Pith. sign in

REVIEW 4 major objections 4 minor 40 references

BUSTR: Descriptor-Aware Vision-Language Learning for Breast Ultrasound Report Generation

T0 review · 4 major / 4 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read BUSTR trains a vision–language model to write breast ultrasound reports from structured lesion descriptors alone, without paired image–report data, and reports consistent gains over five baselines on two public datasets.

desk verdict BUSTR is a solidly engineered methods paper whose central claim is undercut by a closed evaluation loop: the reference reports are generated from the same descriptors used for training and metric parsing, so the gains show template mimicry, not clinical report quality. read the letter →

arxiv 2511.20956 v2 pith:ZE5XSW3Q submitted 2025-11-26 cs.CV cs.AIcs.LG

Rawa Mohammed , Mina Attin , Laxmi Gewali , Bryar Shareef This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords breastultrasoundreportgenerationBI-RADSvision-languagemodelmultitasklearninglargelanguagemodelsradiomicslow-resourcemedicalimaging
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

Breast ultrasound (BUS) report generation is bottlenecked by the scarcity of paired image–radiologist-report datasets. This paper's central claim is that structured lesion descriptors — BI-RADS category, shape, margin, echogenicity, posterior features, pathology, histology, plus radiomics from lesion masks — can substitute for that missing supervision. BUSTR first has a frozen large language model rewrite descriptor–value pairs into natural-language 'reports,' then trains a multi-head Swin vision encoder to predict all descriptors, and finally fine-tunes the encoder while keeping the language model frozen, using a dual loss of token cross-entropy and cosine alignment. On the BrEaST and BUS-BRA datasets the method outperforms five existing baselines across standard text-similarity metrics and recovers key clinical facts (BI-RADS, pathology, margin) with higher precision/sensitivity/F1 at inference time, using only the image. A sympathetic reader would care because, if true, any annotation-rich BUS dataset — not just ones with narrative reports — becomes usable for training interpretable report generators.

What carries the argument

Three components carry the argument: (1) zero-shot supervisory report construction, where a frozen LLM formats descriptor–value pairs and radiomics features into a natural-language report; (2) a descriptor-aware multi-head Swin encoder trained with multitask losses over the per-dataset descriptor sets, including a bundled margin loss and a regression head for tumor size; and (3) a dual-level generation objective that combines token-level cross-entropy with a cosine-similarity alignment loss between the LLM's final hidden states and its input embeddings. The encoder forces image features to represent each BI-RADS descriptor explicitly, the synthetic reports supply the textual supervision, and

What would settle it

Obtain genuine radiologist-written reports for a subset of BrEaST or BUS-BRA images (e.g., 50–100 cases) and have clinicians rate BUSTR's generated reports (trained only on synthetic descriptors) against simple template-filled reports; if BUSTR is not rated more clinically accurate, complete, or safe than the templates, the claim that descriptor-derived synthetic supervision yields clinical-grade narrative reports is falsified.

Watch

Extended reading notes

Core claim

The paper claims that a report-generation model for breast ultrasound can be trained without a single paired image–report example. The trick is to manufacture supervision from structured descriptors: a frozen LLM converts each image's descriptor–value pairs (BI-RADS, shape, margin, echogenicity, posterior features, and for some datasets pathology/histology, plus radiomics derived from the lesion mask) into a natural-language report that restates exactly those facts and nothing more. A multi-head Swin Transformer is then trained, via per-descriptor classification heads and a regression head for tumor size, to predict all available descriptors; the shared visual tokens are projected into the L

Load-bearing premise

The load-bearing premise is that a narrative generated by a frozen LLM from descriptor–value pairs is a faithful proxy for a radiologist-written report, so that training and evaluating against it reflects clinically meaningful report quality.

Editorial extensions

If this is right

  • Public BUS datasets with structured annotations and masks — even without narrative reports — become usable training data for report generation.
  • Encoding descriptors explicitly during vision pretraining improves downstream recovery of clinically critical attributes (BI-RADS, pathology, margin) compared with generic pretraining.
  • The cosine-alignment loss offers a recipe for keeping frozen LLMs grounded in visual inputs when paired data is scarce.
  • Radiologists would receive narrative reports that connect lesion descriptors to the final BI-RADS assessment, potentially improving interpretability over score-only outputs.

Reading between the lines

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

  • A testable extension: on a small set of real radiologist-written reports for the same images, compare BUSTR's outputs against reports from a model trained on real pairs. If the synthetic-reports model matches or beats it on clinician ratings, descriptor-derived supervision is validated; if not, the proxy-reality assumption fails.
  • The alignment loss likely trades fluency for groundedness; varying the alignment weight across values (e.g., 0.2–0.8) would reveal whether hallucination rate drops monotonically — a knob the paper does not turn.
  • Because BUSTR inserts predicted tumor size from a regression head, an immediate extension is to evaluate size estimation as an additional clinical-efficacy metric; the paper reports no size F1, so the apparent success on other descriptors may not carry over to continuous attributes.
  • The same descriptor-to-report construction could be applied to multi-site datasets with partially overlapping annotation sets, letting a single model exploit all available labels — a direction the paper's architecture enables but does not demonstrate.
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

4 major / 4 minor

Summary. The paper proposes BUSTR, a framework for breast ultrasound (BUS) report generation that does not require paired image–report data. Synthetic supervisory reports are created by feeding structured descriptors (BI-RADS, shape, margin, echogenicity, posterior features, pathology, histology) and radiomics features through a formatting function and a frozen LLaMA model (Eqs. 1–2). A Swin Transformer vision encoder is trained with multitask descriptor classification, and report generation is trained with a dual-level loss combining token-level cross-entropy and cosine alignment between input and output representations (Eqs. 8–10). The method is evaluated on BrEaST and BUS-BRA using NLG metrics (BLEU, ROUGE-L, METEOR, CIDEr) and clinical efficacy metrics (precision, sensitivity, F1 for descriptor recovery). The authors report consistent improvements over five baselines on both datasets, with significance tests on some NLG metrics, and an ablation study attributing gains to both the descriptor-aware encoder and the dual-level loss.

Significance. If the central claim were established, the work would be a useful step toward automated BUS reporting in low-resource settings, where only structured annotations and masks are available. The paper has several strengths: it is transparent about the lack of real narrative reports and reader studies; it reports statistical significance tests; it includes an ablation study isolating the contributions of the vision encoder and the loss; and it promises public code. However, the evaluation rests on synthetic reports that are deterministically derived from the same descriptors used for supervision. Consequently, the reported NLG and clinical efficacy numbers primarily measure how well the model reproduces a descriptor template, not whether the generated text is clinically meaningful or acceptable to radiologists. The significance of the headline claim is therefore conditional on external validation with human-written reports or a reader study.

major comments (4)
  1. [Sec. 2.2, Eqs. (1–2) and Sec. 3.3.1, Table 3] The NLG ground truth is the synthetic report R_gen = F_LLM(P), where P is built solely from structured descriptor–value pairs and radiomics. The same descriptors supervise the vision encoder (Sec. 2.3) and are restated in the training text. Thus BLEU, ROUGE-L, METEOR, and CIDEr scores measure agreement with a template paraphrase of the training labels, not agreement with any radiologist-written report. The paper acknowledges the absence of narrative reports (Sec. 4), but the abstract and conclusion still claim improvements in 'report similarity' and 'clinically meaningful' report generation. As written, this is a load-bearing overstatement. To support the claim, the authors need either independent, human-written reference reports or a reader study; otherwise the claims should be reframed as 'structured descriptor recovery' or 'template report generation.'
  2. [Sec. 3.3.2, Tables 4–5] The clinical efficacy metrics parse generated text and compare the extracted descriptor values to the very descriptors used to construct the synthetic training reports. Because the supervisory reports are constrained to restate those descriptor values, high precision/sensitivity/F1 partly reflect how well the language model copies the template. This is a closed evaluation loop: ground-truth descriptors → F_format + frozen LLaMA → synthetic report → CE metric compares back to the same descriptors. The results in Tables 4–5 are therefore not evidence of clinical efficacy in the sense of diagnostic utility; they are measures of factual consistency with the synthetic template. This limitation must be stated prominently wherever CE results are interpreted.
  3. [Sec. 3.4, Table 3] It is not specified whether all baselines (R2Gen, R2GenCMN, TSGET, R2GenGPT, Li) were trained on the same synthetic supervisory reports, with the same descriptor sets, the same five-fold splits, and the same preprocessing. If baselines were trained on different synthetic text or without the multitask descriptor supervision, the comparison in Table 3 is not an apples-to-apples evaluation of the proposed contributions. The authors should state the exact training protocol for each baseline, including the supervisory text and the descriptor labels used. This is necessary to interpret the claim that BUSTR 'consistently outperforms all baselines.'
  4. [Sec. 2.3–2.4 and Abstract] The claim 'without requiring paired image–report supervision' is misleading as stated. The method uses paired image–descriptor supervision to train the vision encoder and then generates synthetic reports from those same descriptors, training the language model against them. It does not use paired image–narrative-report supervision, which is true and important, but it does use a form of paired supervision that is tightly coupled to the evaluation. The distinction should be made explicit in the abstract and introduction to avoid implying that no image–text pairs are used in training.
minor comments (4)
  1. [Sec. 3.5] The paired t-tests compare BUSTR to only one baseline per dataset (the strongest CIDEr baseline). For a complete picture, either report significance against all baselines or state that the selected comparison is the most competitive. Multiple-testing correction should be considered if many comparisons are made.
  2. [Tables 4 and 5] Use 'BUSTR' consistently instead of 'Ours' in the table headers. Also, some numbers in Table 4 appear to have typographical issues (e.g., missing spaces, overlapping values such as '0.6020.689'); these should be cleaned.
  3. [Eq. (9)] Please clarify whether the cosine alignment loss is computed over all L positions, including vision tokens and prompt tokens, or only over report tokens. Define precisely how H and Z are aligned in length and what masking is applied.
  4. [General] The two abstracts in the paper (one in the arXiv metadata and one at the start of the full text) appear to be near-duplicates but are not identical. Please unify them and ensure the title formatting is correct (spacing in 'BREASTULTRASOUNDTEXTREPORTING').

Circularity Check

2 steps flagged · score 6.0 of 10

Synthetic-report evaluation loop: NLG ground truth is generated from the same descriptors that supervise the vision encoder, and CE metrics check recovery of exactly those descriptors; the narrative-report claim is not independently validated.

  1. self definitional [Sec. 2.2 (Eq. 1–2); Sec. 3.3.1]
    "In the absence of narrative radiology reports for breast ultrasound (BUS) images, we utilize a Llama-based model [30] to generate reports from structured, fact-based BUS descriptors... In what follows, we refer to these automatically generated texts simply as reports."

    Equations 1–2 define the NLG ground truth as Rgen = F_LLM(P), where P = F_format(ŷ_descriptors, R), i.e., a template paraphrase of the exact descriptor–value pairs used to supervise the vision heads (Eq. 4–6) and to construct the training text (Eq. 7–8). Section 3.3.1 then measures generated reports against this same synthetic text using BLEU/ROUGE/METEOR/CIDEr. Thus the NLG scores measure fidelity to the descriptor template, not independent clinical narrative quality; the evaluation loop is closed by construction.

  2. fitted input called prediction [Sec. 3.3.2 (CE metrics), with Eq. 1 and Eqs. 3–6]
    "These metrics quantify how accurately the generated reports recover the underlying structured facts and, together with qualitative examples (e.g., Fig. 2), provide a simple form of post-hoc interpretability by verifying whether the generated text correctly recovers key lesion descriptors and BI-RADS categories."

    The 'underlying structured facts' are the descriptors ŷ_descriptors that were the input to Eq. 1 when constructing Rgen, and the same values are the targets of the multitask vision loss (Eq. 4–6). CE therefore checks whether the generated text restates the exact labels used to build the reference reports and to train the vision encoder. This is a closed descriptor-recovery check: it confirms template fidelity, but it does not validate narrative quality against human-written reports, a limitation the authors concede in Sec. 4.

full rationale

The paper does not rely on load-bearing self-citation or imported uniqueness theorems; the self-citations present are to the authors' earlier segmentation/classification work and are not used to justify BUSTR's core mechanism. The central circularity is evaluative: because no paired radiologist-written reports exist, the authors synthesize reports from structured descriptors (Eq. 1–2) and then use those same synthetic reports as the NLG ground truth (Sec. 3.3.1), while CE metrics (Sec. 3.3.2) check recovery of the very descriptor values fed into Eq. 1 and supervised in Eq. 4–6. This makes the headline 'report generation' claim partially circular: the numbers quantify how well the model reproduces a deterministic descriptor template, not whether the output is clinically meaningful narrative. The image-to-descriptor classification component itself is independently meaningful (test descriptors are real annotations and are held out), so the circularity is not total. The authors explicitly acknowledge the missing external validation: 'neither dataset provides real narrative radiology reports... there are no reader studies' (Sec. 4). Under the rubric, this warrants a score of 6: one or more 'predictions' (narrative report quality) reduce by construction to reproducing the inputs used to build the references, but the framework still contains an independent image-understanding task.

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

The central claim rests on five hand-chosen design elements (loss weights, margin-loss weighting, combination-function selection, size normalization, prompt template) plus domain assumptions about LLaMA's report quality, the BI-RADS-as-composite hypothesis, pretrained weights, and the multitask averaging scheme. No new physical or conceptual entities are introduced; the 'descriptor-aware multi-head encoder' is a method component, not an invented entity.

free parameters (5)
  • dual loss weights λ_CE, λ_Align = λ_CE = λ_Align = 0.5
    Eq. (10): the final objective is a weighted sum of token CE and cosine alignment; the equal-weight choice is asserted without search, and the combination function F (CE+Cos, CE*Cos, mean, max) was then selected by test performance in Table 6.
  • margin combined-loss weights = 0.5 main + 0.5 mean of 4 subtype losses
    Eq. (5): equal weighting of the main margin class and the mean of the four non-circumscribed subtype losses, stated to 'match the implementation'; no justification given.
  • combination function F for dual loss = mean(CE, Cos)
    Table 6: the mean was adopted as default because it yields the best NLG scores on the test folds — a post-hoc selection of a design choice on the evaluation data.
  • tumor size normalization = scaled by maximum value during training
    Sec. 2.3: 'we scale tumor size by the maximum value during training, consistent with the implementation' — a hand-chosen preprocessing choice that affects the size regression head.
  • instruction-style prompt template F_format = exact text not given
    Eq. (1): the formatting function converts descriptors and radiomics into the prompt P for the frozen LLM; the exact template is not specified yet it determines report phrasing and thus all NLG scores.
assumptions (5)
  • domain assumption A frozen LLM (Llama 2) can turn descriptor–value pairs into narrative reports whose text is suitable as ground truth for training a report generator.
    Sec. 2.2, Eq. (2): Rgen = F_LLM(P). The entire supervision pipeline rests on the quality and suitability of LLaMA-generated restatements.
  • domain assumption BI-RADS category can be predicted from the concatenation of shape/margin/posterior/echo hidden states.
    Sec. 2.3: the descriptor-aware BI-RADS head takes concat(h_shape, h_margin, h_posterior, h_echo), 'mirroring the ACR guidelines' — this equates BI-RADS to a function of the other four descriptors and is a modeling hypothesis.
  • domain assumption Pre-trained Swin Transformer and LLaMA-2 weights provide valid starting points, with the frozen LLM contributing the narrative style.
    Sec. 2.3–2.4: relies on external pretrained models whose behavior and biases are taken as given; no analysis of style transfer from LLM pretraining to BUS reporting.
  • domain assumption The multitask vision loss as a simple average over active tasks is a valid joint objective.
    Eq. (6): equal-weight averaging across descriptor tasks is asserted without analysis of task conflict, dominance, or class imbalance (BrEaST has 15 anechoic vs 148 hypoechoic cases).
  • domain assumption Synthetic reports 'should not introduce any new findings beyond what is present in the structured fields'.
    Sec. 2.2: the design principle assumes the frozen LLM obeys the constraint and does not hallucinate or paraphrase beyond the descriptor values.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BUSTR: Descriptor-Aware Vision-Language Learning for Breast Ultrasound Report Generation." pith.science (2026). https://pith.science/paper/ZE5XSW3Q

@misc{pith2026251120956,
  author       = {Pith},
  title        = {Pith review of: BUSTR: Descriptor-Aware Vision-Language Learning for Breast Ultrasound Report Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZE5XSW3Q}},
  note         = {Machine review of arXiv:2511.20956}
}
read the original abstract

Breast ultrasound (BUS) reporting relies on clinically meaningful lesion descriptors, including BI-RADS category, lesion shape, margin, echogenicity, posterior features, pathology, and histology. However, many public BUS datasets provide structured annotations and lesion masks without paired radiologist-written reports, limiting the development of vision--language models for BUS report generation. We propose BUSTR, a descriptor-aware vision--language framework that uses structured lesion information to enable report generation under limited report supervision. BUSTR first constructs descriptor-derived reports from available annotations and radiomics features extracted from lesion masks. It then trains a multi-head Swin Transformer encoder with multitask supervision to learn descriptor-aware visual representations across datasets with partially overlapping annotation sets. The projected visual tokens condition a frozen LLaMA-based language model, and training is guided by a dual-level objective combining token-level cross-entropy with representation-level cosine alignment. At inference, BUSTR generates reports from BUS images without access to structured descriptors, lesion masks, or radiomics features. We evaluate BUSTR on the public BrEaST and BUS-BRA datasets using natural language generation and clinical efficacy metrics. BUSTR improves report similarity and descriptor recovery compared with representative report-generation baselines, with notable gains for lesion shape, margin, posterior features, and pathology, as well as improved BI-RADS sensitivity and F1-score on BrEaST. These results suggest that structured BUS descriptors, lesion masks, and radiomics features can provide useful supervision for descriptor-aware BUS report generation when paired radiologist-written reports are unavailable.

Figures

Figures reproduced from arXiv: 2511.20956 by the authors.

Figure 1
Figure 1. Overall BUSTR architecture. (a) Zero-shot construction of reports from BUS descriptors and radiomics [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Qualitative comparison between top-performing models. Different colors are used to show each predicted [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 4 linked inside Pith

  1. [1]

    Thomas M Kolb, Jacob Lichy, and Jeffrey H Newhouse. Comparison of the performance of screening mammog- raphy, physical examination, and breast us and evaluation of factors that influence them: an analysis of 27,825 patient evaluations.Radiology, 225(1):165–175, 2002

    2002

  2. [2]

    The complexities of physician supply and demand: Projections from 2015 to 2030

    IHS Markit. The complexities of physician supply and demand: Projections from 2015 to 2030. Technical report, 2017. 11 Breast Ultrasound Text Reporting

    2015

  3. [3]

    Challenge-aware u-net for breast lesion segmentation in ultrasound images.Pattern Recognition, page 111851, 2025

    Dengdi Sun, Changxu Dong, Yuchen Yan, Bo Jiang, Yayang Duan, Zhengzheng Tu, and Chaoxue Zhang. Challenge-aware u-net for breast lesion segmentation in ultrasound images.Pattern Recognition, page 111851, 2025

    2025

  4. [4]

    Freer, and Min Xian

    Bryar Shareef, Aleksandar Vakanski, Phoebe E. Freer, and Min Xian. Estan: Enhanced small tumor-aware network for breast ultrasound image segmentation.Healthcare, 10(11):2262, 2022

    2022

  5. [5]

    TR Mahesh, Surbhi Bhatia Khan, Kritika Kumari Mishra, Saeed Alzahrani, and Mohammed Alojail. Enhancing diagnostic precision in breast cancer classification through efficientnetb7 using advanced image augmentation and interpretation techniques.International Journal of Imaging Systems and Technology, 35(1):e70000, 2025

    2025

  6. [6]

    Zhang, M

    Y . Zhang, M. Xian, H.D. Cheng, B. Shareef, J. Ding, F. Xu, K. Huang, B. Zhang, C. Ning, and Y . Wang. Busis: A benchmark for breast ultrasound image segmentation.Healthcare (Basel), 10(4):729, 2022

    2022

  7. [7]

    Bcs-net: Multi-task breast cancer screening network enhanced by multi-modality attention

    Ruili Li, Ruiyu Li, Eichi Takaya, Zizhen Lin, Tomoya Kobayashi, Nanako Mtsuda, and Takuya Ueda. Bcs-net: Multi-task breast cancer screening network enhanced by multi-modality attention. InICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5, 2025

    2025

  8. [8]

    An alignment and imputation network (ainet) for breast cancer diagnosis with multimodal multi-view ultrasound images.IEEE Transactions on Medical Imaging, pages 1–1, 2025

    Haoyuan Chen, Yonghao Li, Jiadong Zhang, Long Yang, Yiqun Sun, Yaling Chen, Shichong Zhou, Zhenhui Li, Xuejun Qian, Qi Xu, and Dinggang Shen. An alignment and imputation network (ainet) for breast cancer diagnosis with multimodal multi-view ultrasound images.IEEE Transactions on Medical Imaging, pages 1–1, 2025

    2025

Show all 40 references
  1. [9]

    Breast ultrasound tumor classification using a hybrid multitask cnn–transformer network

    Bryar Shareef, Min Xian, Aleksandar Vakanski, and Haotian Wang. Breast ultrasound tumor classification using a hybrid multitask cnn–transformer network. InMedical Image Computing and Computer-Assisted Intervention (MICCAI), 2023

    2023

  2. [10]

    An svm confidence-based approach to medical image annotation

    Tatiana Tommasi, Francesco Orabona, and Barbara Caputo. An svm confidence-based approach to medical image annotation. InWorkshop of the Cross-Language Evaluation Forum for European Languages, pages 696–703. Springer, 2008

    2008

  3. [11]

    Query and document expansion with medical subject headings terms at medical imageclef 2008

    Julien Gobeill, Patrick Ruch, and Xin Zhou. Query and document expansion with medical subject headings terms at medical imageclef 2008. InEvaluating Systems for Multilingual and Multimodal Information Access: 9th Workshop of the Cross-Language Evaluation Forum, CLEF 2008, Aarh...

    2008

  4. [12]

    Attention based automated radiology report generation using cnn and lstm.Plos one, 17(1):e0262209, 2022

    Mehreen Sirshar, Muhammad Faheem Khalil Paracha, Muhammad Usman Akram, Norah Saleh Alghamdi, Syeda Zainab Yousuf Zaidi, and Tatheer Fatima. Attention based automated radiology report generation using cnn and lstm.Plos one, 17(1):e0262209, 2022

    2022

  5. [13]

    A survey of deep learning-based radiology report generation using multimodal data.arXiv preprint arXiv:2405.12833, 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 data.arXiv preprint arXiv:2405.12833, 2025

    2025 arXiv

  6. [14]

    METransformer: Radiology report generation by transformer with multiple learnable expert tokens

    Zhanyu Wang, Lingqiao Liu, Lei Wang, and Luping Zhou. METransformer: Radiology report generation by transformer with multiple learnable expert tokens. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16302–16312, 2023

    2023

  7. [15]

    and Liu, L

    Wang, Z. and Liu, L. and Wang, L. and Zhou, L. R2gengpt: Radiology report generation with frozen llms. Meta-Radiology, 1(3):100033, 2023

    2023

  8. [16]

    Evaluating gpt-4 on impressions generation in radiology reports.Radiology, 307(5):e231259, 2023

    Zhaoyi Sun, Hanley Ong, Patrick Kennedy, Liyan Tang, Shirley Chen, Jonathan Elias, Eugene Lucas, George Shih, and Yifan Peng. Evaluating gpt-4 on impressions generation in radiology reports.Radiology, 307(5):e231259, 2023

    2023

  9. [17]

    Gemini goes to med school: exploring the capabilities of multimodal large language models on medical challenge problems & hallucinations

    Ankit Pal and Malaikannan Sankarasubbu. Gemini goes to med school: exploring the capabilities of multimodal large language models on medical challenge problems & hallucinations. InProceedings of the 6th Clinical Natural Language Processing Workshop, pages 21–46, 2024

    2024

  10. [18]

    Improving radiology report generation systems by removing hallucinated references to non-existent priors

    Vignav Ramesh, Nathan A Chi, and Pranav Rajpurkar. Improving radiology report generation systems by removing hallucinated references to non-existent priors. InMachine Learning for Health, pages 456–473. PMLR, 2022

    2022

  11. [19]

    RadAlign: Advancing radiology report generation with vision-language concept alignment.arXiv preprint arXiv:2501.07525, 2025

    Difei Gu, Yunhe Gao, Yang Zhou, Mu Zhou, and Dimitris Metaxas. RadAlign: Advancing radiology report generation with vision-language concept alignment.arXiv preprint arXiv:2501.07525, 2025

    2025 arXiv

  12. [20]

    Improving radiology report generation quality and diversity through reinforcement learning and text augmentation.Bioengineering, 11(4):351, 2024

    Daniel Parres, Alberto Albiol, and Roberto Paredes. Improving radiology report generation quality and diversity through reinforcement learning and text augmentation.Bioengineering, 11(4):351, 2024

    2024

  13. [21]

    Xbusnet: Text-guided breast ultrasound segmentation via multimodal vision– language learning.Diagnostics, 15(22):2849, 2025

    Raja Mallina and Bryar Shareef. Xbusnet: Text-guided breast ultrasound segmentation via multimodal vision– language learning.Diagnostics, 15(22):2849, 2025. 12 Breast Ultrasound Text Reporting

    2025

  14. [22]

    Boyu Zhang, Aleksandar Vakanski, and Min Xian. Bi-rads-net-v2: a composite multi-task neural network for computer-aided diagnosis of breast cancer in ultrasound images with semantic and quantitative explanations.IEEE Access, 11:79480–79494, 2023

    2023

  15. [23]

    Margolies, Gaurav Pandey, Eliot R

    Laurie R. Margolies, Gaurav Pandey, Eliot R. Horowitz, and David S. Mendelson. Breast imaging in the era of big data: Structured reporting and data mining.AJR American Journal of Roentgenology, 206(2):259–264, 2016

    2016

  16. [24]

    Ai-assisted method for efficiently generating breast ultrasound screening reports.Current Medical Imaging, 19(2):149–157, 2023

    Shuang Ge, Qiongyu Ye, Wenquan Xie, Desheng Sun, Huabin Zhang, Xiaobo Zhou, and Kehong Yuan. Ai-assisted method for efficiently generating breast ultrasound screening reports.Current Medical Imaging, 19(2):149–157, 2023

    2023

  17. [25]

    Ultrasound report generation with cross-modality feature alignment via unsupervised guidance.IEEE Transactions on Medical Imaging, 2024

    Jun Li, Tongkun Su, Baoliang Zhao, Faqin Lv, Qiong Wang, Nassir Navab, Ying Hu, and Zhongliang Jiang. Ultrasound report generation with cross-modality feature alignment via unsupervised guidance.IEEE Transactions on Medical Imaging, 2024

    2024

  18. [26]

    Chung-Ming Lo and Hui-Ru Chen. Automated breast imaging report generation based on the integration of multiple image features in a metadata format for shared decision-making.Health informatics journal, 30(3):14604582241288460, 2024

    2024

  19. [27]

    Computer-aided diagnosis system for breast ultrasound reports generation and classification method based on deep learning.Applied Sciences, 13(11):6577, 2023

    Haojun Qin, Lei Zhang, and Quan Guo. Computer-aided diagnosis system for breast ultrasound reports generation and classification method based on deep learning.Applied Sciences, 13(11):6577, 2023

    2023

  20. [28]

    Ai-powered synthesis of structured multimodal breast ultrasound reports integrating radiologist annotations and deep learning analysis

    Khadija Azhar, Byoung-Dai Lee, Shi Sub Byon, Kyu Ran Cho, and Sung Eun Song. Ai-powered synthesis of structured multimodal breast ultrasound reports integrating radiologist annotations and deep learning analysis. Bioengineering, 11(9):890, 2024

    2024

  21. [29]

    Breast ultrasound report generation using LangChain.arXiv preprint arXiv:2312.03013, 2023

    Jaeyoung Huh, Hyun Jeong Park, and Jong Chul Ye. Breast ultrasound report generation using LangChain.arXiv preprint arXiv:2312.03013, 2023

    2023 arXiv

  22. [30]

    Llama 2: Open foundation and fine-tuned chat models

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  23. [31]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. InProceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021

    2021

  24. [32]

    Curated benchmark dataset for ultrasound based breast lesion analysis.Scientific Data, 11(1):148, 2024

    Anna Pawłowska, Anna ´Cwierz-Pie´nkowska, Agnieszka Domalik, Dominika Jagu ´s, Piotr Kasprzak, Rafał Matkowski, Łukasz Fura, Andrzej Nowicki, and Norbert ˙Zołek. Curated benchmark dataset for ultrasound based breast lesion analysis.Scientific Data, 11(1):148, 2024

    2024

  25. [33]

    Bus-bra: a breast ultrasound dataset for assessing computer-aided diagnosis systems.Medical Physics, 51(4):3110–3123, 2024

    Wilfrido Gómez-Flores, Maria Julia Gregorio-Calas, and Wagner Coelho de Albuquerque Pereira. Bus-bra: a breast ultrasound dataset for assessing computer-aided diagnosis systems.Medical Physics, 51(4):3110–3123, 2024

    2024

  26. [34]

    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

  27. [35]

    Rouge: A package for automatic evaluation of summaries

    Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText summarization branches out, pages 74–81, 2004

    2004

  28. [36]

    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

  29. [37]

    Cider: Consensus-based image description evaluation

    Ramakrishna Vedantam, C Lawrence Zitnick, and Devi Parikh. Cider: Consensus-based image description evaluation. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4566–4575, 2015

    2015

  30. [38]

    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

  31. [39]

    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...

    2021

  32. [40]

    Tsget: Two-stage global enhanced transformer for automatic radiology report generation.IEEE Journal of Biomedical and Health Informatics, 28(4):2152–2162, 2024

    Xiulong Yi, You Fu, Ruiqing Liu, Hao Zhang, and Rong Hua. Tsget: Two-stage global enhanced transformer for automatic radiology report generation.IEEE Journal of Biomedical and Health Informatics, 28(4):2152–2162, 2024. 13

    2024

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.