pith. sign in

arxiv: 2506.03178 · v2 · pith:ZUF6SDU7new · submitted 2025-05-29 · 📡 eess.IV · cs.AI· cs.CV

LLaMA-XR: A Novel Framework for Radiology Report Generation using LLaMA and QLoRA Fine Tuning

Md. Zihad Bin Jahangir , Muhammad Ashad Kabir , Sumaiya Akter , Israt Jahan , Minh Chau This is my paper

Pith reviewed 2026-05-19 13:36 UTC · model grok-4.3

classification 📡 eess.IV cs.AIcs.CV
keywords radiology report generationLLaMAQLoRAchest X-raymedical image captioningfine-tuningnatural language generationclinical accuracy
0
0 comments X

The pith

LLaMA-XR generates more coherent and clinically accurate radiology reports from chest X-rays by pairing LLaMA 3.1 with DenseNet-121 embeddings and QLoRA fine-tuning.

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

The paper presents LLaMA-XR as a system that takes chest radiographs, extracts visual features with DenseNet-121, and feeds them into a LLaMA 3.1 model that has been adapted with QLoRA for efficient training. The authors report that this produces reports scoring 0.433 on ROUGE-L and 0.336 on METEOR on the IU X-ray dataset, beating prior methods while using less memory and running faster. A reader would care because automated reports could lighten the routine workload for radiologists without demanding expensive hardware. The work frames this as a practical step toward reliable AI assistance in diagnostic imaging.

Core claim

LLaMA-XR integrates LLaMA 3.1 with DenseNet-121-based image embeddings and Quantized Low-Rank Adaptation (QLoRA) fine-tuning. On the IU X-ray benchmark dataset it reaches a ROUGE-L score of 0.433 and a METEOR score of 0.336, outperforming existing methods in coherence and clinical accuracy while preserving computational efficiency through optimized parameter utilization and reduced memory overhead.

What carries the argument

QLoRA-adapted LLaMA 3.1 conditioned on DenseNet-121 image embeddings, which enables memory-efficient fine-tuning for medical report generation from radiographs.

If this is right

  • Outperforms prior state-of-the-art methods on the standard IU X-ray benchmark.
  • Produces reports with greater coherence and clinical accuracy.
  • Generates reports faster while requiring lower computational resources.
  • Provides enhanced clinical utility and reliability for automated radiology reporting.

Where Pith is reading between the lines

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

  • If the metric gains translate to real clinical settings, hospitals could deploy similar systems to draft initial reports and let radiologists focus on ambiguous cases.
  • The quantized adaptation technique may allow other large language models to be specialized for additional medical imaging modalities without large compute budgets.
  • Testing the same architecture on larger, multi-institutional radiology datasets would reveal whether the reported improvements hold outside the IU X-ray collection.

Load-bearing premise

Higher scores on automatic similarity metrics such as ROUGE-L and METEOR reliably indicate improved clinical accuracy and usefulness in the generated reports.

What would settle it

A head-to-head evaluation in which practicing radiologists rate the clinical accuracy, completeness, and diagnostic utility of LLaMA-XR reports against both human-written ground truth and outputs from prior models, showing no meaningful advantage for the new system.

Figures

Figures reproduced from arXiv: 2506.03178 by Israt Jahan, Md. Zihad Bin Jahangir, Minh Chau, Muhammad Ashad Kabir, Sumaiya Akter.

Figure 1
Figure 1. Figure 1: Overview of the proposed model architecture for radiology report generation. The X-ray images (AP and LAT views) are processed [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the “DenseNet121-res224-all” output classes. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The heatmap highlights the critical regions in a chest X-ray image that influenced the Densenet-121 model’s classification decision. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example prompt for fine-tuning and gradient stability, the batch size was set to 8, with gradient accumulation steps of 4, effectively simulating a batch size of 32. The model was trained for three complete epochs, which was sufficient for convergence, given the size of the training dataset and the complexity of the task. A learning rate of 2 × 10−6 was used, optimized using the AdamW 8-bit optimizer (opti… view at source ↗
read the original abstract

Automated radiology report generation holds significant potential to reduce radiologists' workload and enhance diagnostic accuracy. However, generating precise and clinically meaningful reports from chest radiographs remains challenging due to the complexity of medical language and the need for contextual understanding. Existing models often struggle with maintaining both accuracy and contextual relevance. In this paper, we present LLaMA-XR, a novel framework that integrates LLaMA 3.1 with DenseNet-121-based image embeddings and Quantized Low-Rank Adaptation (QLoRA) fine-tuning. LLaMA-XR achieves improved coherence and clinical accuracy while maintaining computational efficiency. This efficiency is driven by an optimization strategy that enhances parameter utilization and reduces memory overhead, enabling faster report generation with lower computational resource demands. Extensive experiments conducted on the IU X-ray benchmark dataset demonstrate that LLaMA-XR outperforms a range of state-of-the-art methods. Our model achieves a ROUGE-L score of 0.433 and a METEOR score of 0.336, establishing new performance benchmarks in the domain. These results underscore LLaMA-XR's potential as an effective and efficient AI system for automated radiology reporting, offering enhanced clinical utility and reliability.

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.

Referee Report

2 major / 1 minor

Summary. The paper introduces LLaMA-XR, a framework integrating LLaMA 3.1 with DenseNet-121 image embeddings and QLoRA fine-tuning for automated generation of radiology reports from chest X-rays. It claims improved coherence and clinical accuracy with computational efficiency, reporting ROUGE-L of 0.433 and METEOR of 0.336 on the IU X-ray benchmark while outperforming state-of-the-art methods.

Significance. If properly validated, the use of QLoRA for efficient adaptation of LLaMA to medical report generation could offer a practical contribution to resource-efficient LLM fine-tuning in radiology. However, the current results rest on automatic lexical metrics without demonstrated links to clinical utility, limiting the work's immediate significance for diagnostic applications.

major comments (2)
  1. [Abstract] Abstract: The claim that ROUGE-L = 0.433 and METEOR = 0.336 establish 'improved coherence and clinical accuracy' plus 'enhanced clinical utility' is unsupported. These metrics quantify n-gram overlap with reference reports and do not assess omission of critical findings, hallucinated pathologies, or diagnostic correctness; no radiologist scoring or factuality metrics (e.g., RadGraph, CheXbert) are referenced to bridge this gap.
  2. [Abstract] Abstract: The assertion of outperforming 'a range of state-of-the-art methods' supplies no information on the exact baselines, statistical significance tests, ablation studies isolating the contribution of DenseNet-121 embeddings or QLoRA, or details on train/validation/test splits and data handling for the IU X-ray dataset, leaving the central empirical claim without visible supporting evidence.
minor comments (1)
  1. The methods section should include explicit details on QLoRA rank, scaling factors, learning rate schedule, and exact training procedure to support reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below, indicating where revisions to the manuscript are planned.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that ROUGE-L = 0.433 and METEOR = 0.336 establish 'improved coherence and clinical accuracy' plus 'enhanced clinical utility' is unsupported. These metrics quantify n-gram overlap with reference reports and do not assess omission of critical findings, hallucinated pathologies, or diagnostic correctness; no radiologist scoring or factuality metrics (e.g., RadGraph, CheXbert) are referenced to bridge this gap.

    Authors: We agree that ROUGE-L and METEOR are lexical overlap metrics and do not directly measure clinical accuracy, factuality, omission of findings, or hallucination of pathologies. The abstract phrasing overstated the clinical implications of these scores. In the revised manuscript we will rephrase the abstract to report the metric values as performance on standard automatic evaluation benchmarks without claiming direct clinical accuracy or utility. We will also add a limitations paragraph that explicitly notes the scope of these metrics and identifies clinical validation and factuality metrics (such as CheXbert-based entity extraction) as important directions for future work. revision: yes

  2. Referee: [Abstract] Abstract: The assertion of outperforming 'a range of state-of-the-art methods' supplies no information on the exact baselines, statistical significance tests, ablation studies isolating the contribution of DenseNet-121 embeddings or QLoRA, or details on train/validation/test splits and data handling for the IU X-ray dataset, leaving the central empirical claim without visible supporting evidence.

    Authors: The experimental section of the manuscript contains the full set of baseline comparisons, but the abstract is too concise to convey the necessary details. We will revise the abstract to name the primary state-of-the-art methods against which improvements are reported. We will also ensure the methods and results sections clearly document the train/validation/test splits used on IU X-ray, any statistical significance testing performed, and ablation experiments that isolate the contributions of the DenseNet-121 encoder and QLoRA adaptation. These details will be summarized or cross-referenced so that the empirical claims are fully supported. revision: yes

Circularity Check

0 steps flagged

No circularity: standard empirical fine-tuning and benchmark evaluation

full rationale

The paper presents an empirical ML framework that combines LLaMA 3.1, DenseNet-121 image embeddings, and QLoRA fine-tuning, then reports ROUGE-L and METEOR scores on the IU X-ray dataset after training. No mathematical derivation chain exists that reduces claimed outputs to inputs by construction. Performance numbers are obtained via conventional train/test splits and standard NLP metrics; they are not self-defined, fitted parameters renamed as predictions, or justified solely by self-citations. The central claim of improved performance rests on external benchmark comparison rather than tautological redefinition, satisfying the criteria for a self-contained empirical result.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the validity of standard NLP metrics as proxies for clinical quality and on the representativeness of the IU X-ray benchmark for real-world radiology reporting.

free parameters (1)
  • QLoRA rank and scaling factors
    Low-rank adaptation parameters are selected to control memory use and are not derived from first principles.
axioms (1)
  • domain assumption ROUGE-L and METEOR scores are adequate proxies for clinical accuracy of radiology reports.
    Invoked when the abstract equates higher metric values with improved clinical accuracy and utility.

pith-pipeline@v0.9.0 · 5771 in / 1485 out tokens · 85707 ms · 2026-05-19T13:36:22.483067+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

64 extracted references · 64 canonical work pages · 4 internal anchors

  1. [1]

    X. Wang, Y . Peng, L. Lu, Z. Lu, R. M. Summers, Tienet: Text-image embedding network for common thorax disease classification and reporting in chest x-rays, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 9049–9058. doi:10.1109/cvpr.2018.00943

    work page doi:10.1109/cvpr.2018.00943 2018

  2. [2]

    R. M. MR, et al., Acquired heart disease in adults: what can a chest x-ray tell us?, Radiologia 59 (2017) 446–459

    work page 2017

  3. [3]

    S. Bahl, T. Ramzan, R. Maraj, Interpretation and documentation of chest x-rays in the acute medical unit, Clinical Medicine 20 (2020) s73

    work page 2020

  4. [4]

    Liu, T.-M

    G. Liu, T.-M. H. Hsu, M. McDermott, W. Boag, W.-H. Weng, P. Szolovits, M. Ghassemi, Clinically accurate chest x-ray report generation, in: Machine Learning for Healthcare Conference, PMLR, 2019, pp. 249–269

    work page 2019

  5. [5]

    Sloan, P

    P. Sloan, P. Clatworthy, E. Simpson, M. Mirmehdi, Automated radiology report generation: A review of recent advances, IEEE Reviews in Biomedical Engineering (2024). doi:10.1109/RBME.2024.3408456

    work page doi:10.1109/rbme.2024.3408456 2024

  6. [6]

    Q. You, H. Jin, Z. Wang, C. Fang, J. Luo, Image captioning with semantic attention, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 4651–4659. doi:10.1109/CVPR.2016.503

    work page doi:10.1109/cvpr.2016.503 2016

  7. [7]

    F. Liu, X. Ren, Y . Liu, H. Wang, X. Sun, simnet: Stepwise image-topic merging network for generating detailed and comprehensive image captions, 2018. arXiv:1808.08732

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  8. [8]

    Vinyals, A

    O. Vinyals, A. Toshev, S. Bengio, D. Erhan, Show and tell: A neural image caption generator, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3156–3164. doi:10.1109/CVPR.2015.7298935

    work page doi:10.1109/cvpr.2015.7298935 2015

  9. [9]

    Iftikhar, Iqra naz, anmol zahra, and syeda zainab yousuf zaidi

    S. Iftikhar, Iqra naz, anmol zahra, and syeda zainab yousuf zaidi. 2022. report generation of lungs diseases from chest x-ray using nlp”, International Journal of Innovations in Science & Technology 3 (2022) 223–233

    work page 2022

  10. [10]

    Ranjit, G

    M. Ranjit, G. Ganapathy, R. Manuel, T. Ganu, Retrieval augmented chest x-ray report generation using openai gpt models, in: Machine Learning for Healthcare Conference, PMLR, 2023, pp. 650–666

    work page 2023

  11. [11]

    L. C. Adams, D. Truhn, F. Busch, A. Kader, S. M. Niehues, M. R. Makowski, K. K. Bressem, Leveraging gpt-4 for post hoc transformation of free-text radiology reports into structured reporting: a multilingual feasibility study, Radiology 307 (2023) e230725. doi:10.1148/radiol. 230725

    work page doi:10.1148/radiol 2023

  12. [12]

    Buckley, J

    T. Buckley, J. Diao, R. Adam, A. Manrai, Accuracy of a vision-language model on challenging medical cases, 2023. arXiv:2311.05591

    work page arXiv 2023

  13. [13]

    Z. Liu, Y . Huang, X. Yu, L. Zhang, Z. Wu, C. Cao, H. Dai, L. Zhao, Y . Li, P. Shu, F. Zeng, L. Sun, W. Liu, D. Shen, Q. Li, T. Liu, D. Zhu, X. Li, Deid-gpt: Zero-shot medical text de-identification by gpt-4, 2023. arXiv:2303.11032

    work page arXiv 2023

  14. [14]

    T. H. Kung, M. Cheatham, A. Medenilla, C. Sillos, L. De Leon, C. Elepa ˜no, M. Madriaga, R. Aggabao, G. Diaz-Candido, J. Maningo, et al., Performance of chatgpt on usmle: potential for ai-assisted medical education using large language models, PLoS digital health 2 (2023) e0000198. doi:10.1371/journal.pdig.0000198

    work page doi:10.1371/journal.pdig.0000198 2023

  15. [15]

    & Chen, C

    T. Tanida, P. M ¨uller, G. Kaissis, D. Rueckert, Interactive and explainable region-guided radiology report generation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 7433–7442. doi:10.1109/CVPR52729.2023.00718

    work page doi:10.1109/cvpr52729.2023.00718 2023

  16. [16]

    Xu, Medicalgpt: Training medical gpt model, https://github.com/shibing624/MedicalGPT, 2023

    M. Xu, Medicalgpt: Training medical gpt model, https://github.com/shibing624/MedicalGPT, 2023

    work page 2023

  17. [17]

    Radford, K

    A. Radford, K. Narasimhan, T. Salimans, I. Sutskever, et al., Improving language understanding by generative pre-training (2018). xxii

    work page 2018

  18. [18]

    Devlin, M.-W

    J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, Bert: Pre-training of deep bidirectional transformers for language understanding, in: Proceed- ings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), 2019, pp. 4171–4186

    work page 2019

  19. [19]

    LLaMA: Open and Efficient Foundation Language Models

    H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozi `ere, N. Goyal, E. Hambro, F. Azhar, et al., Llama: Open and efficient foundation language models, arXiv preprint arXiv:2302.13971 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  20. [20]

    The Llama 3 Herd of Models

    A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Yang, A. Fan, et al., The llama 3 herd of models, arXiv preprint arXiv:2407.21783 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  21. [21]

    Nicolson, J

    A. Nicolson, J. Dowling, B. Koopman, Improving chest x-ray report generation by leveraging warm starting, Artificial intelligence in medicine 144 (2023) 102633. doi:10.1016/j.artmed.2023.102633

    work page doi:10.1016/j.artmed.2023.102633 2023

  22. [22]

    Y . Tao, L. Ma, J. Yu, H. Zhang, Memory-based cross-modal semantic alignment network for radiology report generation, IEEE Journal of Biomedical and Health Informatics (2024). doi:10.1109/JBHI.2024.3393018

    work page doi:10.1109/jbhi.2024.3393018 2024

  23. [23]

    J. P. Cohen, J. D. Viviano, P. Bertin, P. Morrison, P. Torabian, M. Guarrera, M. P. Lungren, A. Chaudhari, R. Brooks, M. Hashir, et al., Torchxrayvision: A library of chest x-ray datasets and models, in: International Conference on Medical Imaging with Deep Learning, PMLR, 2022, pp. 231–249

    work page 2022

  24. [24]

    H. T. N. Nguyen, D. Nie, T. Badamdorj, Y . Liu, Y . Zhu, J. Truong, L. Cheng, Automated generation of accurate & fluent medical x-ray reports, 2021. arXiv:2108.12126

    work page arXiv 2021

  25. [25]

    Dettmers, A

    T. Dettmers, A. Pagnoni, A. Holtzman, L. Zettlemoyer, Qlora: E fficient finetuning of quantized llms, Advances in Neural Information Processing Systems 36 (2024)

    work page 2024

  26. [26]

    Demner-Fushman, M

    D. Demner-Fushman, M. D. Kohli, M. B. Rosenman, S. E. Shooshan, L. Rodriguez, S. Antani, G. R. Thoma, C. J. McDonald, Preparing a collection of radiology examinations for distribution and retrieval, Journal of the American Medical Informatics Association 23 (2016) 304–310. doi:10.1093/jamia/ocv080

    work page doi:10.1093/jamia/ocv080 2016

  27. [27]

    Vaswani, Attention is all you need, Advances in Neural Information Processing Systems (2017)

    A. Vaswani, Attention is all you need, Advances in Neural Information Processing Systems (2017)

    work page 2017

  28. [28]

    Y . Li, X. Liang, Z. Hu, E. P. Xing, Hybrid retrieval-generation reinforced agent for medical image report generation, Advances in neural information processing systems 31 (2018)

    work page 2018

  29. [29]

    B. Jing, Z. Wang, E. Xing, Show, describe and conclude: On exploiting the structure information of chest x-ray reports, arXiv preprint arXiv:2004.12274 (2020)

    work page arXiv 2004

  30. [30]

    Zhang, X

    Y . Zhang, X. Wang, Z. Xu, Q. Yu, A. Yuille, D. Xu, When radiology report generation meets knowledge graph, in: Proceedings of the AAAI conference on artificial intelligence, volume 34, 2020, pp. 12910–12917. doi:10.1609/aaai.v34i07.6989

    work page doi:10.1609/aaai.v34i07.6989 2020

  31. [31]

    Z. Chen, Y . Song, T.-H. Chang, X. Wan, Generating radiology reports via memory-driven transformer, arXiv preprint arXiv:2010.16056 (2020)

    work page arXiv 2010

  32. [32]

    F. Liu, X. Wu, S. Ge, W. Fan, Y . Zou, Exploring and distilling posterior and prior knowledge for radiology report generation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 13753–13762. doi:10.1109/CVPR46437.2021.01354

    work page doi:10.1109/cvpr46437.2021.01354 2021

  33. [33]

    J. Li, S. Li, Y . Hu, H. Tao, A self-guided framework for radiology report generation, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2022, pp. 588–598. doi:10.1007/978-3-031-16452-1_56

    work page doi:10.1007/978-3-031-16452-1_56 2022

  34. [34]

    F. Liu, S. Ge, Y . Zou, X. Wu, Competence-based multimodal curriculum learning for medical report generation, arXiv preprint arXiv:2206.14579 (2022)

    work page arXiv 2022

  35. [35]

    Z. Chen, Y . Shen, Y . Song, X. Wan, Cross-modal memory networks for radiology report generation, 2022.arXiv:2204.13258

    work page arXiv 2022

  36. [36]

    J. You, D. Li, M. Okumura, K. Suzuki, Jpg-jointly learn to align: Automated disease prediction and radiology report generation, in: Proceedings of the 29th international conference on computational linguistics, 2022, pp. 5989–6001

    work page 2022

  37. [37]

    B. Yan, M. Pei, M. Zhao, C. Shan, Z. Tian, Prior guided transformer for accurate radiology reports generation, IEEE Journal of Biomedical and Health Informatics 26 (2022) 5631–5640. doi:10.1109/JBHI.2022.3197162

    work page doi:10.1109/jbhi.2022.3197162 2022

  38. [38]

    L. Wang, M. Ning, D. Lu, D. Wei, Y . Zheng, J. Chen, An inclusive task-aware framework for radiology report generation, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2022, pp. 568–577. doi: 10.1007/978-3-031- xxiii 16452-1_54

    work page doi:10.1007/978-3-031- 2022

  39. [39]

    M. Li, B. Lin, Z. Chen, H. Lin, X. Liang, X. Chang, Dynamic graph enhanced contrastive learning for chest x-ray report generation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 3334–3343. doi: 10.1109/CVPR52729. 2023.00325

    work page doi:10.1109/cvpr52729 2023

  40. [40]

    H. Qin, Y . Song, Reinforced cross-modal alignment for radiology report generation, in: Findings of the Association for Computational Linguistics: ACL 2022, 2022, pp. 448–458. doi:10.18653/v1/2022.findings-acl.38

    work page doi:10.18653/v1/2022.findings-acl.38 2022

  41. [41]

    Najdenkoska, X

    I. Najdenkoska, X. Zhen, M. Worring, L. Shao, Variational topic inference for chest x-ray report generation, in: Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part III 24, Springer, 2021, pp. 625–635. doi:10.1007/978-3-030-87199-4_59

    work page doi:10.1007/978-3-030-87199-4_59 2021

  42. [42]

    F. Zeng, Z. Lyu, Q. Li, X. Li, Enhancing llms for impression generation in radiology reports through a multi-agent system, arXiv preprint arXiv:2412.06828 (2024). doi:10.48550/arXiv.2412.06828

    work page doi:10.48550/arxiv.2412.06828 2024

  43. [43]

    Y . Li, B. Yang, X. Cheng, Z. Zhu, H. Li, Y . Zou, Unify, align and refine: Multi-level semantic alignment for radiology report generation, in: Proceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 2863–2874. doi:10.48550/arXiv.2303.15932

    work page doi:10.48550/arxiv.2303.15932 2023

  44. [44]

    C. Yin, B. Qian, J. Wei, X. Li, X. Zhang, Y . Li, Q. Zheng, Automatic generation of medical imaging diagnostic report with hierarchical recurrent neural network, in: 2019 IEEE international conference on data mining (ICDM), IEEE, 2019, pp. 728–737. doi: 10.1109/ICDM. 2019.00083

    work page doi:10.1109/icdm 2019

  45. [45]

    Islam, A

    S. Islam, A. Dash, A. Seum, A. H. Raj, T. Hossain, F. M. Shah, Exploring video captioning techniques: A comprehensive survey on deep learning methods, SN Computer Science 2 (2021) 1–28. doi: 10.1007/s42979-021-00487-x

    work page doi:10.1007/s42979-021-00487-x 2021

  46. [46]

    K. R. Suresh, A. Jarapala, P. Sudeep, Image captioning encoder–decoder models using cnn-rnn architectures: A comparative study, Circuits, Systems, and Signal Processing 41 (2022) 5719–5742. doi:10.1007/s00034-022-02050-2

    work page doi:10.1007/s00034-022-02050-2 2022

  47. [47]

    Zhang, P

    K. Zhang, P. Li, J. Wang, A review of deep learning-based remote sensing image caption: Methods, models, comparisons and future directions, Remote Sensing 16 (2024) 4113. doi: 10.3390/rs16214113

    work page doi:10.3390/rs16214113 2024

  48. [48]

    G. Xu, S. Niu, M. Tan, Y . Luo, Q. Du, Q. Wu, Towards accurate text-based image captioning with content diversity exploration, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 12637–12646. doi: 10.1109/CVPR46437. 2021.01245

    work page doi:10.1109/cvpr46437 2021

  49. [49]

    L. Chen, Z. Jiang, J. Xiao, W. Liu, Human-like controllable image captioning with verb-specific semantic roles, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 16846–16856. doi:10.1109/CVPR46437.2021.01657

    work page doi:10.1109/cvpr46437.2021.01657 2021

  50. [50]

    A. Tran, A. Mathews, L. Xie, Transform and tell: Entity-aware news image captioning, in: Proceedings of the IEEE /CVF conference on computer vision and pattern recognition, 2020, pp. 13035–13045. doi:10.1109/CVPR42600.2020.01305

    work page doi:10.1109/cvpr42600.2020.01305 2020

  51. [51]

    Jiang, C

    Y . Jiang, C. Chen, D. Nguyen, B. M. Mervak, C. Tan, Gpt-4v cannot generate radiology reports yet, 2024. arXiv:2407.12176

    work page arXiv 2024

  52. [52]

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chen, Lora: Low-rank adaptation of large language models, 2021. arXiv:2106.09685

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  53. [53]

    B leu: a Method for Automatic Evaluation of Machine Translation

    K. Papineni, S. Roukos, T. Ward, W.-J. Zhu, Bleu: a method for automatic evaluation of machine translation, in: Proceedings of the 40th annual meeting of the Association for Computational Linguistics, 2002, pp. 311–318. doi:10.3115/1073083.1073135

    work page doi:10.3115/1073083.1073135 2002

  54. [54]

    Lin, Rouge: A package for automatic evaluation of summaries, in: Text summarization branches out, 2004, pp

    C.-Y . Lin, Rouge: A package for automatic evaluation of summaries, in: Text summarization branches out, 2004, pp. 74–81

    work page 2004

  55. [55]

    Denkowski, A

    M. Denkowski, A. Lavie, Meteor 1.3: Automatic metric for reliable optimization and evaluation of machine translation systems, in: Proceedings of the sixth workshop on statistical machine translation, 2011, pp. 85–91

    work page 2011

  56. [56]

    Banerjee, A

    S. Banerjee, A. Lavie, Meteor: An automatic metric for mt evaluation with improved correlation with human judgments, in: Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, 2005, pp. 65–72

    work page 2005

  57. [57]

    Nguyen, C

    D. Nguyen, C. Chen, H. He, C. Tan, Pragmatic radiology report generation, in: Machine Learning for Health (ML4H), PMLR, 2023, pp. 385–402

    work page 2023

  58. [58]

    A. E. Johnson, T. J. Pollard, S. J. Berkowitz, N. R. Greenbaum, M. P. Lungren, C.-y. Deng, R. G. Mark, S. Horng, Mimic-cxr, a de-identified publicly available database of chest radiographs with free-text reports, Scientific data 6 (2019) 317. doi:10.1038/s41597-019-0322-0 . xxiv

    work page doi:10.1038/s41597-019-0322-0 2019

  59. [59]

    Kim, C.-k

    S. Kim, C.-k. Lee, S.-s. Kim, Large language models: a guide for radiologists, Korean Journal of Radiology 25 (2024) 126. doi: 10.3348/ kjr.2023.0997

    work page arXiv 2024

  60. [60]

    Dikici, M

    E. Dikici, M. Bigelow, L. M. Prevedello, R. D. White, B. S. Erdal, Integrating ai into radiology workflow: levels of research, production, and feedback maturity, Journal of Medical Imaging 7 (2020) 016502–016502. doi:10.1117/1.JMI.7.1.016502

    work page doi:10.1117/1.jmi.7.1.016502 2020

  61. [61]

    L. Guo, L. Xia, Q. Zheng, B. Zheng, S. Jaeger, M. L. Giger, J. Fuhrman, H. Li, F. Y . Lure, H. Li, et al., Can ai generate diagnostic reports for radiologist approval on cxr images? a multi-reader and multi-case observer performance study, Journal of X-Ray Science and Technology (2024) 1–16. doi:10.3233/XST-240051

    work page doi:10.3233/xst-240051 2024

  62. [62]

    Watanabe, S

    A. Watanabe, S. Ketabi, K. Namdar, F. Khalvati, Improving disease classification performance and explainability of deep learning models in radiology with heatmap generators, Frontiers in radiology 2 (2022) 991683. doi: 10.3389/fradi.2022.991683

    work page doi:10.3389/fradi.2022.991683 2022

  63. [63]

    Granata, F

    V . Granata, F. De Muzio, C. Cutolo, F. Dell’Aversana, F. Grassi, R. Grassi, I. Simonetti, F. Bruno, P. Palumbo, G. Chiti, et al., Structured reporting in radiological settings: pitfalls and perspectives, Journal of Personalized Medicine 12 (2022) 1344. doi: 10.3390/jpm12081344

    work page doi:10.3390/jpm12081344 2022

  64. [64]

    Ahluwalia, M

    M. Ahluwalia, M. Abdalla, J. Sanayei, L. Seyyed-Kalantari, M. Hussain, A. Ali, B. Fine, The subgroup imperative: chest radiograph classifier generalization gaps in patient, setting, and pathology subgroups, Radiology: Artificial Intelligence 5 (2023) e220270. doi: 10.1148/ryai. 220270. xxv

    work page doi:10.1148/ryai 2023