Breast ultrasound (BUS) is an essential tool for diagnosing breast lesions, with millions of examinations per year. However, publicly available high-quality BUS benchmarks for AI development are limited in data scale and annotation richness. In this work, we present BUS-CoT, a BUS dataset for chain-of-thought (CoT) reasoning analysis, which contains 11,439 ultrasound images from 11,850 lesions and 4,838 patients, covering all 99 WHO-defined histopathology categories. For model training and evaluation, we provide a curated high-quality subset of 5,163 lesion-focused images annotated by experienced radiologists. To facilitate research on incentivizing CoT reasoning, we construct the reasoning processes based on observation, feature, diagnosis and pathology labels, annotated and verified by experienced experts. Moreover, by covering lesions of all histopathology types, we aim to facilitate robust AI systems in rare cases, which can be error-prone in clinical practice. The data and code are publicly available at https://doi.org/10.6084/m9.figshare.30838715.

Objective To evaluate the application value of the Peyton four-step teaching method in the standardized training of intestinal ultrasound and compare it with traditional teaching methods,so as to provide an optimized approach for clinical ultrasound training.Methods Participants from the Department of Ultra-sound at Peking Union Medical College Hospital between September 2024 and March 2025 were randomly as-signed to either the traditional group or Peyton group.The traditional group followed the conventional"lecture-demonstration-practice"model,while the Peyton group implemented the standardized"demonstration-decon-struction-comprehension-execution"four-step approach.All training focused on standard intestinal ultrasound scanning techniques.After the training,the operational skills were independently evaluated by the instructors.To verify the reproducibility of the teaching method,the participants in traditional teaching group received addi-tional Peyton method training after the initial assessment and underwent a second evaluation.Results A total of 18 participants were included in this study,with 9 in the traditional teaching group and 9 in the Peyton teach-ing group.Participants in the Peyton group demonstrated significantly higher scores than those in the traditional group at every anatomical site assessed(all P＜0.05),including the terminal ileum[7.3(6.5,8)vs.(3.1±1.4)],ileocecal region[7(6,8)vs.(3.7±1.3)],appendix[8(7,8.3)vs.(5.3±0.8)],colon[7(6.5,7.7)vs.5.3(2.8,5.3)]and small intestine[(7.3±0.5)vs.(3.7±1.2)].The overall score was also significantly higher in the Peyton group[(7.2±0.9)vs.(3.7±1.0),P＜0.001].After receiving supplemental Peyton method training,the traditional group showed significant improvement in all anatomical site scores,with the overall score increasing to[(7.5±0.4),P=0.008].Conclusions The Peyton four-step method is significantly more effective than traditional teaching in improving residents'intestinal ultrasound skills,demonstrating its suitability as the preferred approach for standardized training programs.

OBJECTIVE:Ultrasound-based viscoelastic imaging enables the real-time characterization of tissue properties. In this study, we aimed to investigate the correlation between viscoelastic mechanical properties and tumor proliferation in invasive ductal breast cancer (IDC). METHODS:This prospective study consecutively enrolled patients with IDC during October 2024 and January 2025. All patients underwent preoperative shear-wave elastography (SWE) and ultrasound-based viscoelastic imaging. The viscoelastic mechanical properties (viscosity coefficients [Vmax and Vmean] and dispersion coefficients [Dmax and Dmean]) and SWE (Emax) were measured within the tumor and in a 1-mm peritumoral region (Vtmax, Vtmean, et al.). Patients were stratified into high (Ki-67 ≥ 14%) and low (Ki-67 < 14%) proliferation groups. Viscoelastic mechanical properties and SWE between groups were compared. RESULTS:Among the 124 female patients with IDC, 94 exhibited high Ki-67 expression while 30 had low Ki-67 expression. No significant differences were observed in SWE-based elasticity between the two groups (p > 0.05). Vmean and Vtmean were lower in the high Ki-67 group (1.1 [0.7-1.7] vs. 1.5 [1.0-2.3], p = 0.020; 1.3 [0.8-1.8] vs. 1.7 [1.1-2.4], p = 0.026, respectively). Using Vmean ≤ 1.2 Pa·s as cutoff value, a significantly higher proportion of high Ki-67 expression was found (64.9% vs. 35.1%, p = 0.016). CONCLUSION:Viscoelastic mechanical properties, rather than traditional SWE elasticity, correlated with Ki-67 expression in IDC. A lower Vmean may be correlated with higher proliferative activity, potentially serving as a non-invasive imaging biomarker. These findings support further exploration of ultrasound-based viscoelastic imaging for the assessment of tumor microenvironment heterogeneity.

Foundation models have emerged as powerful tools for addressing various tasks in clinical settings. However, their potential development for breast ultrasound analysis remains untapped. Here we present BUSGen, the first foundation generative model designed for breast ultrasound image analysis. Pretrained on over 3.5 million breast ultrasound images, BUSGen has acquired extensive knowledge of breast structures, pathological features and clinical variations. With few-shot adaptation, BUSGen can generate repositories of realistic and informative task-specific data, facilitating the development of models for a wide range of downstream tasks. Extensive experiments highlight BUSGen's exceptional adaptability, significantly exceeding real-data-trained foundation models in breast cancer screening, diagnosis and prognosis. In breast cancer early diagnosis, our approach outperformed all board-certified radiologists (n = 9), achieving an average sensitivity improvement of 16.5% (P < 0.0001). In addition, we characterized the scaling effect of using synthetic data. Finally, BUSGen enabled de-identified data sharing, making progress forward in secure medical data utilization.

Axillary lymph node (ALN) metastasis is a critical factor influencing prognosis and treatment strategies in breast cancer patients. However, traditional methodsu2014ranging from physical examination to ultrasoundu2014often lack the precision required for clinical decision-making. In recent years, ultrasound radiomics and deep learning have emerged as promising solutions, leveraging high-throughput quantitative features from ultrasound images to enhance detection accuracy. This review explores the development and application of radiomics and deep learning across multiple ultrasound modalities (grayscale, elastography, and contrast-enhanced ultrasound), as well as in multimodal imaging approaches that integrate ultrasound with MRI and PET/CT, underscoring the benefits of incorporating clinicopathological variables to boost predictive performance. These studies provide a vital foundation for personalized treatment and precision medicine in breast cancer management.

To determine a direct method for diagnosing axillary lymph node (ALN) tumor burden preoperatively in cT1-2N0 breast cancer patients, we developed and validated a deep learning (DL) model based on ultrasound (US) images of sentinel lymph nodes (SLNs) detected by contrast-enhanced lymphatic ultrasound (CEUS). Women with cT1-T2N0 breast cancer who received CEUS were enrolled prospectively from Peking Union Medical College Hospital between April 2020 and July 2021 and from Sichuan Cancer Hospital between April 2022 and July 2022. Heavy ALN tumor burden was defined as > 2 metastatic lymph nodes according to the Z0011 criteria. We developed a DL model, the modality-adaptive network with clinicopathological information (MAN + C), using grayscale or color Doppler US images and radioclinicopathological information to predict heavy tumor burden. A total of 595 SLNs from 374 patients met the inclusion criteria. The areas under the receiver operating characteristic curve (AUCs) were calculated to evaluate the predictive performance of the model, yielding values of 0.91[95

The 70-gene signature (70-GS; MammaPrint) assay is useful for prognosis assessment in HR+/HER2- early breast cancer, but limited accessibility motivates development of noninvasive alternatives. We retrospectively enrolled 219 women with preoperative grayscale ultrasound and 70-GS results, including a development cohort (n = 125), an internal validation cohort (n = 53), and a temporally independent validation cohort (n = 41). Radiomic features were extracted from manually delineated ROIs using PyRadiomics, and a radiomics score was derived after LASSO selection. Candidate radiomics-only, clinicopathologic-only, and full clinicoradiomic models were explored. To reduce overfitting, we selected a parsimonious model combining the radiomics score and Ki67 as the primary model. The simplified model achieved AUCs of 0.878, 0.816, and 0.831 in the development, internal validation, and temporally independent validation cohorts, respectively. In 1000 bootstrap resamples, the optimism-corrected AUC was 0.872 and the corrected calibration slope was 0.953. Adding the radiomics score to a Ki67-only model significantly improved model fit (likelihood-ratio chi-square = 17.14, df = 1, p < 0.001). An ultrasound radiomics and Ki67 model may provide a noninvasive reference for estimating MammaPrint risk categorization, but it should be considered only as a supportive adjunct and not as a replacement for genomic testing.

The rapid advancement of large language models (LLMs) has unlocked transformative potential for role-playing emotional companion products, enabling systems that support emotional well-being, educational development, and therapeutic applications. However, existing approaches often lack sustained personalization and contextual adaptability, limiting their effectiveness in real-world settings. In this paper, we introduce iPET, an LLM-powered virtual pet agent designed to enhance user engagement through rich, dynamic pet behaviors and interactions tailored to individual preferences. iPET comprises three core components: a dialogue module that instantiates virtual pet agents for emotionally interactive conversations; a memory module that stores and synthesizes records of both agent and user experiences; and a world simulation module that generates diverse, preference-driven pet behaviors guided by high-level reflections. Deployed for over 200 days in a real-world, non-commercial product, iPET has served millions of users – providing emotional support to psychologically distressed individuals and demonstrating its effectiveness in practical applications.

This study aimed to develop a new ultrasonographic dating formula to estimate gestational age (GA) based on fetal crown–rump length (CRL) in a Chinese population, evaluate model accuracy and compare its performance with established dating formulas. A prospective, multicenter study was conducted across mainland China. Participants included healthy, low-risk women with spontaneously conceived singleton pregnancies and a regular menstrual cycle in the preceding year. Ultrasonography was performed between 11 and 14 weeks of gestation, with GA determined based on the last menstrual period. Participants were randomly assigned to a development or validation cohort in a 7:3 ratio. A best-fit regression model was constructed for GA estimation based on CRL in the development cohort. For validation, mean differences between the new estimated GA and menstrual age were calculated and compared with those obtained using five established CRL-based dating formulas in the validation cohort. All participants were followed through to delivery. The study recruited 4,710 women with singleton pregnancies, with 3,297 in the development cohort and 1,413 women in the validation cohort. The mean and standard deviation values of CRL changed linearly with GA during 11–14 weeks. CRL demonstrated a linear relationship with GA between 11 and 14 weeks, yielding the regression equation GA = 59.590085 + 0.458539×CRL (R2 = 0.8042). The mean difference between estimated GA and menstrual age was 0.32 days (95

Background:The accurate intraoperative localization of lung cancers presenting as ground-glass opacities (GGOs) in lung tissue remains challenging. Ultrahigh-frequency ultrasound (UHFUS), which allows for the visualization of micron-scale structures, may help address this issue. This study aimed to evaluate whether UHFUS can serve as a reliable intraoperative tool for real-time detection and localization of GGOs in excised lung specimens. Methods:This prospective observational study included patients with suspected lung cancer who underwent surgery between June 2023 and March 2024. Each excised GGO was sequentially detected and localized intraoperatively via palpation and UHFUS (22-38 MHz). The UHFUS features were independently examined and evaluated by two radiologists. Comparisons between localization rate and time consumption were analyzed with the McNemar and Wilcoxon signed-rank tests. Results:In total, 36 patients (55±10 years; 9 males) comprising 58 GGOs were included, of which 50 were small (≤1 cm) and 8 were general (>1 cm) GGOs; when grouped by density, 37 were considered pure and 21 mixed GGOs. UHFUS, as compared to palpation, demonstrated a superior localization rate for both small nodules (UHFUS: n=48, 96.0%; palpation: n=40, 80.0%; P=0.02) and pure GGOs (UHFUS: n=33, 94.3%; palpation: n=25; 71.4%; P=0.02). Even in the micronodule subgroup (≤5 mm), UHFUS showed better localization ability (n=15, 100%) than did palpation (n=12, 80%). For small GGOs, the median localization time of UHFUS [5 s, interquartile range (IQR) 5-8 s] was significantly shorter (P=0.003) than that for palpation (5 s, IQR 5-15 s); this difference was more pronounced (P=0.004) in pure GGOs (UHFUS: median 5 s, IQR 5-10 s; palpation: 12.5 s, IQR 5-20 s). On UHFUS, 83.3% of GGOs appeared as indistinct hypoechoic areas with posterior shadowing. UHFUS could detect new GGOs, and the agreement in diameter between UHFUS and pathology surpassed that between CT and pathology. Conclusions:This prospective observational trial supports the use of real-time, noninvasive, and radiation-free UHFUS for the intraoperative localization of GGOs in lung tissue. It has the potential to enhance the efficiency of lung surgery in detecting small tumors and identify new nodules.

The rise of social media platforms has increased the demand for semantic-rich services, such as event and storyline attribution. However, most existing research focuses on clip-level event understanding, mainly through basic captioning tasks, without addressing the causal relationships between events across an entire movie. This presents a significant challenge, as even advanced multimodal large language models (MLLMs) struggle with extensive multimodal information due to limited context length. To tackle this, we propose a Two-Stage Prefix-Enhanced MLLM (TSPE) approach for event attribution, which connects events through their causal semantics in movie videos. In the local stage, we introduce an interaction-aware prefix to guide the model’s focus on relevant multimodal cues within a single clip, briefly summarizing each event. In the global stage, we enhance event connections using an inferential knowledge graph and design an event-aware prefix to focus on relevant events, not all preceding clips, leading to accurate event attribution. Extensive evaluations on two real-world datasets demonstrate that our framework outperforms state-of-the-art methods.

Unicentric Castleman disease (UCD) is a rare group of non-neoplastic lymphoproliferative disorders. This study aims to summarize the specific ultrasonic manifestations of UCD. This retrospective study included patients who underwent preoperative ultrasound for enlarged lymph nodes and were later diagnosed with UCD between January 2016 and March 2024. Ultrasound features, including lymph node size, cortical characteristics, corticomedullary interface, hyperechoic regions, and Doppler flow signals, were recorded. Pathological types were classified as hyaline vascular (HV), plasma cell (PC), or mixed. The ultrasonic features of each UCD subtype were systematically analyzed. A total of 41 patients were enrolled in the study, comprising 29 with HV-type, 4 with PC-type, and 8 with a mixed type. All patients presented with enlarged lymph nodes (LNs) characterized by a solitary mass, well-defined margins, and increased cortical thickness. Among these, 95.12

INTRODUCTION:Gastrointestinal ultrasound (GIUS) is recommended for monitoring Crohn's disease (CD). GIUS scores are used to quantify CD activity. Among them, International Bowel Ultrasound Segmental Activity Score (IBUS-SAS), Bowel Ultrasound Score (BUSS), Simple Ultrasound Score, and Simple Ultrasound Score for Crohn's Disease are most commonly used. The aim of this study was to compare and correlate the performance of such indicators with endoscopic activity and to calculate interobserver agreement. METHODS:Consecutive patients with CD at our hospital between June 2015 and July 2021 were retrospectively enrolled. All patients underwent ileocolonoscopy after medical treatment. GIUS was performed within 2 weeks, and 4 GIUS scores were independently calculated. Receiver operating characteristic curve analyses were used to determine a cutoff value. Cohen kappa (κ) coefficient was calculated to estimate the agreement between GIUS findings. RESULTS:A total of 106 patients with CD were enrolled. 80.2% (85/106) were endoscopic active (Simple Endoscopic Score for Crohn's disease ≥3), and 8.49% (9/106) were severe cases (Simple Endoscopic Score for Crohn's disease ≥9). All GIUS features (bowel wall thickness, color Doppler signs, bowel wall stratification, inflammatory signals at the mesentery) were statistically significant in assessing CD activity ( P < 0.05). IBUS-SAS showed the highest area under the curve (0.98; 95% CI: 0.96-1.00) and specificity (95.2%) for a cutoff value of 46.50. However, IBUS-SAS had only moderate agreement (Cohen κ = 0.427; P < 0.001). BUSS had substantial interobserver agreement (Cohen κ = 0.947; P < 0.001), with a similar diagnostic value (sensitivity, 100.0%; accuracy, 95.3%; area under the curve of 0.96 [95% CI: 0.91-1.00] for a cutoff value of 4.58). DISCUSSION:GIUS score is an efficient and reliable method to assess CD activity. BUSS achieved a high accuracy and excellent interobserver agreement, which is more suitable for treatment assessment.

Abstract Objectives Creeping fat (CF) is associated with stricture formation in Crohn’s disease (CD). This study evaluated the feasibility of intestinal ultrasound (IUS) for semiquantitative analysis of CF and compared the agreement between IUS and computed tomography enterography (CTE). Methods In this retrospective study, we recruited consecutive CD patients who underwent IUS and CTE. CF wrapping angle was analyzed on the most affected bowel segment and was independently evaluated by IUS and CTE. We evaluated the wrapping angle of CF in the cross- and vertical sections of the diseased bowel. CF wrapping angle was divided into < 180° and ≥ 180°. IUS performance was assessed using CTE as a reference standard, and IUS interobserver consistency was evaluated. Results We enrolled 96 patients. CTE showed that CF wrapping angle was < 180° in 35 patients and ≥ 180° in 61 patients. We excluded three cases in which the observation positions were inconsistent between the IUS and CTE. Excellent agreement was shown between US and CTE (82/93, 88.2%). The eleven remaining cases showed inconsistencies mostly in the terminal ileum (n = 5) and small intestine (n = 4). Total agreement between IUS observers was 89.6% (86/96, κ = 0.839, p = 0.000), with perfect agreement for the ileocecal and colonic segments (35/37, 94.6% and 20/21, 95.2%, respectively) and moderate agreement for small intestinal segments (16/21, 76.2%). Conclusions IUS could be of value and complementary to CTE for assessing CF, particularly in patients with affected terminal ileum and colon. IUS is a non-invasive technique for monitoring CD patients. Critical relevance statement In our study, excellent agreement was shown between intestinal US observers as well as between US and CT enterography (CTE) for assessing creeping fat (CF), which showed that ultrasound could be of value and complementary to CTE. Key Points Creeping fat (CF) is a potential therapeutic target in Crohn’s disease. Excellent agreement was shown between US and CT Enterography (CTE) for assessing CF. Ultrasound could be complementary to CTE for assessing CF. Graphical Abstract

Background: The Oncotype DX (ODX) recurrence score (RS), a 21-gene assay, has been proven to recognize patients at high risk of recurrence (RS >= 26) who would benefit from chemotherapy. However, it has limited availability and high costs. Our study thus aimed to identify ultrasound (US) imaging biomarkers and develop a prediction model for identifying patients with a high ODX RS. Methods: In this retrospective study, consecutive patients with T1-3N0-1M0 breast cancer who were hormone receptor positive and human epidermal growth factor receptor 2 (HER2) negative who had an available ODX RS were reviewed. Patients treated from May 2012 and December 2015 were placed into a training cohort, and those treated from January 2016 to January 2017 were placed in a validation cohort. Clinicopathologic data were collected, and preoperative US scans were analyzed. Univariable and multivariable regression analyses were performed to evaluate the independent predictors for a high-risk of breast cancer in the training cohort, and a nomogram was developed and evaluated with the area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis (DCA). Results: A total of 363 patients were in the training cohort and 160 in the validation cohort, with the proportion with a high RS (RS 26-100) being 14% and 13.1%, respectively. Echogenic halo, enhanced posterior echo, low level of progesterone receptor (PR), and high Ki-67 index were identified as independent risk factors for high RS (all P values <0.05). The nomogram was constructed based on the combined model, which showed a better discrimination ability than did the clinicopathological model [combined model: AUC P=0.001] and greater clinical benefit according to DCA. Furthermore, the nomogram was found to be effective in the validation cohort (AUC =0.90, 95% CI: 0.84-0.94), especially in patients with stage T1N0M0 disease (AUC =0.91, 95% CI: 0.84-0.95). Conclusions: US features may serve as valuable imaging biomarkers for the prediction of high recurrence risk in patients with T1-3N0-1M0 breast cancer and hormone receptor (HR)-positive and HER2-negative status. A nomogram incorporating PR status, Ki-67 index, and US imaging biomarkers showed a good discrimination ability in the early selection of patients at high risk of recurrence, especially in those with stage T1N0M0 disease.

The authors' objective was to obtain information on the current status of thyroid ultrasound in mainland China and analyze the factors affecting the accuracy of thyroid cancer ultrasound. A survey was designed and sent to 1,810 ultrasound departments across all 31 provinces in mainland China; responses were received from 308 ultrasound departments across 29 provinces. Each department was required to submit its basic quality control data and 10 randomly selected consecutive thyroid cases from March 2023. Experts from the National Ultrasound Quality Control Center would calculate the following data for the 10 cases of each department (3,080 total cases), which includes (1) ultrasound image qualification rate, (2) ultrasound reporting qualification rate, (3) Thyroid Imaging Reporting and Data System (TI-RADS) and American Thyroid Association (ATA) utilization rate, and (4) ultrasound accuracy rate. Moreover, to examine the variables affecting the accuracy of ultrasound, a multivariate logistic regression model was established. A review of the cases and the survey results finds that a total of 2,328.5 ultrasound examinations were performed annually per sonologist on average. The average TI-RADS/ATA utilization rate was 97.5%+/- 13.7%. The thyroid ultrasound image qualification rate was 82.3%+/- 28.1%, and the thyroid ultrasound reporting qualification rate was 92.7%+/- 16.3%, which were both notably different between tier 3 (the grade level that indicates the highest quality, based on hospital scale, management level, and hospital medical level) and tier 2/1 hospitals (P=0.0028; P<0.001). More ultrasound departments in tier 3 hospitals performed thyroid contrast-enhanced ultrasound (CEUS) and elastography than in tier 2/1 hospitals (52.0% vs. 17.3%, P<0.001; 43.6% vs. 21.0%, P=0.001). The accuracy of thyroid cancer ultrasound diagnosis was 87.2%+/- 17.5%. The ultrasound reporting qualification rate and the ultrasound image qualification rate were independently and significantly related to the accuracy of ultrasound diagnosis of thyroid cancer (both P<0.001). The authors summarized the present state of thyroid ultrasound in mainland China and discovered an effective correlation between the ultrasound reporting/image qualification and the accuracy of thyroid cancer ultrasound diagnosis. This understanding of the quality of thyroid ultrasound in China could help in the development of the discipline and contribute to improved diagnostic accuracy.

Despite the significant progress of large language models (LLMs) in various tasks, they often produce factual errors due to their limited internal knowledge. Retrieval-Augmented Generation (RAG), which enhances LLMs with external knowledge sources, offers a promising solution. However, these methods can be misled by irrelevant paragraphs in retrieved documents. Due to the inherent uncertainty in LLM generation, inputting the entire document may introduce off-topic information, causing the model to deviate from the central topic and affecting the relevance of the generated content. To address these issues, we propose the Retrieve-Plan-Generation (RPG) framework. RPG generates plan tokens to guide subsequent generation in the plan stage. In the answer stage, the model selects relevant fine-grained paragraphs based on the plan and uses them for further answer generation. This plan-answer process is repeated iteratively until completion, enhancing generation relevance by focusing on specific topics. To implement this framework efficiently, we utilize a simple but effective multi-task prompt-tuning method, enabling the existing LLMs to handle both planning and answering. We comprehensively compare RPG with baselines across 5 knowledge-intensive generation tasks, demonstrating the effectiveness of our approach.

The role of cross-sectional imaging in the management of acute appendicitis (AA) is contentious. This study aimed to investigate the current usage and diagnostic performance of ultrasound (US) and computed tomography (CT). A national survey was conducted by a core group from The National Ultrasound Quality and Control Center of China among radiologists practicing in medical institutions equipped with emergency departments and regularly performing appendectomies. Radiologists participated by completing the survey online from August 2022 to August 2023 after reviewing medical records of at least 40 patients with suspected AA. Sensitivity, specificity, positive predictive value, and negative predictive value and likelihood ratios were calculated for US and CT, respectively. Diagnostic performance of US between hospital subgroups were also compared. A total of 141 questionnaires were submitted and 118 were eligible, each representing a distinct hospital. A total of 1844 children and 4165 adults were included. There were 76.4