Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Affective-ROPTester: Capability and Bias Analysis of LLMs in Predicting Retinopathy of Prematurity

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper builds a 993-record Chinese benchmark for predicting retinopathy of prematurity (ROP) risk in preterm infants and reports that large language models over-predict medium and high risk, a bias that positive affective prompts…

desk verdict Useful new Chinese ROP risk benchmark and a credible bias finding, but the affective-framing mitigation claim is confounded by prompt content beyond valence. read the letter →

arxiv 2507.05816 v1 pith:UQ3KMWTK submitted 2025-07-08 cs.AI cs.CEcs.CL

Shuai Zhao , Yulin Zhang , Luwei Xiao , Xinyi Wu , Yanhao Jia , Zhongliang Guo , Xiaobao Wu , Cong-Duy Nguyen
show 2 more authors
Guoming Zhang Anh Tuan Luu
This is my paper · ORCID
classification cs.AIcs.CEcs.CL
keywords retinopathyofprematuritylargelanguagemodelsaffectivebiaspromptengineeringriskpredictionchain-of-thoughtpromptingin-contextlearningChinesemedicalbenchmark
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

The paper builds what it identifies as the first Chinese benchmark, CROP, of 993 premature-infant admission records labeled low, medium, or high risk for retinopathy of prematurity (ROP), and uses it to ask whether large language models can predict ROP risk from text alone. It finds that with only an instruction prompt the models perform near chance and systematically over-assign medium and high risk, especially to low-risk infants. Feeding the models a hand-built chain of thought based on known risk factors, or a few retrieved examples, raises accuracy substantially; the best model reaches 61.33 percent. The paper's most actionable claim is that positive affective framing in the prompt reduces the over-prediction bias and improves accuracy relative to negative or neutral framing. If true, prompt affect is a real, adjustable lever on diagnostic bias, not just a stylistic footnote.

What carries the argument

The load-bearing mechanism is Affective-ROPTester, an evaluation scaffold with three prompting schemes. The Instruction scheme presents the model with a role definition and a restricted three-way label space, probing intrinsic knowledge and bias. The CoT scheme prepends an eight-step reasoning template built from known ROP risk factors, including gestational age, birth weight, oxygen exposure, multiple pregnancy, and maternal complications, thereby supplying external knowledge. The ICL scheme retrieves five nearest training admissions by BM25 and presents them as labeled demonstrations. Superimposed on the Instruction scheme are three hand-written affective framings, negative, neutral, and positive, and this affective component is what carries the paper's bias-mitigation claim.

What would settle it

Re-run the CROP evaluations with several paraphrases of each affective prompt, matched for length and lexical content; if accuracy and bias shifts do not track valence consistently across paraphrases, the affective-framing claim is not supported.

Watch

Extended reading notes

Core claim

The central claim is that LLMs are biased risk predictors for ROP when left to their own knowledge, consistently pushing low- and medium-risk admissions into higher categories, and that this bias can be partially corrected by supplying structured external knowledge or by embedding positive emotion in the prompt. The paper states that prompts infused with positive affective framing help reduce the bias exhibited by LLMs. Accuracy follows the same pattern: negative prompts hurt, positive prompts help, with DeepSeek-R1 rising from 40.18 percent under negative framing to 49.24 percent under positive framing, and with chain-of-thought prompting the best model reaches 61.33 percent accuracy, which the paper still calls insufficient for clinical use.

Load-bearing premise

The result that positive affect reduces bias rests on the assumption that the three hand-written prompts differ only in emotional valence, but the positive prompt also changes wording, length, expected base rate, and clinical advice, so the measured effect could come from content rather than affect.

Editorial extensions

If this is right

  • If correct, LLM-only screening from admission notes is not yet clinically usable for ROP; the ceiling in this benchmark is about 61 percent accuracy with a chain-of-thought prompt.
  • Structured external knowledge, through risk-factor reasoning or demonstrations, is what makes LLM predictions usable at all, shifting the best open models from roughly 39 percent to 61 percent accuracy.
  • The documented over-prediction of medium and high risk means naive deployment would generate false alarms disproportionate to low-risk infants.
  • Positive affective prompts are a cheap, parameter-free intervention that moves accuracy and bias in the desired direction, so prompt wording should be reported in any clinical LLM benchmark.
  • These effects appear across open, closed, and medical models, indicating the bias is not specific to one model family.

Reading between the lines

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

  • Editorial inference: if positive framing works by shifting the model's prior toward better outcomes, then calibration or class-prior adjustment could achieve the same bias reduction without changing prompt content.
  • Editorial inference: the affective result may extend to other clinical risk tasks where LLMs systematically over-predict severe outcomes, such as sepsis or ICU mortality screening, and should be tested there.
  • Editorial inference: because the three prompts differ in wording and content beyond valence, a lexical-control study is needed to separate affect from instruction content; until then the size of the affect effect is uncertain.
  • Editorial inference: the unbalanced-dataset experiment implies the bias is visible as a drop in accuracy when low-risk cases dominate, so a testable extension is reporting calibration, such as Brier score or expected calibration error, per valence.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces CROP, a Chinese benchmark dataset of 993 admission records for predicting retinopathy of prematurity (ROP) risk, and Affective-ROPTester, an evaluation framework with Instruction, Chain-of-Thought (CoT), and In-Context Learning (ICL) prompting schemes, with additional affective framings (negative, neutral, positive) embedded in the Instruction prompt. Experiments across a wide range of open-source, closed-source, and medical LLMs show that (i) LLMs alone predict ROP risk poorly, (ii) CoT and ICL improve accuracy, and (iii) LLMs exhibit a bias toward predicting medium and high risk, which the paper claims is mitigated by positive affective framing.

Significance. If the findings hold, CROP will be a useful first Chinese text-only benchmark for ROP risk prediction, and the Affective-ROPTester framework provides a reproducible protocol for evaluating LLM capability and bias in clinical risk stratification. The paper's breadth of model coverage and its confusion-matrix-based evidence for a consistent high/medium-risk bias are valuable. However, the central affective-framing conclusion is not yet empirically established because the prompts differ in more than emotional valence, and the CoT scheme's hand-coded rubric may confound the interpretation of the performance gains. These issues limit the current contribution until addressed.

major comments (3)
  1. [Section IV-B and Section V-D] The claim that positive affective framing mitigates bias is confounded: the three affective prompts do not differ only in emotional valence. The positive prompt ('You are a pediatric ophthalmologist who is confident in treatment outcomes and believes that most cases can successfully avoid retinopathy') explicitly asserts a low base rate and instructs the model toward low-risk classifications, while the negative prompt presumably asserts the opposite. Consequently, the observed reduction in high-risk predictions under positive framing (e.g., Figure 4's 76.79% vs. 21.43%) could be a prior shift or instruction-following effect rather than an emotional effect. The paper should include multiple phrasings per valence, a lexical control that varies wording without changing valence, or an explicit manipulation of the stated base rate to disentangle these factors. In addition, Table IV provides no significance tests or confidence intervals, and the effect is inconsistent across models (e.g., Qwen-2.5 gains 4.53 accuracy points from neutral to positive, while o4-mini loses 1.51 points), so the claim that positive framing helps is not yet supported.
  2. [Section IV-C, Scheme 2] The CoT scheme encodes a deterministic clinical rubric: Step 7 maps gestational age <28 weeks or weight <1000 g to high risk, 28-32 weeks and 1000-1500 g to medium risk, and >32 weeks and >1500 g to low risk. The reported performance improvements under CoT therefore largely measure how well this hand-coded rubric matches the dataset labels, rather than the LLM's own reasoning about ROP risk. Since the paper interprets the CoT results as evidence that external knowledge enhances LLM capability, it should explicitly acknowledge this confound, analyze how often the model simply echoes the rubric's thresholds, and ideally compare against a baseline where the same rubric is applied non-LLM. This is load-bearing for the paper's second principal conclusion (Section V-B).
  3. [Section V-D and Table IV] The affective-framing results lack any uncertainty quantification or hypothesis testing. No standard deviations, confidence intervals, or significance tests are reported for Table IV, and the small per-class test sizes (108-112 samples) mean that differences of a few percentage points (e.g., Qwen-2.5: 40.78 neutral vs. 42.29 positive) may be within sampling noise. The paper's conclusion that positive affective framing improves accuracy and mitigates bias should be supported by multiple runs, error bars, and a statistical test across the set of evaluated models. This is essential because the claim is one of the paper's three headline conclusions.
minor comments (5)
  1. [Abstract and Conclusion] The abstract lists 'two principal observations' but then enumerates three (limited efficacy, affective bias, positive framing mitigation); the Conclusion likewise contains three separate claims.
  2. [Equation (1)] Equation (1) states that y_j represents the true label of x, but in the argmax expression y_j should denote a candidate label in Y; this is a notation error.
  3. [Section I, Contributions] The bullet list contains typos: 'instrcution' should be 'instruction' and the sentence 'And also give an instrcution for researchers...' is ungrammatical; this should be corrected.
  4. [Table I and Section III-B] The text says the dataset is 'evenly divided' into three categories, but Table I reports 112 low-, 111 medium-, and 108 high-risk test samples; the imbalance is minor but the wording should be adjusted.
  5. [Figure 4] Figure 4 does not specify which model the confusion matrices are for; this should be stated in the caption or text.

Circularity Check

2 steps flagged · score 4.0 of 10

Partial circularity: the CoT prompt embeds the label-mapping rubric and the 'positive affective framing' condition contains an explicit low-base-rate instruction, so two reported gains are partly built into the inputs; the core ROP bias observation remains independent.

  1. other [Section IV-C, Scheme 2 (Steps 2 and 7); results in Section V-B.]
    "Step 2. Analyze the Impact of Gestational Age and Birth Weight: • Gestational Age less than 28 weeks: High risk, especially if the weight is less than 1000 grams. • Gestational Age 28-32 weeks: Medium risk, with weights between 1000-1500 grams. • Gestational Age 32-34 weeks: Low risk, with weights greater than 1500 grams."

    The CoT input itself contains the classification thresholds that are then reported as LLM performance. Section V-B credits the improvement to the CoT's 'external knowledge,' but the external knowledge includes the decision rule; the measured accuracy is largely the accuracy of the hand-written GA/BW rubric executed by the model, not an emergent predictive capability. The conclusion that external knowledge effectively improves accuracy is therefore partly a restatement of the injected rule. Because the thresholds are cited to prior literature rather than fit to CROP labels, this is partial rather than complete circularity.

  2. self definitional [Section IV-B and Section V-D (Table IV, Figure 4).]
    "For example, in the positive emotional prompts, the instruction 'You are a pediatric ophthalmologist who is confident in treatment outcomes and believes that most cases can successfully avoid retinopathy.' is used to guide LLMs away from overly pessimistic predictions."

    The operationalization of 'positive affective framing' is not affect-only: it asserts a low disease base rate and directs the model away from pessimistic/high-risk outputs. The later conclusion that 'prompts infused with positive affective framing help reduce the bias exhibited by LLMs' measures exactly the behavior the prompt text already requests. Since no valence-matched control separates emotion from this base-rate/content statement, the mitigation result is at least partly self-definitional: the treatment contains the desired outcome.

full rationale

Most of the paper is a benchmarking study rather than a derivation chain. The CROP labels are externally defined from clinical diagnoses, and the bias finding (LLMs overpredict medium and high risk) is read off confusion matrices under the Instruction scheme, independent of the CoT and affective manipulations. No load-bearing self-citation or imported uniqueness theorem appears. The circularity is partial and localized: (i) the CoT performance gain is substantially an evaluation of the hand-written threshold rubric placed inside the prompt, and (ii) the positive-affect condition is operationally confounded with an explicit low-base-rate instruction, so the 'affect mitigates bias' claim is partly built into the input. These issues prevent clean causal attribution of the two improvement effects, but they do not make the basic ROP overprediction bias circular. A score of 4 reflects one central claim partially reducing to its input wording plus a secondary CoT-rubric construction, while the dataset and the core bias observation retain independent content.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claims rest on a hand-crafted CoT with manually chosen thresholds, an assumed relation between the risk categories and final diagnosis labels, and a single set of hand-written affective prompts. No new physical entities are introduced.

free parameters (3)
  • Gestational age thresholds in CoT = <28, 28-32, 32-34 weeks
    Chosen by hand in Scheme 2 to map gestational age to risk categories; not derived from data.
  • Birth weight thresholds in CoT = <1000, 1000-1500, >1500 grams
    Chosen by hand in Scheme 2; discrete mapping used in the CoT risk assignment.
  • Number of ICL demonstrations k = 5
    Set to 5 after reporting an ablation over k=1..5 (Table V); no separate validation split is described.
assumptions (4)
  • domain assumption Known ROP risk factors (birth weight, gestational age, oxygen exposure, multiple gestation, pregnancy complications) are sufficient to predict ROP risk from admission notes.
    Invoked in Section III-A and Scheme 2 to justify the task and the content of the CoT.
  • domain assumption Risk labels derived from the final diagnosis are accurate ground truth, and the admission notes do not already reveal the outcome.
    Used in dataset construction (Section III-B); the paper does not discuss label leakage or misclassification.
  • domain assumption LLMs follow the instructed output format so that generated categories are parsed without error.
    Implicit in all evaluation tables; off-format outputs could distort accuracy and F1 scores.
  • ad hoc to paper The CoT's discrete thresholds (GA and weight cutoffs) are a valid clinical approximation.
    Defined in Scheme 2; presented as known risk factors but not validated against Chinese ROP screening guidelines in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Affective-ROPTester: Capability and Bias Analysis of LLMs in Predicting Retinopathy of Prematurity." pith.science (2026). https://pith.science/paper/UQ3KMWTK

@misc{pith2026250705816,
  author       = {Pith},
  title        = {Pith review of: Affective-ROPTester: Capability and Bias Analysis of LLMs in Predicting Retinopathy of Prematurity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UQ3KMWTK}},
  note         = {Machine review of arXiv:2507.05816}
}
read the original abstract

Despite the remarkable progress of large language models (LLMs) across various domains, their capacity to predict retinopathy of prematurity (ROP) risk remains largely unexplored. To address this gap, we introduce a novel Chinese benchmark dataset, termed CROP, comprising 993 admission records annotated with low, medium, and high-risk labels. To systematically examine the predictive capabilities and affective biases of LLMs in ROP risk stratification, we propose Affective-ROPTester, an automated evaluation framework incorporating three prompting strategies: Instruction-based, Chain-of-Thought (CoT), and In-Context Learning (ICL). The Instruction scheme assesses LLMs' intrinsic knowledge and associated biases, whereas the CoT and ICL schemes leverage external medical knowledge to enhance predictive accuracy. Crucially, we integrate emotional elements at the prompt level to investigate how different affective framings influence the model's ability to predict ROP and its bias patterns. Empirical results derived from the CROP dataset yield two principal observations. First, LLMs demonstrate limited efficacy in ROP risk prediction when operating solely on intrinsic knowledge, yet exhibit marked performance gains when augmented with structured external inputs. Second, affective biases are evident in the model outputs, with a consistent inclination toward overestimating medium- and high-risk cases. Third, compared to negative emotions, positive emotional framing contributes to mitigating predictive bias in model outputs. These findings highlight the critical role of affect-sensitive prompt engineering in enhancing diagnostic reliability and emphasize the utility of Affective-ROPTester as a framework for evaluating and mitigating affective bias in clinical language modeling systems.

Figures

Figures reproduced from arXiv: 2507.05816 by the authors.

Figure 1
Figure 1. The ROP risk prediction pipeline, which includes the construction of the CROP benchmark dataset and integrates the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Sample distribution analysis for our CROP dataset. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The confusion matrix of the model’s predicted risk compared with the true label. The [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The confusion matrix of the predicted risk under [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: A case study with ROP risk prediction reasoning process based on Affective-ROPTester’s Instructional scheme, the [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The predicted results with standard deviation under the Instruction, CoT, and ICL schemes. predicting ROP risk. Additionally, the model performance is optimal when the number of demonstration examples is four in the GPT-4o model, indicating that an excessive number of …

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. HRIPBench: Benchmarking LLMs in Harm Reduction Information Provision to Support People Who Use Drugs

    cs.CL 2025-07 conditional novelty 6.0 of 10

    State-of-the-art LLMs are frequently inaccurate, and sometimes dangerous, when answering harm reduction questions about drug use, even when given retrieved source material.

Reference graph

Works this paper leans on

58 extracted references · 45 canonical work pages · cited by 1 Pith paper

  1. [1]

    The llama 4 herd: The beginning of a new era of natively multimodal ai innovation,

    A. Meta, “The llama 4 herd: The beginning of a new era of natively multimodal ai innovation,” 2025

    work page 2025

  2. [2]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,

    D. Guo, D. Yang, H. Zhang, J. Song et al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,” arXiv preprint arXiv:2501.12948, 2025. 12

    arXiv 2025

  3. [3]

    From softmax to nucleusmax: A novel sparse language model for chinese radiology report summarization,

    S. Zhao, Q. Li, Y. Yang, J. Wen, and W. Luo, “From softmax to nucleusmax: A novel sparse language model for chinese radiology report summarization,” ACM Transactions on Asian and Low-Resource Language Information Processing, pp. 1–21, 2023

    work page 2023

  4. [4]

    Uni-Retrieval: A Multi-Style Retrieval Framework for STEM's Education

    Y. Jia, X. Wu, H. Li, Q. Zhang, Y. Hu, S. Zhao, and W. Fan, “Uni-retrieval: A multi-style retrieval framework for stem’s education,” arXiv preprint arXiv:2502.05863, 2025

    work page Pith review arXiv 2025

  5. [5]

    Towards robust evaluation of stem education: Leveraging mllms in project-based learning,

    Y. Jia, X. Wu, Q. Zhang, Y. Qin, L. Xiao, and S. Zhao, “Towards robust evaluation of stem education: Leveraging mllms in project-based learning,” arXiv preprint arXiv:2505.17050, 2025

    arXiv 2025

  6. [6]

    From query to explanation: Uni-rag for multi-modal retrieval-augmented learning in stem,

    X. Wu, Y. Jia, L. Xiao, S. Zhao, F. Chiang, and E. Cambria, “From query to explanation: Uni-rag for multi-modal retrieval-augmented learning in stem,” arXiv preprint arXiv:2507.03868 , 2025. [Online]. Available: https://arxiv.org/abs/2507.03868

    arXiv 2025

  7. [7]

    Bloomberggpt: A large language model for finance,

    S. Wu, O. Irsoy, S. Lu, V. Dabravolski, M. Dredze, S. Gehrmann, P. Kambadur, D. Rosenberg, and G. Mann, “Bloomberggpt: A large language model for finance,” arXiv preprint arXiv:2303.17564, 2023

    arXiv 2023

  8. [8]

    Cross-modal knowledge diffusion-based generation for difference-aware medical vqa,

    Q. Lin, K. He, Y. Zhu, F. Xu, E. Cambria, and M. Feng, “Cross-modal knowledge diffusion-based generation for difference-aware medical vqa,” IEEE Transactions on Image Processing , 2025

    work page 2025

Show all 58 references
  1. [9]

    A survey of large language models for healthcare: from data, technology, and applications to accountability and ethics,

    K. He, R. Mao, Q. Lin, Y. Ruan, X. Lan, M. Feng, and E. Cambria, “A survey of large language models for healthcare: from data, technology, and applications to accountability and ethics,” Information Fusion, p. 102963, 2025

    2025

  2. [10]

    Harnessing the potential of multimodal ehr data: A comprehensive survey of clinical predictive modeling for intelligent healthcare,

    J. Wu, K. He, R. Mao, X. Shang, and E. Cambria, “Harnessing the potential of multimodal ehr data: A comprehensive survey of clinical predictive modeling for intelligent healthcare,” Information Fusion, p. 103283, 2025

    2025

  3. [11]

    Chimed-gpt: A chinese medical large language model with full training regime and better alignment to human preferences,

    Y. Tian, R. Gan, Y. Song, J. Zhang, and Y. Zhang, “Chimed-gpt: A chinese medical large language model with full training regime and better alignment to human preferences,” in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics , 2024, pp. 7156–7173

    2024

  4. [12]

    Megacare: Knowledge-guided multi-view hypergraph predictive framework for healthcare,

    J. Wu, K. He, R. Mao, C. Li, and E. Cambria, “Megacare: Knowledge-guided multi-view hypergraph predictive framework for healthcare,” Information Fusion, vol. 100, p. 101939, 2023

    2023

  5. [13]

    The ability of large language models to generate patient information materials for retinopathy of prematurity: Evaluation of readability, accuracy, and comprehensiveness,

    S. A. Postacı and A. Dal, “The ability of large language models to generate patient information materials for retinopathy of prematurity: Evaluation of readability, accuracy, and comprehensiveness,” Turkish Journal of Ophthalmology , vol. 54, no. 6, p. 330, 2024

    2024

  6. [14]

    International classification of retinopathy of prematurity,

    M. F. Chiang, G. E. Quinn, A. R. Fielder, S. R. Ostmo, R. P. Chan, A. Berrocal et al. , “International classification of retinopathy of prematurity,” Ophthalmology, vol. 128, no. 10, pp. e51–e68, 2021

    2021

  7. [15]

    Research progress on the risk factors of retinopathy of prematurity,

    Q. Hong and H. Wei, “Research progress on the risk factors of retinopathy of prematurity,” Hans Journal of Ophthalmology , 2022

    2022

  8. [16]

    Oxygenation fluctuations associated with severe retinopathy of prematurity: Insights from a multimodal deep learning approach,

    W.-C. Lin, B. K. Jordan, B. Scottoline, S. R. Ostmo et al., “Oxygenation fluctuations associated with severe retinopathy of prematurity: Insights from a multimodal deep learning approach,” Ophthalmology Science , vol. 4, no. 2, p. 100417, 2024

    2024

  9. [17]

    Development and validation of a deep learning model to predict the occurrence and severity of retinopathy of prematurity

    Q. Wu, Y. Hu, Z. Mo, R. Wu, X. Zhang, Y. Yang, B. Liu, Y. Xiao et al., “Development and validation of a deep learning model to predict the occurrence and severity of retinopathy of prematurity.” JAMA Network Open, vol. 5, no. 6, 2022

    2022

  10. [18]

    Exploring cognitive and aesthetic causality for multimodal aspect-based sentiment analysis,

    L. Xiao, R. Mao, S. Zhao, Q. Lin, Y. Jia, L. He, and E. Cambria, “Exploring cognitive and aesthetic causality for multimodal aspect-based sentiment analysis,” IEEE Transactions on Affective Computing , 2025

    2025

  11. [19]

    Seeing sound, hearing sight: Uncovering modality bias and conflict of ai models in sound localization,

    Y. Jia, J. Xie, S. Jivaganesh, H. Li, X. Wu, and M. Zhang, “Seeing sound, hearing sight: Uncovering modality bias and conflict of ai models in sound localization,” arXiv preprint arXiv:2505.11217, 2025

    2025

  12. [20]

    Time matters: Examine temporal effects on biomedical language models,

    W. Liu, Z. He, and X. Huang, “Time matters: Examine temporal effects on biomedical language models,” arXiv preprint arXiv:2407.17638, 2024

    2024 arXiv

  13. [21]

    Has multimodal learning delivered universal intelligence in healthcare? a comprehensive survey,

    Q. Lin, Y. Zhu, X. Mei, L. Huang, J. Ma, K. He, Z. Peng, E. Cambria, and M. Feng, “Has multimodal learning delivered universal intelligence in healthcare? a comprehensive survey,” Information Fusion, p. 102795, 2024

    2024

  14. [22]

    Chain-of-interaction: Enhancing large language models for psychiatric behavior understanding by dyadic contexts,

    G. Han, W. Liu, X. Huang, and B. Borsari, “Chain-of-interaction: Enhancing large language models for psychiatric behavior understanding by dyadic contexts,” arXiv preprint arXiv:2403.13786, 2024

    2024 arXiv

  15. [23]

    Evaluating multiple large language models in pediatric ophthalmology,

    J. Holmes, R. Peng, Y. Li, J. Hu, Z. Liu et al., “Evaluating multiple large language models in pediatric ophthalmology,” arXiv preprint arXiv:2311.04368, 2023

    2023 arXiv

  16. [24]

    Exploring the role of chatgpt-4, bingai, and gemini as virtual consultants to educate families about retinopathy of prematurity,

    C. D. Engin, E. Karatas, and T. Ozturk, “Exploring the role of chatgpt-4, bingai, and gemini as virtual consultants to educate families about retinopathy of prematurity,” Children, vol. 11, no. 6, p. 750, 2024

    2024

  17. [25]

    Ermis, E

    S. Ermis, E. ¨Ozal, M. Karapapak, E. Kumanta s ¸, and S. A. ¨Ozal, “Assessing the responses of large language models (chatgpt-4, claude 3, gemini, and microsoft copilot) to frequently asked questions in retinopathy of prematurity: A study on readability and appropriateness,” J...

    2024

  18. [26]

    Ffa-gpt: an interactive visual question answering system for fundus fluorescein angiography,

    D. Shi, X. Chen, W. Zhang, P. Xu, Z. Zhao, Y. Zheng, and M. He, “Ffa-gpt: an interactive visual question answering system for fundus fluorescein angiography,” 2023

    2023

  19. [27]

    Fine-tuning large language model (llm) artificial intelligence chatbots in ophthalmology and llm-based evaluation using gpt-4,

    T. F. Tan, K. Elangovan, L. Jin, Y. Jie, L. Yong, J. Lim, S. Poh, W. Y. Ng, D. Lim, Y. Ke et al., “Fine-tuning large language model (llm) artificial intelligence chatbots in ophthalmology and llm-based evaluation using gpt-4,” arXiv preprint arXiv:2402.10083, 2024

    2024 arXiv

  20. [28]

    Advancing question-answering in ophthalmology with retrieval augmented generations (rag): Benchmarking open-source and proprietary large language models,

    Q. Nguyen, D.-A. Nguyen, K. Dang, S. Liu, S. Y. Wang, and K. o. Nguyen, “Advancing question-answering in ophthalmology with retrieval augmented generations (rag): Benchmarking open-source and proprietary large language models,” medRxiv, pp. 2024–11, 2024

    2024

  21. [29]

    Deep dive in retinal fundus image segmentation using deep learning for retinopathy of prematurity,

    R. Agrawal, S. Kulkarni, R. Walambe, M. Deshpande, and K. Kotecha, “Deep dive in retinal fundus image segmentation using deep learning for retinopathy of prematurity,” Multimedia Tools and Applications , vol. 81, no. 8, pp. 11 441–11 460, 2022

    2022

  22. [30]

    Deep learning-assisted retinopathy of prematurity (rop) screening,

    V. Kumar, H. Patel, K. Paul, and S. Azad, “Deep learning-assisted retinopathy of prematurity (rop) screening,” ACM Transactions on Computing for Healthcare , vol. 4, no. 3, pp. 1–32, 2023

    2023

  23. [31]

    Automated diagnosis of retinopathy of prematurity from retinal images of preterm infants using hybrid deep learning techniques,

    V. R. Sankari, U. Snekhalatha, A. Chandrasekaran, and P. Baskaran, “Automated diagnosis of retinopathy of prematurity from retinal images of preterm infants using hybrid deep learning techniques,” Biomedical Signal Processing and Control , vol. 85, p. 104883, 2023

    2023

  24. [32]

    Evaluation of a deep learning image assessment system for detecting severe retinopathy of prematurity,

    T. K. Redd, J. P. Campbell, J. M. Brown, S. J. Kim, S. Ostmo, R. V. P. Chan et al., “Evaluation of a deep learning image assessment system for detecting severe retinopathy of prematurity,” British Journal of Ophthalmology, vol. 103, no. 5, pp. 580–584, 2019

    2019

  25. [33]

    Atlantis: Aesthetic-oriented multiple granularities fusion network for joint multimodal aspect-based sentiment analysis,

    L. Xiao, X. Wu, J. Xu, W. Li, C. Jin, and L. He, “Atlantis: Aesthetic-oriented multiple granularities fusion network for joint multimodal aspect-based sentiment analysis,” Information Fusion, vol. 106, p. 102304, 2024

    2024

  26. [34]

    Vanessa: Visual connotation and aesthetic attributes understanding network for multimodal aspect-based sentiment analysis,

    L. Xiao, R. Mao, X. Zhang, L. He, and E. Cambria, “Vanessa: Visual connotation and aesthetic attributes understanding network for multimodal aspect-based sentiment analysis,” in Findings of the Association for Computational Linguistics: EMNLP 2024 , 2024, pp. 11 486–11 500

    2024

  27. [35]

    Emo-bias: A large scale evaluation of social bias on speech emotion recognition,

    Y.-C. Lin, H. Wu, H.-C. Chou, C.-C. Lee, and H.-y. Lee, “Emo-bias: A large scale evaluation of social bias on speech emotion recognition,” in Proc. Interspeech 2024, 2024, pp. 4633–4637

    2024

  28. [36]

    Understanding latent affective bias in large pre-trained neural language models,

    A. Kadan, P. Deepak, S. Bhadra, M. P. Gangan, and V. Lajish, “Understanding latent affective bias in large pre-trained neural language models,” Natural Language Processing Journal , vol. 7, p. 100062, 2024

    2024

  29. [37]

    Active learning with complementary sampling for instructing class-biased multi-label text emotion classification,

    X. Kang, X. Shi, Y. Wu, and F. Ren, “Active learning with complementary sampling for instructing class-biased multi-label text emotion classification,” IEEE Transactions on Affective Computing , vol. 14, no. 1, pp. 523–536, 2020

    2020

  30. [38]

    The biases of pre-trained language models: An empirical study on prompt-based sentiment analysis and emotion detection,

    R. Mao, Q. Liu, K. He, W. Li, and E. Cambria, “The biases of pre-trained language models: An empirical study on prompt-based sentiment analysis and emotion detection,” IEEE transactions on affective computing, vol. 14, no. 3, pp. 1743–1753, 2022

    2022

  31. [39]

    Screening examination of premature infants for retinopathy of prematurity,

    W. M. Fierson, M. F. Chiang, W. Good, D. Phelps, J. Reynolds, S. L. Robbins et al., “Screening examination of premature infants for retinopathy of prematurity,” Pediatrics, vol. 142, no. 6, 2018

    2018

  32. [40]

    Optimised retinopathy of prematurity screening guideline in china based on a 5-year cohort study,

    Q. Yang, X. Zhou, Y. Ni, H. Shan, W. Shi, X. Yin, J. Liu, and X. Huang, “Optimised retinopathy of prematurity screening guideline in china based on a 5-year cohort study,” British Journal of Ophthalmology , vol. 105, no. 6, pp. 819–823, 2021

    2021

  33. [41]

    The probabilistic relevance framework: Bm25 and beyond,

    S. Robertson, H. Zaragoza et al., “The probabilistic relevance framework: Bm25 and beyond,” FOUND TRENDS INF RET , 2009

    2009

  34. [42]

    Chatglm: A family of large language models from glm-130b to glm-4 all tools,

    T. GLM, A. Zeng, B. Xu, B. Wang et al., “Chatglm: A family of large language models from glm-130b to glm-4 all tools,” arXiv preprint arXiv:2406.12793, 2024

    2024 arXiv

  35. [43]

    Gemma 2: Improving open language models at a practical size,

    M. Rivi`ere, S. Pathak, P. G. Sessa, C. Hardin et al., “Gemma 2: Improving open language models at a practical size,” CoRR, 2024

    2024

  36. [44]

    Yi: Open foundation models by 01. ai,

    A. Young, B. Chen, C. Li, C. Huang, G. Zhang, G. Zhang, H. Li, J. Zhu, J. Chen, J. Chang et al., “Yi: Open foundation models by 01. ai,” arXiv preprint arXiv:2403.04652, 2024

    2024 arXiv

  37. [45]

    Mixtral of experts,

    A. Q. Jiang, A. Sablayrolles, A. Roux, A. Mensch, B. Savary, C. Bamford, D. S. Chaplot et al., “Mixtral of experts,” arXiv preprint arXiv:2401.04088, 2024

    2024 arXiv

  38. [46]

    Llama 3 model card,

    AI@Meta, “Llama 3 model card,” 2024

    2024

  39. [47]

    Qwen3: Think Deeper, Act Faster,

    QwenLM Team, “Qwen3: Think Deeper, Act Faster,” 2025

    2025

  40. [48]

    Qwen2.5: A party of foundation models,

    Q. Team, “Qwen2.5: A party of foundation models,” September 2024

    2024

  41. [49]

    Training language models to follow instructions with human feedback,

    L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. Wainwright et al., “Training language models to follow instructions with human feedback,” Advances in neural information processing systems , vol. 35, pp. 27 730–27 744, 2022. 13

    2022

  42. [50]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya et al., “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  43. [51]

    Gpt-4o system card,

    A. Hurst, A. Lerer, A. P. Goucher, A. Perelman, A. Ramesh, A. Clark et al., “Gpt-4o system card,” arXiv preprint arXiv:2410.21276, 2024

    2024 arXiv

  44. [52]

    Doubao pro models,

    D. Team, “Doubao pro models,” 2024

    2024

  45. [53]

    Introducing ernie 3.5: Baidu’s knowledge enhanced foundation model takes a giant leap forward,

    B. Team, “Introducing ernie 3.5: Baidu’s knowledge enhanced foundation model takes a giant leap forward,” 2024

    2024

  46. [54]

    Baichuan,

    Baichuan, “Baichuan,” 2024

    2024

  47. [55]

    The claude 3 model family: Opus, sonnet, haiku,

    Anthropic, “The claude 3 model family: Opus, sonnet, haiku,” 2024

    2024

  48. [56]

    Huatuogpt-o1, towards medical complex reasoning with llms,

    J. Chen, Z. Cai, K. Ji et al., “Huatuogpt-o1, towards medical complex reasoning with llms,” arXiv preprint arXiv:2412.18925, 2024

    2024 arXiv

  49. [57]

    Openbiollms: Advancing open-source large language models for healthcare and life sciences,

    M. S. Ankit Pal, “Openbiollms: Advancing open-source large language models for healthcare and life sciences,” 2024

    2024

  50. [58]

    Lora: Low-rank adaptation of large language models,

    E. J. Hu, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, W. Chenet al., “Lora: Low-rank adaptation of large language models,” in International Conference on Learning Representations , 2021

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.