Pith. sign in

REVIEW 4 major objections 4 minor 97 references

Auditable agentic AI for evidence-grounded thyroid ultrasound diagnosis and reporting

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Treating thyroid ultrasound as an auditable, coordinated evidence workflow lifts segmentation, classification, and reporting consistency across multicentre tests.

desk verdict A solid medical-AI systems paper with a genuinely useful new dataset and workflow, but the abstract's headline AUROC hides the private multicentre cohort's 0.819 and several near-chance centres. read the letter →

arxiv 2608.12590 v1 pith:MM3CMOVQ submitted 2026-08-12 cs.AI cs.CV

Haifan Gong , Shiyu Chen , Bodong Wang , Yuqi Wang , Shijie Wang , Guoliang You , Xinyu Xiong , Haowei Wang
show 6 more authors
Mingzhi Mao Dexing Kong Qinghua Liu Wei Lou Fei Chen Guanbin Li
This is my paper · ORCID
classification cs.AIcs.CV
keywords agenticAIthyroidultrasoundcase-levelevidencerecordnodulesegmentationbenign-malignantclassificationreportgenerationclinician-in-the-loopauditability
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

Thyroid ultrasound diagnosis is a chain of linked steps: locate a nodule, measure it, characterise it, decide on risk, and write a report. This paper argues that AI should coordinate that whole chain around a case-level evidence record instead of firing off isolated predictions. The central claim is that an agentic workflow which routes cases to specialised tools, consolidates masks, probabilities and radiomic descriptors, and stores everything for clinician correction outperforms both standalone models and end-to-end multimodal language models on heterogeneous data. Reported results include a mean Dice of 87.21% for nodule segmentation, a mean AUROC of 0.9466 for benign-malignant classification, and an increase in report diagnostic consistency from 70.3% to 86.2% with reduced physician time. If the approach holds, clinical AI for multi-step diagnostic tasks can be built to be inspectable and correctable rather than a final black-box answer.

What carries the argument

The central object is the case-level evidence record, a structured store of lesion masks, measurements, class probabilities, radiomic descriptors, SHAP explanations, confidence signals and report clauses for one examination. The mechanism that carries the argument is the agent-as-workflow-controller: an LLM router reads structured summaries and metadata, selects the most reliable expert output for each case, writes the evidence to the shared record, and later converts that evidence into report text through template retrieval, slot filling and clause combination. This design makes auditability a property of the workflow itself, because every output traces back to editable evidence and corrections propagate forward.

What would settle it

Look at Table S11: on the private NHC-MISD-TUS cohort, pooled AUROC is 0.819, but several centres sit near chance, including THYB_S_ZJ24 (n=269, AUROC 0.590), THYB_S_JL04 (n=12, AUROC 0.407) and THYB_S_YN05 (n=12, AUROC 0.500). If a validation protocol counted per-centre failures rather than pooled means, and a meaningful fraction of larger centres stayed below 0.6, the multicentre robustness claim would be falsified.

Watch

Extended reading notes

Core claim

The paper's discovery is that the unit of AI assistance should be the auditable case-level evidence record, not a single prediction endpoint. ThyroidXAgent plans each case, routes inputs to specialised segmentation, classification, radiomics and measurement tools, and stores masks, class probabilities, radiomic features, uncertainty signals and report clauses in a shared record. That record can be inspected, corrected by a clinician, and reused downstream: a corrected mask updates SHAP attributions and classification confidence, and the same evidence drives evidence-grounded report assembly. Across seven segmentation test sets and five classification test sets the workflow led on most metrics, and in reader studies physicians using the evidence achieved higher accuracy and consistency while taking less time. The claim is that this coordination-and-evidence architecture, rather than any single new model, is what carries the improvement.

Load-bearing premise

The broad multicentre-generalization claim rests on pooled per-centre metrics; if pooling hides centres that perform at chance, the claim that the workflow is consistent across heterogeneous settings loses force.

Editorial extensions

If this is right

  • A clinician can correct an intermediate lesion mask and the downstream radiomics, classification and report update automatically instead of restarting the pipeline.
  • Report generation can be constrained by structured facts rather than free-form text, lowering the risk of hallucinated lesion locations or measurements.
  • The same segmentation-radiomics-classification tools are reusable for different clinical questions, such as lymph-node metastasis and subtype classification, with only the classifier and task instructions changed.
  • If the reported workload reductions replicate, AI-assisted review could shorten segmentation and reporting time without sacrificing segmentation quality.
  • The evidence record provides a natural audit trail for retrospective review of AI-supported diagnoses.

Reading between the lines

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

  • Editorial inference: pooling per-centre numbers may hide clinically useless performance at individual sites; a deployment-ready system should report and monitor centre-specific AUROC and Dice, especially where the paper's private cohort shows values near chance.
  • Editorial inference: the template-bank approach suggests reports can be adapted to new hospitals or languages by swapping retrieval libraries, without retraining image tools; that is a testable extension.
  • Editorial inference: clinician corrections written back to the evidence store could become a source of continuous model improvement, though privacy, provenance and versioning governance would be required.
  • Editorial inference: the same orchestration pattern could be transferred to other structured ultrasound examinations, such as breast or abdominal imaging, where the evidence objects and report clauses differ but the workflow controller stays the same.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper presents ThyroidXAgent, an agentic AI system that coordinates specialized tools for thyroid ultrasound nodule segmentation, benign-malignant classification, malignant-lesion stratification, and report generation, storing intermediate outputs in an auditable case-level evidence record. The system was trained on a newly assembled OpenThyroidDB resource and evaluated on held-out public test sets plus a private 8,721-case, 35-centre NHC-MISD-TUS cohort. The main reported results are a mean Dice of 87.21% across seven segmentation test sets, a mean AUROC of 0.9466 across five public classification test sets, an AUROC of 0.819 on the private multicentre cohort, and a reader study showing improved physician classification accuracy, increased report diagnostic consistency (70.3% to 86.2%), and reduced segmentation/reporting time. The paper also introduces ThyClinScore, a lesion-level clinical semantic metric for report evaluation, and reports that it correlates better with an LLM judge than overlap-based metrics. The central claim is that an evidence-centred, clinician-correctable agentic workflow outperforms isolated predictive models and supports auditable clinical reporting.

Significance. If the results hold, the paper makes a useful contribution by reframing thyroid ultrasound AI around a case-level evidence record rather than isolated prediction endpoints. The manuscript is strong in its scale and methodology: it provides bootstrap confidence intervals throughout, uses genuinely held-out test sets, includes an independent private multicentre cohort, reports reader studies in a crossover design, and makes code and data publicly available. The strongest conceptual value is the demonstration that an agent can route cases to specialized tools and expose intermediate evidence for clinician correction, which is a plausible path toward auditable clinical AI. The main caveat is that the headline generalization claims are stronger than the evidence: the private-cohort AUROC of 0.819 is materially lower than the public-set mean of 0.9466, and per-centre variation is large; in addition, one of the 'external' public sets shares an institution with training data. The validation of ThyClinScore against an LLM judge rather than human expert scores is also indirect. These issues do not invalidate the paper, but they require careful revision of the claims.

major comments (4)
  1. [§2.2, Fig. 2h, Table S11] The abstract's headline 'mean AUROC of 0.9466' is the unweighted mean over five public test sets and omits the private NHC-MISD-TUS result, which is 0.819 (Table S11, overall row). More importantly, the pooled private-cohort AUROC conceals very large per-centre heterogeneity: THYB_S_ZJ24 (N=269) has AUROC 0.590, THYB_S_BJ09 (N=20) 0.586, THYB_S_JL04 (N=12) 0.407, and THYB_S_YN05 (N=12) 0.500. Since the paper's key generalization claim rests on 'heterogeneous datasets' and 'multicentre robustness', the pooled number alone is insufficient. The authors should report a centre-level analysis, for example a random-effects estimate with per-centre CIs, and explicitly discuss which centres are near chance and why. They should also revise the abstract to include the private-cohort AUROC or temper the generalization language.
  2. [§2.1, Table S1] ZJH-8K is described as an independent external test set, but Table S1 states that it was collected at Zhujiang Hospital, Southern Medical University, the same institution that contributed the TN3K training data and the SMU-HMC report-training data. Institutional overlap with training data weakens the 'external' characterization and makes the cross-dataset generalization claim less strong than presented. The paper should explicitly acknowledge this overlap and re-evaluate the cross-dataset conclusions with ZJH-8K excluded or flagged as a same-institution holdout.
  3. [§4.4, Fig. 4c, §2.5] ThyClinScore is a central evaluation metric for report quality, but its parameters (λ, η, τ, α_s, β_f, the matching threshold, and the component weights w_q and u_r in §4.4) are described as predefined without any sensitivity analysis. The only external validation is a Pearson correlation with a location-aware LLM judge (Fig. 4c, r = 0.696), which is itself not a clinical ground truth. Since the paper uses ThyClinScore to claim that evidence-grounded assembly outperforms LLM baselines, the authors should demonstrate that the ordering of systems under ThyClinScore is stable across reasonable parameter choices, or validate the metric against human expert ratings on a sample of reports.
  4. [§2.6, Fig. 6] The reader study that supports the workflow-level claims of improved diagnostic consistency (70.3% to 86.2%) and reduced reporting time (2.5 to 1.8 min/case) involves only two physicians and 145 cases, and no confidence intervals or significance tests are reported for the consistency change. The crossover design is a strength, but the manuscript should provide at least bootstrap CIs or a paired test for the consistency and time endpoints, and should dampen the strength of the claim given the small number of readers.
minor comments (4)
  1. [Abstract and Introduction] The abstract contains garbled characters such as 'benign ⚶malignant', and the introduction has a duplicated verb in 'ThyroidXAgent improved produced more clinically consistent reports'. These should be corrected.
  2. [§2.2 and Table S3] The text states that ThyroidXAgent obtained the highest Dice on six of seven test sets and the lowest HD95 on all test sets, which is accurate. However, the PKTN result (82.99% Dice versus MedSAM2's 83.46%) is a clear exception and should be acknowledged in the main text when the 'six of seven' claim is made.
  3. [§5 and §6] The Code availability section says the source code is publicly available on GitHub but also says 'The website will be made available after acceptance', which is contradictory. Please clarify which resources are currently accessible.
  4. [Table S3 and elsewhere] There are several typographical issues, such as 'Dice coefficient' in Table S3 and 'Confilts'/'confilt' in Section 9; these should be fixed in proof.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: core segmentation and classification results are held-out test evaluations, and the report-generation comparisons rest on conventional metrics and human reader studies rather than on definitions.

full rationale

ThyroidXAgent's central claims are empirical evaluations on held-out test sets that are not used for fitting: segmentation (Dice 87.21%, HD95 6.90 mm) and classification (AUROC 0.9466) are reported on independent public cohorts including DDTI, RJH-7K and ZJH-8K, and the private NHC-MISD-TUS set is a separate external validation cohort. The reader studies are human experiments with cross-over design and time/consistency measurements, so those results are not reducible to any fitted model output. The only self-referential element is ThyClinScore, which is explicitly introduced as a new metric with predefined weights in Section 4.4 rather than derived from or fitted to the report-generation outputs; its correlation with a location-aware LLM judge is an empirical comparison, and the report-generation ranking is additionally supported by BLEU/ROUGE/METEOR and by the human reader study. Self-citations to prior data resources such as TN3K are dataset provenance and are not load-bearing evidence for the agent's effectiveness. No equation or fitted parameter is renamed as a prediction, and no uniqueness or self-citation chain forces the central results. The headline generalization concern about the private cohort's pooled AUROC of 0.819 and per-centre variability is a correctness/interpretation issue, not a circularity issue, because the private cohort is still an external held-out evaluation rather than an input used in constructing the model.

Assumptions & free parameters 2 free parameters · 3 assumptions · 1 invented entities

The central empirical claims rest on held-out evaluations, which are largely self-contained. The main hand-chosen inputs are the ThyClinScore metric weights and expert-pool configuration choices; the main domain assumptions are the validity of the LLM judge, the TNVideo label construction, and the interpretation of pooled private-cohort metrics.

free parameters (2)
  • ThyClinScore weighting parameters (lambda, eta, tau, alpha_s, beta_f, matching threshold, component weights) = not fully specified in main text
    These hand-chosen constants define the report-quality metric used to rank systems; reported ThyClinScore values (0.5238 on SMU-HMC, 0.4465 on KMVE, 0.4775 on ZJH-TS) depend on them, and the paper does not state all values in the main text.
  • Expert-pool configuration choices (input resolutions, stacked-training composition, dilation rates) = 128/224/448 pixels; dataset1-4 stacks; varied dilation
    The router and expert pool were configured by comparing configurations (Table S6); selection of the final configuration is partly guided by performance, so these are effective hand-chosen choices influencing the headline numbers.
assumptions (3)
  • domain assumption The location-aware LLM judge is a valid reference for clinical report quality.
    Used to validate ThyClinScore in Section 4.4 and Fig. 4c; no external human-rated report corpus or clinical outcome anchors this judge.
  • domain assumption Annotation-derived labels in the TNVideo reader study (bounding-box aggregation into case-level benign/malignant) are accurate ground truth.
    The 70.3 percent to 86.2 percent consistency claim depends on these labels and the prespecified rule-based assignment of report direction (Section 4.6).
  • domain assumption The pooled NHC-MISD-TUS per-centre metrics are a suitable summary of multicentre generalization.
    Overall AUROC 0.819 aggregates centres with very different AUROCs, including near-chance centres (Table S11); treating the pooled mean as multicentre robustness is an interpretive assumption.
invented entities (1)
  • ThyClinScore
    purpose: Report-level clinical semantic metric for lesion matching, size, vascularity, morphology and completeness.
    Introduced in this paper; agreement against a location-aware LLM judge is the only validation, with no external human-rated report corpus or clinical outcome benchmark; metric weights are hand-chosen.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Auditable agentic AI for evidence-grounded thyroid ultrasound diagnosis and reporting." pith.science (2026). https://pith.science/paper/MM3CMOVQ

@misc{pith2026260812590,
  author       = {Pith},
  title        = {Pith review of: Auditable agentic AI for evidence-grounded thyroid ultrasound diagnosis and reporting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MM3CMOVQ}},
  note         = {Machine review of arXiv:2608.12590}
}
read the original abstract

Thyroid ultrasound diagnosis requires coordinated lesion localization, measurement, risk stratification and reporting, yet most AI systems address these tasks in isolation and provide limited support for clinical review. We present ThyroidXAgent, a clinician-interactive agentic AI system that coordinates specialized diagnostic tools and stores their outputs as an auditable case-level evidence record. The system was developed using OpenThyroidDB, a multicentre, multitask resource integrating approximately 0.3 million ultrasound images and 24,000 paired reports, and was evaluated on 28,458 non-overlapping test cases, including 8,721 cases from 35 centres in the private NHC-MISD-TUS cohort. Across heterogeneous datasets, ThyroidXAgent achieved a mean Dice score of 87.21 percent for nodule segmentation and a mean AUROC of 0.9466 for benign-malignant classification. The same workflow supported lymph-node metastasis prediction and follicular versus papillary thyroid carcinoma classification, with AUROCs of 0.864 and 0.805, respectively. For report generation, evidence-grounded assembly outperformed multimodal language-model baselines across three cohorts. ThyClinScore, a lesion-level clinical semantic metric introduced here, showed the strongest correlation with a location-aware language-model judge. ThyroidXAgent improved physician classification accuracy, increased report diagnostic consistency from 70.3 percent to 86.2 percent, and reduced segmentation and reporting time by 35.9 percent and 27.4 percent, respectively. These findings support auditable, clinician-correctable agentic AI for thyroid ultrasound diagnosis and reporting.

Figures

Figures reproduced from arXiv: 2608.12590 by the authors.

Figure 1
Figure 1. Multicenter thyroid ultrasound data and the ThyroidXAgent evidence workflow. a, OpenThyroidDB integrates curated public thyroid ultrasound resources and institutionally governed clinical cohorts into a full-spectrum resource for segmentation, benign-malignant classification, report generation and malignant-lesion stratification across diverse scanners and acquisition settings. b, The clinician-feedback workflow stor… view at source ↗
Figure 2
Figure 2. Agent-routed evidence for thyroid nodule segmentation, benign-malignant classification and clinician review. a, Interactive review workflow: clinicians assess the predicted nodule mask, accept the classification if the mask is reliable, or refine it through box annotation and interactive segmentation; SHAP-based feature attributions are then recomputed from the corrected mask. b-e, Segmentation and classification pe… view at source ↗
Figure 3
Figure 3. Shared ThyroidXAgent tools support malignant-lesion stratification with task-specific radiomic attributions. a, Workflow for SHAP-based interpretation of malignant-lesion tasks. b, Performance comparison between ThyroidXAgent and the corresponding specialist baselines for FTC/PTC subtype classification and lymph node metastasis prediction, reported as AUROC and AUPRC; percentages denote the relative improvement of T… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: ThyClinScore for clinical semantic evaluation of thyroid ultrasound reports. a, ThyClinScore first structures the ground-truth and predicted reports into lesion-level entries, matches corresponding lesions, and scores clinically relevant attributes, including size, vas…
Figure 5
Figure 5. Figure 5: Evidence-grounded report assembly in ThyroidXAgent. a, Case-level thyroid ultrasound input, including video or image sequences, multiple views and anatomical regions, and greyscale and CDFI modalities. b, Input-preparation skills crop regions of interest, parse image c…
Figure 6
Figure 6. Figure 6: Reader study of AI-assisted thyroid ultrasound reporting. a, Cross-over reader-study design. Each ultrasound video was interpreted under both manual and AI-assisted conditions by different physicians, reducing recall bias while enabling paired case-level comparisons. b…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

97 extracted references · 43 canonical work pages

  1. [1]

    Alexander, E. K. & Cibas, E. S. Diagnosis of thyroid nodules.The Lancet Diabetes & Endocrinol. 10, 533–539, DOI: 10.1016/S2213-8587(22)00101-2 (2022)

    work page doi:10.1016/s2213-8587(22)00101-2 2022

  2. [2]

    & Durante, C

    Grani, G., Sponziello, M., Filetti, S. & Durante, C. Thyroid nodules: diagnosis and management.Nat. Rev. Endocrinol. DOI: 10.1038/s41574-024-01025-4 (2024)

    work page doi:10.1038/s41574-024-01025-4 2024

  3. [3]

    Tessler, F. N.et al. Acr thyroid imaging, reporting and data system (ti-rads): White paper of the acr ti-rads committee. J. Am. Coll. Radiol. 14, 587–595, DOI:10.1016/j.jacr.2017.01.046 (2017)

    doi:10.1016/j.jacr.2017.01.046 2017

  4. [4]

    Hoang, J. K.et al. Interobserver variability of sonographic features used in the american college of radiology thyroid imaging reporting and data system. Am. J. Roentgenol. 211, 162–167, DOI:10.2214/AJR.17.19192 (2018)

    work page doi:10.2214/ajr.17.19192 2018

  5. [5]

    Cibas, E. S. & Ali, S. Z. The 2017 bethesda system for reporting thyroid cytopathology.Thyroid 27, 1341–1346, DOI: 10.1089/thy.2017.0500 (2017)

    work page arXiv 2017

  6. [6]

    Topol, E. J. High-performance medicine: the convergence of human and artificial intelligence.Nat. Medicine 25, 44–56, DOI:10.1038/s41591-018-0300-7 (2019)

    doi:10.1038/s41591-018-0300-7 2019

  7. [7]

    & Topol, E

    Rajpurkar, P., Chen, E., Banerjee, O. & Topol, E. J. Ai in health and medicine.Nat. Medicine 28, 31–38, DOI: 10.1038/s41591-021-01614-0 (2022)

    doi:10.1038/s41591-021-01614-0 2022

  8. [8]

    Chen, H., Gomez, C., Huang, C.-M. et al. Explainable medical imaging AI needs human-centered design: Guidelinesandevidencefromasystematicreview. npj Digit. Medicine 5, 156, DOI:10.1038/s41746-022-00699-2 (2022)

    doi:10.1038/s41746-022-00699-2 2022

Show all 97 references
  1. [9]

    K., Gilbert, S

    Wekenborg, M. K., Gilbert, S. & Kather, J. N. Examining human–AI interaction in real-world healthcare beyond the laboratory.npj Digit. Medicine 8, 169, DOI:10.1038/s41746-025-01559-5 (2025)

    2025 doi

  2. [10]

    Giddings, R. et al. Factors influencing clinician and patient interaction with machine learning-based risk prediction models: A systematic review.The Lancet Digit. Heal. 6, e131–e144, DOI:10.1016/S2589-7500(23) 00241-8 (2024)

    2024 doi

  3. [11]

    Gong, H. et al. Multi-task learning for thyroid nodule segmentation with thyroid region prior. In2021 IEEE 18th international symposium on biomedical imaging (ISBI) , 257–261 (2021)

    2021

  4. [12]

    Thyroid region prior guided attention for ultrasound segmentation of thyroid nodules.Comput

    Gong, H.et al. Thyroid region prior guided attention for ultrasound segmentation of thyroid nodules.Comput. biology medicine 155, 106389 (2023)

    2023

  5. [13]

    & Zhao, Q

    Sun, X., Wei, B., Jiang, Y., Mao, L. & Zhao, Q. Clip-tnseg: A multi-modal hybrid framework for thyroid nodule segmentation in ultrasound images.arXiv preprint arXiv:2412.05530 (2024)

    2024 arXiv

  6. [14]

    Gong, H. et al. Less is more: adaptive curriculum learning for thyroid nodule diagnosis. In International Conference on Medical Image Computing and Computer-Assisted Intervention , 248–257 (2022)

    2022

  7. [15]

    Peng, S. et al. Deep learning-based artificial intelligence model to assist thyroid nodule diagnosis and man- agement: a multicentre diagnostic study.The Lancet Digit. Heal. 3, e250–e259, DOI:10.1016/S2589-7500(21) 00041-8 (2021)

    2021 doi

  8. [16]

    Chen, Y. et al. An artificial intelligence model based on acr ti-rads characteristics for us diagnosis of thyroid nodules. Radiology 303, 613–619, DOI:10.1148/radiol.211455 (2022)

    2022 doi

  9. [17]

    Multimodal gpt model for assisting thyroid nodule diagnosis and management.npj Digit

    Yao, J.et al. Multimodal gpt model for assisting thyroid nodule diagnosis and management.npj Digit. Medicine 8, 245, DOI:10.1038/s41746-025-01652-9 (2025). 46/51

    2025 doi

  10. [18]

    Li, J., Su, T. et al. Ultrasound report generation with cross-modality feature alignment via unsupervised guidance. IEEE Transactions on Med. Imaging 44, 19–30 (2024)

    2024

  11. [19]

    Tanno, R., Barrett, D. G. T., Sellergren, A.et al. Collaboration between clinicians and vision–language models in radiology report generation.Nat. Medicine 31, 599–608, DOI:10.1038/s41591-024-03302-1 (2025)

    2025 doi

  12. [20]

    Towards a holistic framework for multimodal llm in 3d brain ct radiology report generation

    Li, C.-Y., Chang, K.-J.et al. Towards a holistic framework for multimodal llm in 3d brain ct radiology report generation. Nat. Commun. 16, 2258 (2025)

    2025

  13. [21]

    Wang, J. et al. Deep learning models for thyroid nodules diagnosis of fine-needle aspiration biopsy: a ret- rospective, prospective, multicentre study in china. The Lancet Digit. Heal. 6, e458–e469, DOI: 10.1016/ S2589-7500(24)00085-2 (2024)

    2024

  14. [22]

    Explainablemultimodaldeeplearningforpredictingthyroidcancerlaterallymphnodemetastasis using ultrasound imaging

    Shen, P.et al. Explainablemultimodaldeeplearningforpredictingthyroidcancerlaterallymphnodemetastasis using ultrasound imaging. Nat. Commun. 16, 7052, DOI:10.1038/s41467-025-62042-z (2025)

    2025 doi

  15. [23]

    Dai, F. et al. Improving ai models for rare thyroid cancer subtype by text guided diffusion models.Nat. Commun. 16, 4449, DOI:10.1038/s41467-025-59478-8 (2025)

    2025 doi

  16. [24]

    Tikhomirov, L. et al. Medical artificial intelligence for clinicians: The lost cognitive perspective.The Lancet Digit. Heal. 6, e589–e594, DOI:10.1016/S2589-7500(24)00095-5 (2024)

    2024 doi

  17. [25]

    & Fan, Y

    You, G., Li, H., Zhang, Y. & Fan, Y. Learning anatomy-grounded CT vision-language representations with organ-hierarchical report knowledge.arXiv preprint arXiv:2607.10953 DOI: 10.48550/arXiv.2607.10953 (2026)

    2026 doi arXiv

  18. [26]

    Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: Decide-ai.Nat

    Vasey, B.et al. Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: Decide-ai.Nat. Medicine 28, 924–933, DOI:10.1038/s41591-022-01772-9 (2022)

    2022 doi

  19. [27]

    Dreyer, M. et al. Mechanistic understanding and validation of large ai models with semanticlens.Nat. Mach. Intell. 7, 1572–1585 (2025)

    2025

  20. [28]

    N., Rosenberg, L., Willcox, G.et al

    Patel, B. N., Rosenberg, L., Willcox, G.et al. Human–machine partnership with artificial intelligence for chest radiograph diagnosis. npj Digit. Medicine 2, 111, DOI:10.1038/s41746-019-0189-7 (2019)

    2019 doi

  21. [29]

    Leibig, C. et al. Combining the strengths of radiologists and AI for breast cancer screening: A retrospective analysis. The Lancet Digit. Heal. 4, e507–e519, DOI:10.1016/S2589-7500(22)00070-X (2022)

    2022 doi

  22. [30]

    Yu, F. et al. Heterogeneity and predictors of the effects of AI assistance on radiologists.Nat. Medicine 30, 837–849, DOI:10.1038/s41591-024-02850-w (2024)

    2024 doi

  23. [31]

    Impact of human and artificial intelligence collaboration on workload reduction in medical image interpretation

    Chen, M., Wang, Y., Wang, Q.et al. Impact of human and artificial intelligence collaboration on workload reduction in medical image interpretation. npj Digit. Medicine 7, 349, DOI: 10.1038/s41746-024-01328-w (2024)

    2024 doi

  24. [32]

    S., Bunning, B

    Everett, S. S., Bunning, B. J., Jain, P.et al. From tool to teammate in a randomized controlled trial of clinician– AI collaborative workflows for diagnosis.npj Digit. Medicine 9, 409, DOI:10.1038/s41746-026-02545-1 (2026)

    2026 doi

  25. [33]

    Human–AI collaboration in healthcare: A scoping review

    Strong, J., Rogers, H., Sun, E.et al. Human–AI collaboration in healthcare: A scoping review. npj Digit. Medicine DOI: 10.1038/s41746-026-02918-6 (2026)

    2026 doi

  26. [34]

    Wiens, J. et al. Do no harm: a roadmap for responsible machine learning for health care.Nat. Medicine 25, 1337–1340, DOI:10.1038/s41591-019-0548-6 (2019)

    2019 doi

  27. [35]

    & Topol, E

    Zou, J. & Topol, E. J. The rise of agentic ai teammates in medicine.The Lancet 405, 457, DOI: 10.1016/ S0140-6736(25)00202-8 (2025). 47/51

    2025

  28. [36]

    Moor, M. et al. Foundation models for generalist medical artificial intelligence.Nature 616, 259–265, DOI: 10.1038/s41586-023-05881-4 (2023)

    2023 doi

  29. [37]

    Kohane, I. S. Injecting artificial intelligence into medicine.NEJM AI 1, 1–3, DOI:10.1056/AIe2300197 (2024)

    2024 doi

  30. [38]

    GPT versus resident physicians—a benchmark based on official board scores

    Katz, U., Cohen, E., Shachar, E.et al. GPT versus resident physicians—a benchmark based on official board scores. NEJM AI 1, DOI:10.1056/AIdbp2300192 (2024)

    2024 doi

  31. [39]

    Zhou, H.-Y. et al. Medversa: A generalist foundation model for diverse medical imaging tasks.NEJM AI 3, DOI: 10.1056/AIoa2500595 (2026)

    2026 doi

  32. [40]

    Towards conversational diagnostic artificial intelligence.Nature 642, 442–450, DOI:10.1038/s41586-025-08866-7 (2025)

    Tu, T., Schaekermann, M., Palepu, A.et al. Towards conversational diagnostic artificial intelligence.Nature 642, 442–450, DOI:10.1038/s41586-025-08866-7 (2025)

    2025 doi

  33. [41]

    Towards accurate differential diagnosis with large language models

    McDuff, D., Schaekermann, M., Tu, T.et al. Towards accurate differential diagnosis with large language models. Nature 642, 451–457, DOI:10.1038/s41586-025-08869-4 (2025)

    2025 doi

  34. [42]

    Deltadahl, S. et al. Deep generative classification of blood cell morphology.Nat. Mach. Intell. 7, 1791–1803 (2025)

    2025

  35. [43]

    Pontikos, N. et al. Next-generation phenotyping of inherited retinal diseases from multimodal imaging with eye2gene. Nat. Mach. Intell. 7, 967–978 (2025)

    2025

  36. [44]

    Qiu, J. et al. Llm-based agentic systems in medicine and healthcare.Nat. Mach. Intell. 6, 1418–1420, DOI: 10.1038/s42256-024-00944-1 (2024)

    2024 doi

  37. [45]

    & Rajpurkar, P

    Moritz, M., Topol, E. & Rajpurkar, P. Coordinated ai agents for advancing healthcare.Nat. Biomed. Eng. 9, 432–438, DOI:10.1038/s41551-025-01363-2 (2025)

    2025 doi

  38. [46]

    Towards autonomous medical artificial intelligence agents.Nature DOI: 10.1038/s41586-026-10675-5 (2026)

    Ferber, D., Hilgers, L., H”oper, C.et al. Towards autonomous medical artificial intelligence agents.Nature DOI: 10.1038/s41586-026-10675-5 (2026)

    2026 doi

  39. [47]

    G., Haider, S

    Collaco, B. G., Haider, S. A., Prabha, S.et al. The role of agentic artificial intelligence in healthcare: A scoping review. npj Digit. Medicine 9, 345, DOI:10.1038/s41746-026-02517-5 (2026)

    2026 doi

  40. [48]

    Kong, Q. et al. Ai agent-based discovery of d-enantiomeric antimicrobial peptides against multidrug-resistant bacterial infection. Biomaterials 123927 (2025)

    2025

  41. [49]

    Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models

    Wang, L., Xu, W.et al. Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models. In Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1: long papers) , 2609–2634 (2023)

    2023

  42. [50]

    AgentClinic: A multimodal benchmark for tool-using clinical AI agents

    Schmidgall, S., Ziaei, R., Harris, C.et al. AgentClinic: A multimodal benchmark for tool-using clinical AI agents. npj Digit. Medicine 9, 499, DOI:10.1038/s41746-026-02674-7 (2026)

    2026 doi

  43. [51]

    I., Jiang, X.et al

    Liu, Y., Carrero, Z. I., Jiang, X.et al. Benchmarking large language model-based agent systems for clinical decision tasks. npj Digit. Medicine 9, 259, DOI:10.1038/s41746-026-02443-6 (2026)

    2026 doi

  44. [52]

    React: Synergizing reasoning and acting in language models

    Yao, S., Zhao, J.et al. React: Synergizing reasoning and acting in language models. InThe eleventh interna- tional conference on learning representations (2022)

    2022

  45. [53]

    An autonomous agentic workflow for clinical detection of cognitive concerns using large language models.npj Digit

    Tian, J., Fard, P., Cagan, C.et al. An autonomous agentic workflow for clinical detection of cognitive concerns using large language models.npj Digit. Medicine 9, 51, DOI:10.1038/s41746-025-02324-4 (2026)

    2026 doi

  46. [54]

    Jiang, Y. et al. Medagentbench: A virtual ehr environment to benchmark medical llm agents. NEJM AI AIdbp2500144, DOI:10.1056/AIdbp2500144 (2025). 48/51

    2025 doi

  47. [55]

    Tn5000: An ultrasound image dataset for thyroid nodule detection and classification

    Zhang, H., Liu, Q., Han, X.et al. Tn5000: An ultrasound image dataset for thyroid nodule detection and classification. Sci. Data 12, 1437, DOI:10.1038/s41597-025-05757-4 (2025)

    2025 doi

  48. [56]

    Van Griethuysen, J. J.et al. Computational radiomics system to decode the radiographic phenotype.Cancer research 77, e104–e107 (2017)

    2017

  49. [57]

    A hierarchical model for data-to-text generation

    Rebuffel, C., Soulier, L.et al. A hierarchical model for data-to-text generation. InEuropean Conference on Information Retrieval, 65–80 (Springer, 2020)

    2020

  50. [58]

    & Gal, Y

    Farquhar, S., Kossen, J., Kuhn, L. & Gal, Y. Detecting hallucinations in large language models using semantic entropy. Nature 630, 625–630 (2024)

    2024

  51. [59]

    Duong, V. H.et al. Thyroidxl: Advancing thyroid nodule diagnosis with an expert-labeled, pathology-validated dataset. InMedical Image Computing and Computer Assisted Intervention – MICCAI 2025 , vol. 15974, 616–626, DOI: 10.1007/978-3-032-05182-0_60 (Springer Nature Switzerland, 2025)

    2025 doi

  52. [60]

    Pedraza, L. et al. An open access thyroid ultrasound image database. In10th International Symposium on Medical Information Processing and Analysis , vol. 9287, 188–193 (2015)

    2015

  53. [61]

    Shusharina, N., Heinrich, M. P. & Huang, R.Segmentation, Classification, and Registration of Multi-modality Medical Imaging Data: MICCAI 2020 Challenges, ABCs 2020, L2R 2020, TN-SCUI 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 4–8, 2020, Proceedings (Spring...

    2020

  54. [62]

    Dai, F. et al. Improving ai models for rare thyroid cancer subtype by text guided diffusion models.Nat. Commun. 16, 4449 (2025)

    2025

  55. [63]

    Liu, Z. & He, K. A decade’s battle on dataset bias: Are we there yet? InInternational Conference on Learning Representations(2025)

    2025

  56. [64]

    Ma, J. et al. Medsam2: Segment anything in 3d medical images and videos.arXiv preprint arXiv:2504.03600 (2025)

    2025 arXiv

  57. [65]

    Jiang, Y. et al. From pretraining to privacy: Federated ultrasound foundation model with self-supervised learning. npj Digit. Medicine 8, 714, DOI:10.1038/s41746-025-02085-0 (2025)

    2025 doi

  58. [66]

    & Ding, G

    Wang, A., Chen, H., Lin, Z., Han, J. & Ding, G. Repvit: Revisiting mobile cnn from vit perspective. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 15909–15920 (2024)

    2024

  59. [67]

    Dong, C. et al. A survey of natural language generation.ACM Comput. Surv. 55, 1–38 (2022)

    2022

  60. [68]

    Gpt-5 system card.https://openai.com/index/gpt-5-system-card/ (2025)

    OpenAI. Gpt-5 system card.https://openai.com/index/gpt-5-system-card/ (2025). Published August 7, 2025

    2025

  61. [69]

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261 (2025)

    Comanici, G., Bieber, E.et al. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261 (2025)

    2025 arXiv

  62. [70]

    Zhang, S. et al. A multimodal biomedical foundation model trained from fifteen million image⚶text pairs. NEJM AI 2, DOI:10.1056/AIoa2400640 (2024)

    2024 doi

  63. [71]

    Sellergren, A. et al. Medgemma technical report. arXiv preprint arXiv:2507.05201 (2025)

    2025 arXiv

  64. [72]

    Lundberg, S. M. & Lee, S.-I. A unified approach to interpreting model predictions. InAdvances in Neural Information Processing Systems, vol. 30 (2017)

    2017

  65. [73]

    Bleu: a method for automatic evaluation of machine translation

    Papineni, K., Roukos, S.et al. Bleu: a method for automatic evaluation of machine translation. InProceedings of the 40th Annual Meeting of the Association for Computational Linguistics , 311–318 (2002)

    2002

  66. [74]

    Rouge: A package for automatic evaluation of summaries

    Lin, C.-Y. Rouge: A package for automatic evaluation of summaries. InText Summarization Branches Out , 74–81 (2004). 49/51

    2004

  67. [75]

    & Lavie, A

    Banerjee, S. & Lavie, A. 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, 65–72 (2005)

    2005

  68. [76]

    & Mullainathan, S

    Obermeyer, Z., Powers, B., Vogeli, C. & Mullainathan, S. Dissecting racial bias in an algorithm used to manage the health of populations.Science 366, 447–453, DOI:10.1126/science.aax2342 (2019)

    2019 doi

  69. [77]

    Gong, H., Lu, Y., Wan, X. & Li, H. Domain generalized medical landmark detection via robust boundary-aware pre-training. In Proceedings of the AAAI Conference on Artificial Intelligence , vol. 39, 3140–3148 (2025)

    2025

  70. [78]

    Gong, H. et al. Intermediate domain alignment and morphology analogy for patent-product image retrieval. Adv. Neural Inf. Process. Syst. 38, 14501–14523 (2026)

    2026

  71. [79]

    Mohammadi, A. et al. Lymphus: A multicenter open-access database of lymph node ultrasound images in patients with papillary thyroid carcinoma for clinical and artificial intelligence research.Data Brief 66, 112694, DOI: 10.1016/j.dib.2026.112694 (2026)

    2026

  72. [80]

    Diagnosis of metastatic lymph nodes in patients with papillary thyroid cancer: A comparative multi-center study of semantic features and deep learning-based models.J

    Abbasian Ardakani, A.et al. Diagnosis of metastatic lymph nodes in patients with papillary thyroid cancer: A comparative multi-center study of semantic features and deep learning-based models.J. Ultrasound Medicine 42, 1211–1221, DOI:10.1002/jum.16131 (2023)

    2023 doi

  73. [81]

    & Efros, A

    Torralba, A. & Efros, A. A. Unbiased look at dataset bias. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 1521–1528 (2011)

    2011

  74. [82]

    Siméoni, O. et al. Dinov3. arXiv preprint arXiv:2508.10104 (2025)

    2025 arXiv

  75. [83]

    Menon, A. K.et al. Long-tail learning via logit adjustment.arXiv preprint arXiv:2007.07314 (2020)

    2020 arXiv

  76. [84]

    Erickson, N. et al. Autogluon-tabular: Robust and accurate automl for structured data. arXiv preprint arXiv:2003.06505 (2020)

    2020 arXiv

  77. [85]

    Selvaraju, R. R.et al. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE International Conference on Computer Vision , 618–626 (2017)

    2017

  78. [86]

    Wunderling, T. et al. Comparison of thyroid segmentation techniques for 3d ultrasound. InMedical Imaging 2017: Image Processing , vol. 10133, 346–352 (2017)

    2017

  79. [87]

    Hou, X. et al. An ultrasonography of thyroid nodules dataset with pathological diagnosis annotation for deep learning. Sci. Data 11, 1272, DOI:10.1038/s41597-024-04156-5 (2024)

    2024 doi

  80. [88]

    Thyroid ultrasound cine-clip dataset, DOI:10.71718/7m5n-rh16 (2024)

    Stanford AIMI. Thyroid ultrasound cine-clip dataset, DOI:10.71718/7m5n-rh16 (2024)

    2024 doi

  81. [89]

    An annotated heterogeneous ultrasound database.Sci

    Yang, Y.et al. An annotated heterogeneous ultrasound database.Sci. Data 12, 148 (2025)

    2025

  82. [90]

    Chen, J. et al. Transunet: Rethinking the u-net architecture design for medical image segmentation through the lens of transformers.Med. Image Analysis 97, 103280, DOI:10.1016/j.media.2024.103280 (2024)

    2024

  83. [91]

    Zhang, S. et al. A generalist foundation model and database for open-world medical image segmentation.Nat. Biomed. Eng. 10, 1026–1041, DOI:10.1038/s41551-025-01497-3 (2026). Published online 5 September 2025

    2026 doi

  84. [92]

    & Sun, J

    He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 770–778 (2016)

    2016

  85. [93]

    & Ding, G

    Wang, A., Chen, H., Lin, Z., Han, J. & Ding, G. Lsnet: See large, focus small. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 9718–9729 (2025)

    2025

  86. [94]

    Qwen3-vl technical report.arXiv preprint arXiv:2511.21631 (2025)

    Bai, S., Cai, Y., Chen, R.et al. Qwen3-vl technical report.arXiv preprint arXiv:2511.21631 (2025). 50/51

    2025 arXiv

  87. [95]

    Gpt-4o system card.arXiv preprint arXiv:2410.21276 (2024)

    OpenAI. Gpt-4o system card.arXiv preprint arXiv:2410.21276 (2024)

    2024 arXiv

  88. [96]

    Qwen3 technical report.arXiv preprint arXiv:2505.09388 (2025)

    Yang, A., Li, A.et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388 (2025)

    2025 arXiv

  89. [97]

    Llava-med: Training a large language-and-vision assistant for biomedicine in one day

    Li, C., Wong, C.et al. Llava-med: Training a large language-and-vision assistant for biomedicine in one day. Adv. Neural Inf. Process. Syst. 36, 28541–28564 (2023). 51/51

    2023

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.