extends data coverage and achieves state-of-the-art performance across multiple evaluation metrics on MIMIC-CXR and IU-Xray. A theoretical bound on the sub-optimality gap is provided under linear reward assumptions.

Significance. If the central claim were established, the paper would be a valuable contribution: it shows that a lightweight RRG model can improve its own outputs without expert annotations, using automatic metrics for preference construction, and it provides a multi-objective conditioning mechanism that can trade off fluency and clinical accuracy. The idea of iterative self-alignment with a small model, as an alternative to using large foundation models for preference data, is timely and of practical interest. However, the significance is currently weakened by the way the empirical results are reported and by the overlap between the training-time preference metrics and the evaluation metrics; the paper's own limitation section concedes that these metrics may not reflect clinician needs. Consequently, the state-of-the-art claim is not substantiated by the evidence as presented.

major comments (4)
  1. [Tables 3-4, Section 1] The 'Ours (iteration N)' rows in Tables 3 and 4 are not the output of a single model. On MIMIC-CXR iteration 3, the values for B1/B4/BERTScore/RadCliQ are taken from the w=(1,0,0) configuration in Table 1, RadGraphF1/CheXbertF1 from w=(0,1,0), and GREEN from w=(0,0,1). No single weight vector produces all seven numbers, and the text uses these per-column maxima to claim 'state-of-the-art performance across multiple evaluation metrics.' Even the assembled row does not beat MPO on BLEU-4 (0.129 vs 0.139 in Table 3). The paper should report a single evaluable checkpoint (e.g., the equal-weight model) or explicitly label the rows as oracle-style per-metric bests, and the SOTA claim in the Abstract and Section 1 must be revised accordingly.
  2. [Section 3.1, Section 5.1, Section 8] The preference labels used for training are constructed by ranking self-generated reports with RadCliQ, RadGraphF1, and GREEN, and the evaluation in Section 5.1 uses exactly those three metrics as the main radiology metrics. Since the policy is optimized against the same scoring functions that define the evaluation, improvements on these metrics—and on correlated metrics such as BLEU/BERTScore (which the authors note in Section 5.2 are linearly combined into RadCliQ) and CheXbertF1 (correlated with RadGraphF1)—may reflect metric overfitting rather than clinically better reports. The paper's own Section 8 admits that these metrics 'may not be consistent with the actual needs of clinicians.' Without an independent evaluation (e.g., radiologist assessment or a clinical metric not used in preference construction), the claim that OISA produces better radiology reports is not supported.
  3. [Appendix C, Eq. (16)] The theoretical guarantee of iterative improvement in Lemma 2 relies on Inequality (16), which is asserted to 'generally hold' without proof. This inequality states that the preference dataset collected from the previous policy covers the new policy at least as well as the ground-truth optimal policy; this is a coverage assumption, not a consequence of the preceding derivation. The theorem should explicitly state this as an assumption (or provide a proof under stated conditions), and the text in Section 4 that claims a 'theoretically grounded improvement' should be tempered accordingly.
  4. [Tables 1-4, Appendix D.2] All experiments are reported from a single run, with no error bars, multiple seeds, or statistical significance tests. The per-iteration gains are small in several cases (e.g., BERTScore 0.885 vs 0.879 and RadCliQ 2.54 vs 2.56 on MIMIC-CXR in Table 1), and the hyperparameter analysis in Table 5 only varies β and shows that β=0.8 gives a slightly higher BLEU-1 than the chosen β=0.5. Without variance estimates, the claimed monotonic improvement across iterations is not statistically established.
minor comments (5)
  1. [Figure 1] Figure 1 contains the stray annotation 'Xiaoting2.16', which appears to be an author comment left in the figure; it should be removed.
  2. [Section 3.1, Section 5.2] There are typos: 'promt set' should be 'prompt set', and 'BLUE' should be 'BLEU'. Also, the dataset name 'Iu-Xray' in the Table 2 caption is inconsistent with 'IU-Xray' used elsewhere.
  3. [Tables 1-2 captions] The captions state 'Bold blue denote the best results' but the table uses bold and underline without a color legend; the formatting description should be clarified.
  4. [Section 5.2, IU-Xray evaluation] The paper states that on IU-Xray the model is 'tested on the entire IU-XRay set' following PromptMRG, but it does not specify whether the model weights are the MIMIC-CXR-trained ones (zero-shot transfer) or a model retrained on IU-Xray. This should be stated explicitly to make the comparison interpretable.
  5. [Appendix D.2] The choice of β=0.5 is not clearly justified by Table 5: β=0.8 gives BLEU-1 0.419 and BLEU-4 0.119, which are marginally higher than for β=0.5. A sentence explaining the selection criterion would be helpful.

Circularity Check

2 steps flagged · score 6.0 of 10

Partial circularity: the reported SOTA row is an oracle-selected per-column maximum across weight configurations, and the main evaluation metrics overlap with the preference-label rewards; the iterative training loop itself is not circular.

  1. fitted input called prediction [Section 5.2, Table 3 vs. Table 1 (MIMIC-CXR comparison)]
    "Ours (iteration 3) 0.428 0.129 0.885 2.54 0.273 0.516 0.341; Table 1 Iteration 3 rows: 1 0 0 0.428* 0.129* 0.885* 2.54* 0.244 0.486 0.322; 0 1 0 0.418 0.117 0.872 2.65 0.273* 0.516* 0.324; 0 0 1 0.417 0.116 0.873 2.66 0.249 0.484 0.341*."

    The 'Ours (iteration 3)' row in Table 3 is exactly the per-column maximum of the three one-hot-weight rows in Table 1: B1/B4/BERTScore/RadCliQ come from w=(1,0,0), RadGraphF1/CheXbertF1 from w=(0,1,0), and GREEN from w=(0,0,1). No single checkpoint attains all seven values simultaneously, yet the row is presented as one model's result. The claimed SOTA prediction is therefore the test-set selection itself, not the output of a single policy; even this assembled row loses to MPO on BLEU-4 (0.139 vs. 0.129).

  2. other [Section 3.1, Section 5.1, Section 5.2]
    "We use three radiology metrics, RadCliQ, RadGraphF1, and GREEN to represent different objectives, resulting in a preference dataset with N=3 objectives ... The performance trends of BLUE and BERTScore align with those of RadCliQ, as RadCliQ is a linear combination of these metrics."

    The exact functions used as rewards to rank self-generated reports (RadCliQ, RadGraphF1, GREEN) are also the primary radiology evaluation metrics in Section 5.1, and RadCliQ is a linear combination containing BLEU and BERTScore. Thus much of the reported multi-metric improvement is the training target itself, not independent confirmation. The paper's own Section 8 concedes the metrics 'may not be consistent with the actual needs of clinicians,' so metric gains do not by themselves establish clinical quality. The held-out test split prevents full circularity, but the 'multiple metrics' claim is partly definitional.

full rationale

Most of OISA's training loop is not circular: preference data are built from the training split, MODPO is an external algorithm, and the theoretical bound is a standard MLE/RLHF argument under a linear-reward assumption. The paper's self-citations (MPO) are prior-work comparisons, not load-bearing justifications. The circularity lies in the presentation of the headline empirical result. The 'Ours (iteration 3)' rows in Tables 3 and 4 are per-column bests across different one-hot weight configurations from Tables 1 and 2, so the state-of-the-art claim is an oracle-selected aggregate rather than the output of a single model. In addition, the same metrics used to build preference labels (RadCliQ, RadGraphF1, GREEN) are the main evaluation metrics, with RadCliQ linearly including BLEU/BERTScore, so several reported metrics are partly training targets. Because the test set is held out and the iterative training itself is coherent, this is partial rather than total circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central empirical result depends on one fitted hyperparameter (beta), two hand-set deduplication thresholds, a hand-set weight sampling grid, and the assumption that automatic radiology metrics proxy clinician preferences. The theoretical guarantee additionally assumes linear rewards and an unproven coverage inequality in Eq. (16). No new physical entities are introduced.

free parameters (4)
  • KL penalty beta = 0.5
    Selected from the candidate set {0.1, 0.5, 0.8, 1.0} in first-iteration experiments on MIMIC-CXR (Appendix D.2); the paper notes results are not very sensitive to it.
  • BERTScore deduplication thresholds = 0.5 and 0.8
    Appendix A discards reports below 0.5 BERTScore and deduplicates pairs above 0.8; the thresholds are described as empirical and affect preference data diversity and quality.
  • Preference weight sampling space = {0.2, 0.4, 0.6, 0.8, 1.0}
    Sampled during training to produce Pareto fronts (Section 5.1); chosen by hand with no sensitivity analysis reported.
  • Preference dataset size K = 10,000 pairs per objective
    Each of the three preference datasets is fixed at 10,000 pairs; chosen for computational cost rather than derived from theory.
assumptions (5)
  • standard math Bradley-Terry model of pairwise preferences
    Used in Eqn. (2) to convert report preferences into a DPO-style loss; standard in preference learning.
  • domain assumption Assumption 1: each objective reward is linear, r_theta(x,y) = theta^T phi(x,y), with phi known and bounded
    Stated in Section 4; needed for Lemma 1 and Theorem 1 regret bound. It is a strong simplification and is not verified in experiments.
  • standard math Lemma 1 from Zhu et al. (2023) bounding MLE estimation error for DPO rewards
    Imported from cited work and used without proof in Section 4; accepted if the source theorem holds.
  • ad hoc to paper Eq. (16): the dataset collected from the previous policy covers the new policy at least as well as the ground-truth optimal policy
    Assumed in the Appendix C proof of Theorem 1 with the phrase 'generally holds'; it is not proved or tested, but it is needed for the claimed theoretical improvement.
  • domain assumption Automatic radiology metrics correlate with radiologist preferences
    Section 3.1 uses RadCliQ, RadGraphF1, and GREEN to label preference data; Section 8 acknowledges these metrics may not match clinician needs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Online Iterative Self-Alignment for Radiology Report Generation." pith.science (2026). https://pith.science/paper/LRQQ2YNS

@misc{pith2026250511983,
  author       = {Pith},
  title        = {Pith review of: Online Iterative Self-Alignment for Radiology Report Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LRQQ2YNS}},
  note         = {Machine review of arXiv:2505.11983}
}
read the original abstract

Radiology Report Generation (RRG) is an important research topic for relieving radiologist' heavy workload. Existing RRG models mainly rely on supervised fine-tuning (SFT) based on different model architectures using data pairs of radiological images and corresponding radiologist-annotated reports. Recent research has shifted focus to post-training improvements, aligning RRG model outputs with human preferences using reinforcement learning (RL). However, the limited data coverage of high-quality annotated data poses risks of overfitting and generalization. This paper proposes a novel Online Iterative Self-Alignment (OISA) method for RRG that consists of four stages: self-generation of diverse data, self-evaluation for multi-objective preference data,self-alignment for multi-objective optimization and self-iteration for further improvement. Our approach allows for generating varied reports tailored to specific clinical objectives, enhancing the overall performance of the RRG model iteratively. Unlike existing methods, our frame-work significantly increases data quality and optimizes performance through iterative multi-objective optimization. Experimental results demonstrate that our method surpasses previous approaches, achieving state-of-the-art performance across multiple evaluation metrics.

Figures

Figures reproduced from arXiv: 2505.11983 by the authors.

Figure 1
Figure 1. The illustration of the proposed OISA pipeline, comprising the Preference Dataset Construction (PDC) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The distribution of preference dataset on dif [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. The multi-objective alignment fronts across three iterations. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 6 canonical work pages

  1. [1]

    Asma Alkhaldi, Raneem Alnajim, Layan Alabdullatef, Rawan Alyahya, Jun Chen, Deyao Zhu, Ahmed Alsinan, and Mohamed Elhoseiny. 2024. https://arxiv.org/abs/2407.04106 Minigpt-med: Large language model as a general interface for radiology diagnosis . arXiv preprint arXiv:2407.04106

    arXiv 2024

  2. [2]

    Ralph Allan Bradley and Milton E Terry. 1952. https://doi.org/10.2307/2334029 Rank analysis of incomplete block designs: I. the method of paired comparisons . Biometrika, 39(3/4):324--345

    doi:10.2307/2334029 1952

  3. [3]

    Yiming Cao, Lizhen Cui, Lei Zhang, Fuqiang Yu, Zhen Li, and Yonghui Xu. 2023. https://doi.org/10.1609/aaai.v37i1.25100 Mmtn: multi-modal memory transformer network for image-report consistent medical report generation . In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 277--285

    doi:10.1609/aaai.v37i1.25100 2023

  4. [4]

    Xiaoyu Chen, Han Zhong, Zhuoran Yang, Zhaoran Wang, and Liwei Wang. 2022. https://doi.org/10.48550/arXiv.2205.11140 Human-in-the-loop: Provably efficient preference-based reinforcement learning with general function approximation . In International Conference on Machine Learning, pages 3773--3793. PMLR

    arXiv doi:10.48550/arxiv.2205.11140 2022

  5. [5]

    Zhihong Chen, Yaling Shen, Yan Song, and Xiang Wan. 2021. https://doi.org/10.18653/v1/2021.acl-long.459 Cross-modal memory networks for radiology report generation . In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), ...

    doi:10.18653/v1/2021.acl-long.459 2021

  6. [6]

    Zhihong Chen, Yan Song, Tsung-Hui Chang, and Xiang Wan. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.112 Generating radiology reports via memory-driven transformer . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1439--1449

    doi:10.18653/v1/2020.emnlp-main.112 2020

  7. [7]

    Zhihong Chen, Maya Varma, Jean-Benoit Delbrouck, Magdalini Paschali, Louis Blankemeier, Dave Van Veen, Jeya Maria Jose Valanarasu, Alaa Youssef, Joseph Paul Cohen, Eduardo Pontes Reis, et al. 2024. https://doi.org/10.48550/arXiv.2401.12208 Chexagent: Towards a foundation model for chest x-ray interpretation . In AAAI 2024 Spring Symposium on Clinical Foun...

    arXiv doi:10.48550/arxiv.2401.12208 2024

  8. [8]

    Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. 2017. https://doi.org/10.48550/arXiv.1706.03741 Deep reinforcement learning from human preferences . Advances in neural information processing systems, 30

    arXiv doi:10.48550/arxiv.1706.03741 2017

Show all 34 references
  1. [9]

    Jean-Benoit Delbrouck, Pierre Chambon, Christian Bluethgen, Emily Tsai, Omar Almusa, and Curtis Langlotz. 2022. https://doi.org/10.18653/v1/2022.findings-emnlp.319 Improving the factual correctness of radiology report generation with semantic rewards . In Findings of the Assoc...

    2022 doi

  2. [10]

    Dina Demner-Fushman, Marc D Kohli, Marc B Rosenman, Sonya E Shooshan, Laritza Rodriguez, Sameer Antani, George R Thoma, and Clement J McDonald. 2016. https://doi.org/10.1093/jamia/ocv080 Preparing a collection of radiology examinations for distribution and retrieval . Journal ...

    2016 doi

  3. [11]

    Dennis Hein, Zhihong Chen, Sophie Ostmeier, Justin Xu, Maya Varma, Eduardo Pontes Reis, Arne Edward Michalson, Christian Bluethgen, Hyun Joo Shin, Curtis Langlotz, et al. 2024. https://doi.org/10.48550/arXiv.2410.07025 Preference fine-tuning for factuality in chest x-ray inter...

    2024 doi arXiv

  4. [12]

    Zhongzhen Huang, Xiaofan Zhang, and Shaoting Zhang. 2023. https://doi.org/10.48550/arXiv.2306.11345 Kiut: Knowledge-injected u-transformer for radiology report generation . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19809--19818

    2023 doi arXiv

  5. [13]

    Saahil Jain, Ashwin Agrawal, Adriel Saporta, Steven QH Truong, Du Nguyen Duong, Tan Bui, Pierre Chambon, Yuhao Zhang, Matthew P Lungren, Andrew Y Ng, et al. 2021. https://doi.org/10.48550/arXiv.2106.14463 Radgraph: Extracting clinical entities and relations from radiology repo...

    2021 doi arXiv

  6. [14]

    Haibo Jin, Haoxuan Che, Yi Lin, and Hao Chen. 2024. https://doi.org/10.48550/arXiv.2308.12604 Promptmrg: Diagnosis-driven prompts for medical report generation . In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 2607--2615

    2024 doi arXiv

  7. [15]

    Alistair EW Johnson, Tom J Pollard, Nathaniel R Greenbaum, Matthew P Lungren, Chih-ying Deng, Yifan Peng, Zhiyong Lu, Roger G Mark, Seth J Berkowitz, and Steven Horng. 2019. https://doi.org/10.48550/arXiv.1901.07042 Mimic-cxr-jpg, a large publicly available database of labeled...

    2019 doi arXiv

  8. [16]

    knowledge is power

    Kaveri Kale, Pushpak Bhattacharyya, Aditya Shetty, Milind Gune, Kush Shrivastava, Rustom Lawyer, and Spriha Biswas. 2023. https://doi.org/10.18653/v1/2023.acl-industry.2 “knowledge is power”: Constructing knowledge graph of abdominal organs and using them for automatic radiolo...

    2023 doi

  9. [17]

    Chang Liu, Yuanhe Tian, Weidong Chen, Yan Song, and Yongdong Zhang. 2024. https://doi.org/10.1609/aaai.v38i17.29826 Bootstrapping large language models for radiology report generation . In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 18635--18643

    2024 doi

  10. [18]

    Moseley, Curtis P

    Sophie Ostmeier, Justin Xu, Zhihong Chen, Maya Varma, Louis Blankemeier, Christian Bluethgen, Arne Md, Michael E. Moseley, Curtis P. Langlotz, Akshay Chaudhari, and Jean - Benoit Delbrouck. 2024. https://aclanthology.org/2024.findings-emnlp.21 GREEN: generative radiology repor...

    2024

  11. [19]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei - Jing Zhu. 2002. https://doi.org/10.3115/1073083.1073135 Bleu: a method for automatic evaluation of machine translation . In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, July 6-12, ...

    2002

  12. [20]

    Han Qin and Yan Song. 2022. https://doi.org/10.18653/v1/2022.findings-acl.38 Reinforced cross-modal alignment for radiology report generation . In Findings of the Association for Computational Linguistics: ACL 2022, pages 448--458

    2022 doi

  13. [21]

    Hongyu Shen, Mingtao Pei, Juncai Liu, and Zhaoxing Tian. 2024. https://doi.org/10.1609/aaai.v38i5.28279 Automatic radiology reports generation via memory alignment network . In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 4776--4783

    2024 doi

  14. [22]

    Akshay Smit, Saahil Jain, Pranav Rajpurkar, Anuj Pareek, Andrew Y Ng, and Matthew Lungren. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.117 Combining automatic labelers and expert annotations for accurate radiology report labeling using bert . In Proceedings of the 2020 C...

    2020 doi

  15. [23]

    Yuanhao Wang, Qinghua Liu, and Chi Jin. 2023. Is rlhf more difficult than standard rl? a theoretical perspective. Advances in Neural Information Processing Systems, 36:76006--76032

    2023

  16. [24]

    Zhanyu Wang, Mingkang Tang, Lei Wang, Xiu Li, and Luping Zhou. 2022. https://doi.org/10.48550/arXiv.2208.10358 A medical semantic-assisted transformer for radiographic report generation . In International Conference on Medical Image Computing and Computer-Assisted Intervention...

    2022 doi arXiv

  17. [25]

    Zhanyu Wang, Luping Zhou, Lei Wang, and Xiu Li. 2021. https://doi.org/10.1109/CVPR46437.2021.00246 A self-boosting framework for automated radiographic report generation . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2433--2442

    2021

  18. [26]

    Ting Xiao, Lei Shi, Peng Liu, Zhe Wang, and Chenjia Bai. 2025. Radiology report generation via multi-objective preference optimization. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 8664--8672

    2025

  19. [27]

    Wei Xiong, Hanze Dong, Chenlu Ye, Ziqi Wang, Han Zhong, Heng Ji, Nan Jiang, and Tong Zhang. 2024. https://doi.org/10.48550/arXiv.2312.11456 Iterative preference learning from human feedback: Bridging theory and practice for rlhf under kl-constraint . In Forty-first Internation...

    2024 doi arXiv

  20. [28]

    Feiyang Yu, Mark Endo, Rayan Krishnan, Ian Pan, Andy Tsai, Eduardo Pontes Reis, Eduardo Kaiser Ururahy Nunes Fonseca, Henrique Min Ho Lee, Zahra Shakeri Hossein Abad, Andrew Y Ng, et al. 2023. https://doi.org/10.1016/j.patter.2023.100802 Evaluating progress in automatic chest ...

    2023

  21. [29]

    Weinberger, and Yoav Artzi

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. https://openreview.net/forum?id=SkeHuCVFDr Bertscore: Evaluating text generation with BERT . In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April...

    2020

  22. [30]

    Hong-Yu Zhou, Subathra Adithan, Juli \'a n Nicol \'a s Acosta, Eric J Topol, and Pranav Rajpurkar. 2024 a . https://doi.org/10.48550/arXiv.2405.07988 A generalist learner for multifaceted medical image interpretation . arXiv preprint arXiv:2405.07988

    2024 doi arXiv

  23. [31]

    Zhanhui Zhou, Jie Liu, Jing Shao, Xiangyu Yue, Chao Yang, Wanli Ouyang, and Yu Qiao. 2024 b . https://doi.org/10.18653/v1/2024.findings-acl.630 Beyond one-preference-fits-all alignment: Multi-objective direct preference optimization . In Findings of the Association for Computa...

    2024 doi

  24. [32]

    Banghua Zhu, Michael Jordan, and Jiantao Jiao. 2023. https://doi.org/10.48550/arXiv.2301.11270 Principled reinforcement learning with human feedback from pairwise or k-wise comparisons . In International Conference on Machine Learning, pages 43037--43067. PMLR

    2023 doi arXiv

  25. [33]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  26. [34]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.