pith. sign in

arxiv: 2605.24371 · v1 · pith:6XC7I6KLnew · submitted 2026-05-23 · 💻 cs.CV · cs.CL

SliceWorld: A Predictive and Controllable World-State Model for CT Report Generation

Yuanhe Tian , Yan Song This is my paper

Pith reviewed 2026-06-30 13:53 UTC · model grok-4.3

classification 💻 cs.CV cs.CL
keywords CT report generationworld-state modellatent factorsslice sequence predictioncontrollable generationmedical image analysisfactor intervention
0
0 comments X

The pith

SliceWorld encodes prefix CT slices into separable latent states of anatomy, lesion, and uncertainty to support future-slice prediction, factor intervention, and report generation.

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

The paper presents SliceWorld as a world-state model that treats a CT volume as an ordered sequence of axial slices along the z-axis. It first encodes the initial slices into factor-aware latent states and projects them into world tokens. These tokens then drive multi-step prediction of later slices, allow controlled changes to lesion factors, and feed an LLM for report text. The model is pretrained with predictive, factor-aware, and counterfactual losses on slice sequences before fine-tuning on paired reports. A reader would care if the approach yields reports that are both more accurate and responsive to specific clinical adjustments than direct image-to-text baselines.

Core claim

SliceWorld encodes prefix CT evidence into factor-aware latent states containing anatomy, lesion, and uncertainty components, projects these states into world tokens, and uses the tokens for multi-step future-slice feature prediction, lesion-factor intervention, and LLM-based report generation. The model is first pretrained on CT slice sequences with predictive, factor-aware, and counterfactual objectives and then fine-tuned on paired CT-report data, producing improved natural language generation metrics and clinically oriented scores on M3D-Cap and CT-RATE together with measurable multi-horizon prediction, factor alignment, reduced-slice robustness, and selective lesion-sensitive report mod

What carries the argument

Factor-aware latent states (anatomy, lesion, uncertainty components) projected into world tokens that enable prediction, intervention, and generation.

If this is right

  • Improved natural language generation metrics and clinically oriented automatic evaluation on M3D-Cap and CT-RATE.
  • Multi-horizon future-slice prediction from prefix evidence.
  • Measurable alignment between latent factors and clinical variables.
  • Maintained report quality when the number of input slices is reduced.
  • Selective modulation of reports when lesion-related factors are changed.

Where Pith is reading between the lines

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

  • The same latent-state representation might support 'what-if' queries by radiologists who want to see how a report would change if a lesion were larger or smaller.
  • The sequence modeling approach could transfer to other slice-based modalities such as MRI where the ordering axis is spatial rather than temporal.
  • Pretraining on unlabeled slice sequences may lower the amount of paired report data needed for acceptable performance.
  • If the uncertainty component can be read out separately, it might offer a built-in signal for flagging low-confidence regions in the generated report.

Load-bearing premise

Prefix CT evidence can be encoded into cleanly separable factor-aware latent states whose components do not bleed into each other during projection and use.

What would settle it

Intervening on the lesion component of the latent state produces no corresponding change in the predicted future-slice features or in the generated report text, as measured by the same clinical metrics used in the paper.

Figures

Figures reproduced from arXiv: 2605.24371 by Yan Song, Yuanhe Tian.

Figure 1
Figure 1. Figure 1: Overview of SliceWorld. Ordered axial CT slices are encoded into prefix states [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: State-level diagnostics of SliceWorld with Qwen3-1.7B. The left two panels show multi [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: CT-RATE sensitivity to the future-prediction horizon [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative counterfactual lesion intervention examples. Panel (a) shows a lesion-positive [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Additional qualitative examples for factor-aware state behavior. The lesion-positive example [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
read the original abstract

CT report generation (CTRG) requires models to summarize three-dimensional anatomical context and pathological findings from hundreds of axial slices. Existing methods typically learn a direct image-to-text mapping, providing limited mechanisms for modeling how CT evidence evolves across slices or how reports respond to controlled changes in latent lesion-related factors. We propose SliceWorld, a CT-specific world-state framework that treats an axial CT scan as an ordered sequence along the z-axis. SliceWorld encodes prefix CT evidence into factor-aware latent states containing anatomy, lesion, and uncertainty components, and projects these states into world tokens used for multi-step future-slice feature prediction, lesion-factor intervention, and LLM-based report generation. The model is first pretrained on CT slice sequences with predictive, factor-aware, and counterfactual objectives, and is then fine-tuned on paired CT-report data. Experiments on M3D-Cap and CT-RATE show that SliceWorld improves natural language generation metrics and clinically oriented automatic evaluation. Further analyses demonstrate multi-horizon future-slice prediction, measurable factor alignment, reduced-slice robustness, and selective lesion-sensitive report modulation.

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 / 0 minor

Summary. The paper introduces SliceWorld, a world-state model for CT report generation (CTRG) that treats axial CT volumes as z-axis sequences. Prefix slices are encoded into factor-aware latent states comprising anatomy, lesion, and uncertainty components; these states are projected to world tokens supporting multi-horizon future-slice prediction, lesion-factor intervention, and LLM-based report generation. The model is pretrained with predictive, factor-aware, and counterfactual objectives before fine-tuning on paired CT-report data. Experiments on M3D-Cap and CT-RATE report gains in NLG metrics and clinical automatic evaluations, plus supporting analyses for multi-horizon prediction, factor alignment, reduced-slice robustness, and selective lesion-sensitive modulation.

Significance. If the claimed factor separability and controllability hold, the work would advance beyond direct image-to-text CTRG by enabling explicit prediction, intervention, and robustness testing in a medical imaging setting. The combination of world-state modeling with LLM report generation and the reported multi-horizon and modulation capabilities would be a substantive contribution to controllable medical report generation.

major comments (2)
  1. [Abstract] Abstract: The central construction encodes prefix evidence into 'factor-aware latent states containing anatomy, lesion, and uncertainty components' whose projections are asserted to support lesion-factor intervention and selective report modulation. No mechanism (orthogonality penalty, mutual-information term, architectural bottleneck, or explicit loss term) is described that would enforce independence among the three components. This separability is load-bearing for the 'measurable factor alignment' and 'lesion-sensitive report modulation' results; without it, those behaviors could arise from spurious correlations rather than controllable, disentangled states.
  2. [Abstract] Abstract (pretraining description): The model is 'first pretrained on CT slice sequences with predictive, factor-aware, and counterfactual objectives.' The factor-aware objective is listed but not shown to contain any term that penalizes correlation between anatomy/lesion/uncertainty factors. If the factor-aware loss only supervises individual factors without an independence constraint, the downstream interventional claims rest on an unverified assumption.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and for identifying the need for explicit mechanisms supporting factor separability. We address each comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central construction encodes prefix evidence into 'factor-aware latent states containing anatomy, lesion, and uncertainty components' whose projections are asserted to support lesion-factor intervention and selective report modulation. No mechanism (orthogonality penalty, mutual-information term, architectural bottleneck, or explicit loss term) is described that would enforce independence among the three components. This separability is load-bearing for the 'measurable factor alignment' and 'lesion-sensitive report modulation' results; without it, those behaviors could arise from spurious correlations rather than controllable, disentangled states.

    Authors: We agree that the current abstract and methods description do not specify an explicit independence constraint. The components are produced by separate projection heads and receive distinct supervision signals under the factor-aware objective, while the counterfactual objective performs targeted interventions on the lesion component. Nevertheless, the absence of a decorrelation term leaves the separability claim open to the interpretation raised. We will revise the manuscript to add an orthogonality penalty to the factor-aware loss, include the updated formulation in the methods, and report an ablation measuring the effect on factor alignment and modulation metrics. revision: yes

  2. Referee: [Abstract] Abstract (pretraining description): The model is 'first pretrained on CT slice sequences with predictive, factor-aware, and counterfactual objectives.' The factor-aware objective is listed but not shown to contain any term that penalizes correlation between anatomy/lesion/uncertainty factors. If the factor-aware loss only supervises individual factors without an independence constraint, the downstream interventional claims rest on an unverified assumption.

    Authors: The referee correctly notes that the factor-aware objective, as currently described, applies separate supervision to each component without an explicit term that penalizes inter-factor correlation. We will therefore revise the pretraining section to incorporate a correlation penalty (e.g., an orthogonality regularizer) into the factor-aware loss, update the abstract to reflect this change, and add the corresponding experimental results. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation self-contained against external benchmarks

full rationale

The abstract and description present a pretraining stage using predictive/factor-aware/counterfactual objectives followed by fine-tuning on paired CT-report data, with evaluation on independent datasets (M3D-Cap, CT-RATE). No equations, self-citations, or fitted-parameter renamings are supplied that would reduce any claimed prediction or factor projection to an input by construction. The separability premise is an assumption, not a definitional loop; downstream behaviors are asserted to be measured experimentally rather than forced by the training objective itself. This meets the default expectation of no significant circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no equations, training details, or modeling assumptions are provided, so the ledger cannot be populated beyond noting the implicit assumption of separable latent factors.

pith-pipeline@v0.9.1-grok · 5712 in / 1152 out tokens · 36355 ms · 2026-06-30T13:53:57.166776+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

43 extracted references · 14 canonical work pages · 8 internal anchors

  1. [1]

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

    2024

  2. [2]

    CT2Rep: Automated radiology report generation for 3d medical imaging

    Ibrahim Ethem Hamamci, Sezgin Er, and Bjoern Menze. CT2Rep: Automated radiology report generation for 3d medical imaging. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 476–486, 2024

    2024

  3. [3]

    Mvketr: chest ct report generation with multi-view perception and knowledge enhance- ment.IEEE Journal of Biomedical and Health Informatics, 2025

    Xiwei Deng, Xianchun He, Jianfeng Bao, Yudan Zhou, Shuhui Cai, Congbo Cai, and Zhong Chen. Mvketr: chest ct report generation with multi-view perception and knowledge enhance- ment.IEEE Journal of Biomedical and Health Informatics, 2025

    2025

  4. [4]

    Automatic radiology report generation with deep learning: a comprehensive review of methods and advances.Artificial Intelligence Review, 58(11):344, 2025

    Yilin Li, Chao Kong, Guosheng Zhao, and Zijian Zhao. Automatic radiology report generation with deep learning: a comprehensive review of methods and advances.Artificial Intelligence Review, 58(11):344, 2025

    2025

  5. [5]

    A survey of deep-learning-based radiology report generation using multimodal inputs.Medical Image Analysis, 103:103627, 2025

    Xinyi Wang, Grazziela Figueredo, Ruizhe Li, Wei Emma Zhang, Weitong Chen, and Xin Chen. A survey of deep-learning-based radiology report generation using multimodal inputs.Medical Image Analysis, 103:103627, 2025

    2025

  6. [6]

    Ct-graph: Hierarchical graph attention network for anatomy-guided ct report generation

    Hamza Kalisch, Fabian Hörst, Jens Kleesiek, Ken Herrmann, and Constantin Seibold. Ct-graph: Hierarchical graph attention network for anatomy-guided ct report generation. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 6775–6784, 2025

    2025

  7. [7]

    Large language model with region-guided referring and grounding for ct report generation.IEEE Transactions on Medical Imaging, 2025

    Zhixuan Chen, Yequan Bie, Haibo Jin, and Hao Chen. Large language model with region-guided referring and grounding for ct report generation.IEEE Transactions on Medical Imaging, 2025

    2025

  8. [8]

    Recurrent visual feature extraction and stereo attentions for ct report generation.arXiv preprint arXiv:2506.19665, 2025

    Yuanhe Tian, Lei Mao, and Yan Song. Recurrent visual feature extraction and stereo attentions for ct report generation.arXiv preprint arXiv:2506.19665, 2025

    work page arXiv 2025

  9. [9]

    Feature decomposition via shared low-rank matrix recovery for ct report generation.IEEE Transactions on Medical Imaging, 2025

    Yuanhe Tian and Yan Song. Feature decomposition via shared low-rank matrix recovery for ct report generation.IEEE Transactions on Medical Imaging, 2025

    2025

  10. [10]

    World Models

    David Ha and Jürgen Schmidhuber. World models.arXiv preprint arXiv:1803.10122, 2(3):440, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  11. [11]

    Mastering Diverse Domains through World Models

    Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Timothy Lillicrap. Mastering diverse domains through world models.arXiv preprint arXiv:2301.04104, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  12. [12]

    Deeplesion: automated mining of large-scale lesion annotations and universal lesion detection with deep learning.Journal of medical imaging, 5(3):036501–036501, 2018

    Ke Yan, Xiaosong Wang, Le Lu, and Ronald M Summers. Deeplesion: automated mining of large-scale lesion annotations and universal lesion detection with deep learning.Journal of medical imaging, 5(3):036501–036501, 2018

    2018

  13. [13]

    M3D: Advancing 3d medical image analysis with multi-modal large language models.arXiv preprint arXiv:2404.00578, 2024

    Fan Bai, Yuxin Du, Tiejun Huang, Max Q-H Meng, and Bo Zhao. M3D: Advancing 3d medical image analysis with multi-modal large language models.arXiv preprint arXiv:2404.00578, 2024

    work page arXiv 2024

  14. [14]

    Generalist foundation models from a multimodal dataset for 3d computed tomography.Nature Biomedical Engineering, pages 1–19, 2026

    Ibrahim Ethem Hamamci, Sezgin Er, Chenyu Wang, Furkan Almas, Ayse Gulnihan Simsek, Sevval Nil Esirgun, Irem Dogan, Omer Faruk Durugol, Benjamin Hou, Suprosanna Shit, et al. Generalist foundation models from a multimodal dataset for 3d computed tomography.Nature Biomedical Engineering, pages 1–19, 2026. 10

    2026

  15. [15]

    PMC-VQA: Visual Instruction Tuning for Medical Visual Question Answering

    Xiaoman Zhang, Chaoyi Wu, Ziheng Zhao, Weixiong Lin, Ya Zhang, Yanfeng Wang, and Weidi Xie. Pmc-vqa: Visual instruction tuning for medical visual question answering.arXiv preprint arXiv:2305.10415, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  16. [16]

    Med-flamingo: a multimodal medical few-shot learner

    Michael Moor, Qian Huang, Shirley Wu, Michihiro Yasunaga, Yash Dalmia, Jure Leskovec, Cyril Zakka, Eduardo Pontes Reis, and Pranav Rajpurkar. Med-flamingo: a multimodal medical few-shot learner. InMachine Learning for Health (ML4H), pages 353–367, 2023

    2023

  17. [17]

    Towards generalist foundation model for radiology.arXiv preprint arXiv:2308.02463, 2023

    Chaoyi Wu, Xiaoman Zhang, Ya Zhang, Yanfeng Wang, and Weidi Xie. Towards generalist foundation model for radiology.arXiv preprint arXiv:2308.02463, 2023

    work page arXiv 2023

  18. [18]

    Developing generalist foundation models from a multimodal dataset for 3d computed tomography

    Ibrahim Ethem Hamamci, Sezgin Er, Furkan Almas, Ayse Gulnihan Simsek, Sevval Nil Esirgun, Irem Dogan, Muhammed Furkan Dasdelen, Omer Faruk Durugol, Bastian Wittmann, Tamaz Amiranashvili, et al. Developing generalist foundation models from a multimodal dataset for 3d computed tomography. 2024

    2024

  19. [19]

    DINOv2: Learning Robust Visual Features without Supervision

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision.arXiv preprint arXiv:2304.07193, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  20. [20]

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  21. [21]

    Qwen3 Technical Report

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  22. [22]

    Ministral 3

    Alexander H Liu, Kartik Khandelwal, Sandeep Subramanian, Victor Jouault, Abhinav Rastogi, Adrien Sadé, Alan Jeffares, Albert Jiang, Alexandre Cahill, Alexandre Gavaudan, et al. Ministral 3.arXiv preprint arXiv:2601.08584, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  23. [23]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models. volume 1, page 3, 2022

    2022

  24. [24]

    Generating radiology reports via memory-driven transformer

    Zhihong Chen, Yan Song, Tsung-Hui Chang, and Xiang Wan. Generating radiology reports via memory-driven transformer. InProceedings of the 2020 conference on empirical methods in natural language processing (EMNLP), pages 1439–1449, 2020

    2020

  25. [25]

    Cross-modal memory networks for radiology report generation

    Zhihong Chen, Yaling Shen, Yan Song, and Xiang Wan. Cross-modal memory networks for radiology report generation. InProceedings 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), pages 5904–5914, 2021

    2021

  26. [26]

    A systematic review of deep learning-based research on radiology report generation.arXiv preprint arXiv:2311.14199, 2023

    Chang Liu, Yuanhe Tian, and Yan Song. A systematic review of deep learning-based research on radiology report generation.arXiv preprint arXiv:2311.14199, 2023

    work page arXiv 2023

  27. [27]

    Bootstrapping large language models for radiology report generation

    Chang Liu, Yuanhe Tian, Weidong Chen, Yan Song, and Yongdong Zhang. Bootstrapping large language models for radiology report generation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 18635–18643, 2024

    2024

  28. [28]

    Diffusion networks with task-specific noise control for radiology report generation

    Yuanhe Tian, Fei Xia, and Yan Song. Diffusion networks with task-specific noise control for radiology report generation. InProceedings of the 32nd ACM International Conference on Multimedia, pages 1771–1780, 2024

    2024

  29. [29]

    Extractive radiology reporting with memory-based cross-modal representations.IEEE Transactions on Medical Imaging, 2025

    Yuanhe Tian, Zexuan Yan, Nenan Lyu, and Yan Song. Extractive radiology reporting with memory-based cross-modal representations.IEEE Transactions on Medical Imaging, 2025

    2025

  30. [30]

    Merlin: a computed tomography vision–language foundation model and dataset.Nature, pages 1–11, 2026

    Louis Blankemeier, Ashwin Kumar, Joseph Paul Cohen, Jiaming Liu, Longchao Liu, Dave Van Veen, Syed Jamal Safdar Gardezi, Hongkun Yu, Magdalini Paschali, Zhihong Chen, et al. Merlin: a computed tomography vision–language foundation model and dataset.Nature, pages 1–11, 2026. 11

    2026

  31. [31]

    Scaling self-supervised and cross-modal pretraining for volumetric ct transformers

    Cris Claessens, Christiaan Viviers, Giacomo D’Amicantonio, Egor Bondarev, and Fons van der Sommen. Scaling self-supervised and cross-modal pretraining for volumetric ct transformers. arXiv preprint arXiv:2511.17209, 2025

    work page arXiv 2025

  32. [32]

    R2gen-mamba: A selective state space model for radiology report generation

    Yongheng Sun, Yueh Z Lee, Genevieve A Woodard, Hongtu Zhu, Chunfeng Lian, and Mingxia Liu. R2gen-mamba: A selective state space model for radiology report generation. pages 1–4, 2025

    2025

  33. [33]

    Making the most of text semantics to improve biomedical vision–language processing

    Benedikt Boecking, Naoto Usuyama, Shruthi Bannur, Daniel C Castro, Anton Schwaighofer, Stephanie Hyland, Maria Wetscherek, Tristan Naumann, Aditya Nori, Javier Alvarez-Valle, et al. Making the most of text semantics to improve biomedical vision–language processing. In European conference on computer vision, pages 1–21, 2022

    2022

  34. [34]

    Learning semantic relationship among instances for image-text matching

    Zheren Fu, Zhendong Mao, Yan Song, and Yongdong Zhang. Learning semantic relationship among instances for image-text matching. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15159–15168, 2023

    2023

  35. [35]

    Improving image captioning via predicting structured concepts

    Ting Wang, Weidong Chen, Yuanhe Tian, Yan Song, and Zhendong Mao. Improving image captioning via predicting structured concepts. InProceedings of the 2023 conference on empirical methods in natural language processing, pages 360–370, 2023

    2023

  36. [36]

    Chexworld: Exploring image world modeling for radiograph representation learning

    Yang Yue, Yulin Wang, Chenxin Tao, Pan Liu, Shiji Song, and Gao Huang. Chexworld: Exploring image world modeling for radiograph representation learning. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 20778–20788, 2025

    2025

  37. [37]

    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

    2002

  38. [38]

    ROUGE: A Package for Automatic Evaluation of Summaries

    Chin-Yew Lin. ROUGE: A Package for Automatic Evaluation of Summaries. InText Summa- rization Branches Out, pages 74–81, Barcelona, Spain, July 2004

    2004

  39. [39]

    Meteor: An automatic metric for mt evaluation with improved correlation with human judgments

    Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evaluation 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

    2005

  40. [40]

    BERTScore: Evaluating Text Generation with BERT

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. Bertscore: Evaluating text generation with bert.arXiv preprint arXiv:1904.09675, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1904

  41. [41]

    Ratescore: A metric for radiology report generation

    Weike Zhao, Chaoyi Wu, Xiaoman Zhang, Ya Zhang, Yanfeng Wang, and Weidi Xie. Ratescore: A metric for radiology report generation. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 15004–15019, 2024

    2024

  42. [42]

    Green: Generative radiology report evaluation and error notation.arXiv preprint arXiv:2405.03595, 2024

    Sophie Ostmeier, Justin Xu, Zhihong Chen, Maya Varma, Louis Blankemeier, Christian Bluethgen, Arne Edward Michalson, Michael Moseley, Curtis Langlotz, Akshay S Chaud- hari, et al. Green: Generative radiology report evaluation and error notation.arXiv preprint arXiv:2405.03595, 2024. A CT-RATE HU-to-Image Preprocessing CT-RATE provides 3D NIfTI volumes who...

    work page arXiv 2024

  43. [43]

    The decoder backbones are Qwen3-1.7B, Qwen3-4B, and Ministral-3-3B

    The state-factor dimensions are danat = 256 , dlesion = 192 , and dunc = 64 , and the world- token dimension is 512. The decoder backbones are Qwen3-1.7B, Qwen3-4B, and Ministral-3-3B. Qwen3-1.7B has 28 Transformer layers with hidden size 2,048, 16 attention heads, and 8 key-value heads; Qwen3-4B has 36 layers with hidden size 2,560, 32 attention heads, a...

    1935