ObjectivesThis study aimed to develop and validate intra-tumoral and peri-tumoral radiomics models based on dynamic contrast-enhanced ultrasound (CEUS) to preoperatively predict lymph node metastasis (LNM) in thyroid cancer patients with type 2 diabetes.Materials and methodsA total of 203 pathologically confirmed diabetic thyroid cancer patients from three centers were retrospectively included and divided into a training cohort and two external validation cohorts. Radiomics features were extracted from CEUS parameters—time to enhancement (TTE), time to half-peak (TTHP), time to peak (TTP), and washout time (WT). Feature dimensionality reduction was performed using Variance Threshold, SelectKBest, and LASSO regression. Key LNM-related features were screened in the training cohort, and the optimal peri-tumoral region (1 mm vs 2 mm) was determined. Intra-tumoral and peri-tumoral radiomics scores were constructed and integrated into a multivariate logistic regression model. Model performance, calibration, and clinical utility were evaluated across all cohorts.ResultsThe 2 mm peri-tumoral region yielded a higher AUC than the 1 mm region in all cohorts, with statistical significance in the training cohort and a consistent trend in the external validation cohorts. The combined radiomics model achieved AUCs of 0.930 (95% CI: 0.876–0.964), 0.907 (95% CI: 0.796–0.968), and 0.865 (95% CI: 0.739–0.941) in the training and external validation cohorts. Calibration curves showed good agreement between predicted and actual outcomes, and decision curve analysis demonstrated substantial clinical benefit.ConclusionsThe CEUS-based combined radiomics model using intra-tumoral and 2 mm peri-tumoral features provides an effective tool for preoperative LNM prediction in thyroid cancer patients with type 2 diabetes.

To evaluate the clinical utility of superb microvascular imaging (SMI) for noninvasive diagnosis of renal interstitial fibrosis (IF) and compare its performance with conventional hemodynamic indicators, we prospectively enrolled 106 patients, categorized into minimal/mild IF (n = 71) and moderate/severe IF (n = 35). Collected measures included SMI-derived vascular density, capsule-to-terminal vessel distance (SMI distance), estimated glomerular filtration rate (eGFR), and conventional Doppler indices. Univariate and multivariable logistic regression identified independent predictors, which were integrated into a multivariable prediction model. Diagnostic performance was evaluated using receiver operating characteristic (ROC) analysis. Decision curve analysis (DCA) assessed net benefit across clinical thresholds. A nomogram was constructed for individualized prediction. SMI vascular density (OR = 0.955; p = 0.013), SMI distance (OR = 3.161; p = 0.036), and eGFR (OR = 0.976; p = 0.040) were independent predictors. Their areas under the curve (AUCs) were 0.779, 0.742, and 0.733, with sensitivities of 82.9, 77.1, and 77.1% and specificities of 64.8, 67.6, and 57.7%, respectively. The multivariable prediction model achieved an AUC of 0.837, with 68.6% sensitivity, 85.9% specificity, and 80.2% accuracy. DCA confirmed superior net benefit of the multivariable prediction model across most thresholds, consistently outperforming single parameters. SMI provides high sensitivity for early IF detection and, combined with eGFR, enhances diagnostic accuracy. This multivariable prediction model demonstrates clinical utility and may serve as a practical noninvasive tool for fibrosis assessment in chronic kidney disease.

Differentiating IgA nephropathy (IgAN) from membranous nephropathy (MN) often requires an invasive renal biopsy. This study aimed to develop and validate an explainable machine learning model for non-invasive discrimination between IgAN and MN using integrated clinical and ultrasound features. We enrolled 308 patients, randomly splitting them into training (n = 216) and test (n = 92) sets. Clinical and renal ultrasound features were collected. Fifteen machine learning algorithms were evaluated, including ensemble methods, traditional classifiers, and advanced techniques. Feature selection was performed with LASSO, and model interpretability was provided via SHAP analysis. Baseline characteristics were balanced between training and test sets (all P > 0.05). LASSO identified 17 key variables, with SHAP analysis highlighting albumin, age, and eGFR as top predictors. Among all models, Discriminant Analysis achieved the best performance, with a training AUC of 0.962 and test AUC of 0.968. It also attained 94.0

The effects of prenatal exposure to fine particulate matter (PM2.5) on fetal birth weight (FBW) results have been inconsistent, and the crucial components of PM2.5 may be responsible for these harmful effects. We aimed to explore the effects of prenatal exposure to PM2.5 components on FBW and determine the pregnancy susceptible exposure windows for PM2.5 components. We analyzed data from a birth cohort of 10,164 mother-infant pairs in Hefei, China. The study included singleton live births delivered between 2017 and 2021, with complete exposure and covariate data. We assessed pregnancy exposure to PM2.5 and its components (NH4+: ammonium; NO3−: nitrate; SO42−: sulfate; OM: organic matter; BC: black carbon) with spatial datasets matched to participants’ residential addresses, and assessed green space using the Normalized Difference Vegetation Index (NDVI). Generalized linear regression assessed the crucial components of PM2.5 affecting FBW Z-scores, and to further explore the susceptibility windows. Stratified analyses were conducted to explore potential moderating effects of green space. Our study found that pregnancy exposure to PM2.5 and its components was associated with reduced FBW Z-scores, with black carbon (BC) having a stronger negative effect. Each 1 µg/m3 increase in BC was associated with an expected FBW Z-score decrease of β 95

OBJECTIVE:This study aimed to develop and validate an ultrasound (US)-based deep transfer learning radiomics model, integrated with explainable machine learning, for the preoperative malignant risk prediction of parotid gland tumors (PGTs). METHODS:Data from 1,191 patients were retrospectively collected from three medical centers, and postoperative histopathological examination was used as the reference standard. Radiomics features and deep transfer learning features (ResNet50, Inception_V3, Vgg19) were extracted from the US images. Key predictive variables were selected using principal component analysis (PCA) and the least absolute shrinkage and selection operator (LASSO). Six classifiers-decision tree, gradient boosting machine, k-nearest neighbors, logistic regression, naïve Bayes, and random forest-were employed to construct models based on five feature sets: Clinical model, radiomics (Rad) model, deep transfer learning radiomics (DTLR) model, combined deep transfer learning and radiomics (DTLR-Rad) model, and a comprehensive combined model (CM Clinical + DTLR-Rad). Model performance was evaluated using the area under the curve (AUC). Feature importance was interpreted using SHapley Additive exPlanations (SHAP). A web application for real-time, personalized risk prediction was developed. RESULTS:In external test sets 1 and 2, the CM Clinical + DTLR-Rad model based on the random forest classifier achieved the highest AUCs among the evaluated models, with 0.922 (95% CI: 0.890-0.954) and 0.959 (95% CI: 0.932-0.985), respectively. The integrated model outperformed the clinical-only and single-modality models in both external test sets. SHAP visualizations demonstrated the contribution of individual features. The web application provided both prediction probabilities and feature-level interpretability. CONCLUSION:The CM Clinical + DTLR-Rad model demonstrated good predictive performance. The integration of interpretable machine learning and a web-based application may enhance preoperative risk stratification for PGTs and support clinical decision-making.

Background:Accurate preoperative prediction of cervical lymph node metastasis (LNM) in papillary thyroid microcarcinoma (PTMC) remains challenging, particularly because occult nodal disease is common and conventional ultrasound (US) assessment is operator-dependent. This study aimed to develop and validate a multicenter US-based deep learning radiomics nomogram integrating intratumoral, peritumoral, and clinical features for individualized LNM risk assessment in patients with PTMC. Methods:We retrospectively and prospectively collected multi-center data from three hospitals. Patients who underwent total thyroidectomy or lobectomy with lymph node dissection were allocated to a training set (763 cases), an external test set (118 cases), and a prospective validation set (94 cases). Radiomics features from within the tumor and deep learning (DL) features from the peritumoral region were extracted from the largest cross-sectional US image. Twelve machine learning (ML) models were built using the integrated features and evaluated on the test set to identify the optimal one. The machine learning prediction score (ML-score) was incorporated with clinical factors into a nomogram, and its performance was evaluated using area under the receiver operating characteristic curve (AUC), decision curve analysis (DCA), and clinical impact curves (CICs) analysis. Results:The support vector machine (SVM) model showed the best overall performance among the twelve ML algorithms and was used to generate the SVM-score. Multivariable logistic regression identified age, sex, maximum tumor diameter, genetic mutation status, and SVM-score as independent predictors of LNM. The integrated nomogram achieved AUCs of 0.894 in the training cohort, 0.842 in the external test cohort, and 0.856 in the prospective validation cohort. Calibration curves showed good agreement between predicted and observed LNM risk, with Hosmer-Lemeshow test P values of 0.189, 0.383, and 0.254 in the three cohorts, respectively. DCA and CIC demonstrated that the nomogram provided greater clinical net benefit than the clinical model or SVM-score model alone. Conclusions:The US-based DL radiomics nomogram integrating imaging features and clinical factors showed good performance for preoperative prediction of LNM in patients with PTMC. This non-invasive tool may assist individualized cervical lymph node risk stratification and support clinical decision-making.

To develop and validate a combined ultrasound-based radiomics-clinical model for differentiating benign and malignant breast lesions. A total of 3142 patients from eight hospitals between February 2012 and September 2024 were included in this multicenter retrospective development and validation study, with an additional single-center prospective test cohort. Lesions were manually segmented, and radiomics features were automatically extracted to construct five machine learning models. The best-performing radiomics model was combined with clinical features to build a combined model. Model performance and its impact on Breast Imaging Reporting and Data System (BI-RADS)-based biopsy decisions were evaluated. Logistic regression (LR) showed the best radiomics performance, with area under the curves (AUCs) of 0.83, 0.82, 0.81, and 0.82 across the training, internal test, external test, and prospective test sets. The clinical model achieved AUCs of 0.87, 0.85, 0.87, and 0.86, whereas the combined model achieved AUCs of 0.92, 0.90, 0.92, and 0.93, significantly outperforming both single-modality models (all p < 0.01). Decision curve analysis (DCA) showed that the combined model had a higher net benefit than the other models across a broad range of threshold probabilities (0.05–0.95) in this study. Performance remained stable across lesion size and age subgroups. In the reclassification analysis, the model suggested the potential to influence biopsy recommendations without a significant reduction in sensitivity and to increase the malignancy yield in BI-RADS 4a. Shapley additive explanations (SHAP) analysis provided clinically interpretable feature contributions. The interpretable ultrasound-based radiomics model enables reliable, noninvasive breast lesion diagnosis and may reduce unnecessary biopsies. This work developed an interpretable radiomics-clinical combined model in a multicenter retrospective development and validation study, with additional testing in a single-center prospective cohort, and may support breast lesion risk stratification and biopsy decision-making after further prospective clinical utility evaluation.

The co-occurrence of multiple cardiometabolic conditions has been mechanistically linked to impaired insulin signaling, yet the relative utility of surrogate insulin resistance (IR) markers in stratifying cardiometabolic multimorbidity (CMM) risk has not been systematically established. This study aimed to systematically evaluate and compare the associations of 14 IR indices with CMM incidence in a longitudinal Chinese cohort, with external replication in a U.S. nationally representative sample. This study included 8,522 participants from the China Health and Retirement Longitudinal Study (CHARLS) without CMM at baseline (2011). CMM was defined as the concurrent presence of at least two of the following three cardiometabolic conditions: type 2 diabetes, heart disease, and stroke. All 14 IR indices were evaluated both at baseline and as cumulative time-weighted averages derived from repeated measurements in 2011 and 2015. Associations between IR indices and CMM risk were examined using multivariable logistic regression, with dose–response relationships characterized through restricted cubic spline (RCS) modeling. Discriminatory capacity was quantified via receiver operating characteristic (ROC) curve analysis, complemented by net reclassification improvement (NRI) and integrated discrimination improvement (IDI) metrics. To verify the robustness of primary findings, Cox proportional hazards regression and pre-defined subgroup analyses were performed as supplementary sensitivity analyses. External cross-sectional replication was performed in the National Health and Nutrition Examination Survey (NHANES; 1999–2018). During follow-up through 2020, 591 CHARLS participants (6.9

BACKGROUND:The effect of AI-assisted diagnosis on sonographer performance in real-world prenatal settings remains unclear. This study aimed to assess the Prenatal Ultrasound Diagnosis Artificial Intelligence Conduct System (PAICS) in detecting AI-recognisable intracranial malformations and its effect on identifying anomalies beyond its recognition scope in high-risk pregnancies. METHODS:This multicentre, self-crossover, randomised controlled trial was done at five Chinese centres. High risk of fetal malformation singleton pregnancies (11-32 weeks' gestation) were examined by sonographers with 3 to less than 8 years of experience. Participants were randomly assigned 1:1 to two diagnostic sequences: independent real-time diagnosis followed by PAICS-assisted offline review, or PAICS-assisted real-time diagnosis followed by independent offline review, with a 4-week washout period. Allocation was concealed. An expert panel's diagnosis on video review served as the reference standard. Sonographers were masked to fetal anomaly status but were aware of AI assistance during scanning. Outcome assessors and the independent expert panel responsible for the reference standard diagnosis were masked to group assignments. Primary outcomes were sensitivity and specificity in detecting ten specific fetal intracranial malformations, with sensitivity assessed for superiority and specificity against a 5% margin of non-inferiority. The trial is registered with the Chinese Clinical Trial Registry (ChiCTR2200063424). FINDINGS:Between Sept 6, 2022 and Nov 1, 2023, 1584 scans were completed. PAICS-assisted diagnosis improved sensitivity for detecting specific fetal intracranial malformations. In the fetal-based analysis, sensitivity increased by 0·087 (95% CI 0·029 to 0·147; psuperiority<0·0001), with specificity meeting the non-inferiority criterion (difference 0·009, 95% CI -0·006 to 0·023; pnon-inferiority<0·0001). In the malformation-targeted analysis, sensitivity improved by 0·118 (95% CI 0·054 to 0·181; psuperiority<0·0001), and specificity was also non-inferior (difference 0·001, 95% CI -0·001 to 0·002; pnon-inferiority<0·0001). No adverse events related to the diagnostic procedure or the AI assistance were reported. INTERPRETATION:PAICS improves sonographers' sensitivity for detecting targeted fetal intracranial malformations while preserving their specificity, which supports the integration of AI assistance into fetal anomaly screening in high-risk clinical settings. FUNDING:National Natural Science Foundation of China, Guangdong Provincial Basic and Applied Basic Research Fund Project, Guangzhou Science and Technology Program, and Sun Yat-sen University Fundamental Research Funds for the Junior Faculty Program.

We aimed to develop and validate a radiopathomics model for predicting extrathyroidal extension (ETE) in papillary thyroid carcinoma (PTC). This retrospective study included 388 PTC patients with preoperative ultrasound and 400× cytology images from five medical centers between June 2017 and April 2024. We analyzed ultrasound and cytology images using Python and CellProfiler to extract features. Feature selection was performed using univariate analysis, Spearman correlation, and LASSO regression. The XGBoost algorithm was then used to build radiomics, pathomics, and combined radiopathomics models. The diagnostic performance of the radiopathomics model was compared with that of radiologists in an external validation cohort. Model and radiologist performance was evaluated using the area under the receiver operating characteristic curve (AUC). The radiopathomics model was visualized and interpreted through SHAP analysis. The radiopathomics model selected 21 features for construction. The AUC of the radiopathomics model was 0.887, 0.857, and 0.873 in the training, internal validation, and external validation cohorts, respectively, exceeding those of the single radiomics model (0.824, 0.787, and 0.804) and the pathomics model (0.809, 0.811, and 0.794). Compared with radiologists, the radiopathomics model improved the mean accuracy from 0.661 to 0.821. SHAP analysis showed that radiomics features played a major role in diagnosing ETE, while pathomics features provided additional support. The radiopathomics model serves as a promising auxiliary tool for preoperative ETE risk stratification and can help improve radiologists’ diagnostic performance. Not Applicable.

Gestational anemia significantly elevates the risk of adverse maternal and neonatal outcomes, necessitating early predictive tools for targeted intervention. This study aimed to develop and validate a robust machine learning (ML) framework to forecast perinatal complications and facilitate early risk identification. Perinatal mortality, preterm birth, low birth weight and macrosomia are defined as adverse outcomes. Analyzing a retrospective cohort of 5,710 pregnant women, we identified 22 initial variables using Lasso regression integrated with seven ML-based screening algorithms. Subsequently, eight predictive models were constructed and benchmarked via internal and external validation. Model performance was rigorously evaluated using receiver operating characteristic (ROC), precision‑recall (PR), calibration, and decision curves. Seven key predictors were identified, including gestational hypertension, obstetric history, and hepatic markers (Albumin, Alanine Aminotransferase, Globulin). The XGBoost model consistently outperformed its counterparts, demonstrating superior discriminative power (area under the curve (AUC) and F1-score) and clinical utility, as confirmed by the DeLong test and Kolmogorov‑Smirnov (KS) statistics. Based on XGBoost probabilities, we established a three-tier risk stratification: low-risk (< 0.28), medium-risk (0.28–0.52), and high-risk (≥ 0.53). Our ML-based framework offers a reliable tool for early risk assessment in gestational anemia, enabling clinicians to implement individualized management strategies through precise risk stratification.

PURPOSE:To develop and externally validate an ultrasound-based habitat subregional radiomics model for preoperative prediction of invasive breast cancer with concomitant ductal carcinoma in situ (IBC-DCIS). METHODS:A total of 1063 pathologically confirmed breast cancer patients from two centers were retrospectively enrolled and divided into a training cohort (n = 637) and an external validation cohort (n = 426). Tumor regions of interest were manually delineated on two-dimensional ultrasound images and further partitioned into three intratumoral habitat subregions using unsupervised clustering. Radiomics features were extracted from the whole tumor and each subregion. Feature selection was performed using Pearson correlation analysis and least absolute shrinkage and selection operator regression. Multiple machine learning models were constructed and evaluated using the area under the receiver operating characteristic curve (AUC) with 95% confidence intervals, calibration curves, and decision curve analysis. Model comparisons were conducted using the DeLong test. RESULTS:The support vector machine-based combined model achieved the highest AUC in the external validation cohort, with an AUC of 0.910 (95% CI: 0.883-0.936), and showed acceptable calibration and clinical net benefit across a broad range of threshold probabilities. DeLong test results showed that the combined model significantly outperformed imaging-based and single-region radiomics models (p < 0.05). To account for potential class imbalance, model performance was further assessed using multiple complementary metrics, including sensitivity, specificity, predictive values, balanced accuracy, F1-score, PR-AUC, and Brier score. CONCLUSION:An ultrasound-based habitat subregional radiomics model showed favorable performance for the preoperative prediction of IBC-DCIS and may provide supplementary information for preoperative risk stratification.

Ovarian cancer (OC) is a leading cause of gynecologic cancer mortality, with survival prediction limited by existing prognostic models that fail to capture tumor heterogeneity. Conventional methods lack precision for individualized risk assessment. Deep learning (DL) addresses these issues by integrating diverse data to improve survival prediction and risk stratification. This study introduces OvcaSurvivor, a novel multimodal DL framework for R0-resected OC patients, integrating whole-slide images (WSI), ultrasound (US), and clinical data from 543 patients. It uses advanced neural networks (CHIEF for WSI, ResNet50 for US) and an attention-guided fusion module. OvcaSurvivor showed superior performance, with C-indices of 0.81 (internal), 0.76 (external 1), and 0.70 (external 2). Time-dependent AUCs for 1-, 3-, and 5-year survival were highly accurate. WSI features drove prediction, and the model stratified patients into high/low-risk groups, highlighting clinical utility. This multimodal fusion advances OC precision oncology, enabling robust postoperative management.

Metabolic dysfunction-associated steatotic liver disease (MASLD) is associated with increased cardiovascular disease (CVD) risk. Lipoprotein(a) [Lp(a)] and insulin resistance (IR) are established cardiovascular risk factors, yet their joint associations with cardiovascular outcomes in MASLD remain poorly understood. We analyzed data from the UK Biobank and included 101,348 adults with MASLD for CVD mortality analyses and 94,089 individuals without baseline CVD for incident CVD analyses. IR was assessed using the triglyceride-glucose (TyG) index. Participants were categorized according to Lp(a) levels (< 125 vs. ≥ 125 nmol/L) and TyG index (low vs. high, defined by the 75th percentile) and further classified into four joint Lp(a)/IR groups, with low Lp(a)/low IR serving as the reference group. Cox proportional hazards models were used to evaluate associations of Lp(a), TyG, and their combined categories with incident CVD and CVD mortality. During a median follow-up of 15.7 years, elevated Lp(a) and higher TyG levels were each independently associated with increased risks of incident CVD and CVD mortality, regardless of each other’s status. In the fully adjusted model, each SD increase in Lp(a) was associated with a 14

Rationale and Objectives This study aimed to develop a deep learning model using a novel pixel-level radiomics approach based on two-dimensional (2D) and strain elastography (SE) ultrasound images to predict Ki-67 expression in breast cancer (BC). Methods This multicenter study included 1031 BC patients, who were divided into training (n = 616), internal validation (n = 265), and external test (n = 150) cohorts. An additional 63 patients were prospectively enrolled for further validation. The deep learning model, termed Vision-Mamba, predicts Ki67 expression by integrating ultrasound (2D and SE) images with pixel-level radiomics feature maps (RFMs). A combined model was subsequently constructed by incorporating independent clinical predictors. Model performance was assessed using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were applied to enhance interpretability. Results We developed a Vision-Mamba-US-RFMs-Clinical (V-MURC) model that integrates ultrasound images, RFMs, and clinical data for accurate prediction of Ki-67 expression in BC. The area under the ROC curve (AUC) values for the internal validation, external test, and prospective validation cohorts were 0.954 (95% CI, 0.929 - 0.975), 0.941 (95% CI, 0.903 - 0.975), and 0.945 (95% CI, 0.883 - 0.989), respectively, demonstrating excellent discrimination and calibration. Compared with individual models, the V-MURC model achieved significantly superior performance across all datasets (Delong test, P < 0.05). Calibration curves and DCA further supported its clinical applicability. SHAP analysis provided visual interpretability of the model's decision-making process. Conclusion The V-MURC model based on pixel-level RFMs can accurately predict Ki-67 expression in BC and may serve as a valuable tool for individualized treatment decision-making in clinical practice.

Background: Chronic prostatitis/chronic pelvic pain syndrome (CP/CPPS) is a common urinary tract disorder in males. Low-intensity pulsed ultrasound (LIPUS), a non-invasive therapeutic modality, has shown potential in alleviating inflammation associated with this condition. However, the underlying mechanisms through which LIPUS exerts its effects on CP/CPPS remain largely unknown. This study aimed to investigate the impact of LIPUS at varying intensities and durations on pain relief in a rat model of CP/CPPS. Methods: A CP/CPPS model was established by injecting 1% carrageenan into the prostate of rats. LIPUS at three different energy intensities (T1, T2 and T3: 0.5, 1.0 and 1.5 W/cm2, respectively) was applied to the pelvic region for 10 min daily over a period of 2 or 4 weeks. Pain relief and prostate injury were evaluated to assess the therapeutic efficacy of LIPUS. The mechanism study focuses on the role of LIPUS in regulating the Th17/Treg balance (by detecting CD4+ T cells in the prostate through flow cytometry and the expression of Foxp3/interleukin-17A (IL-17A) in prostate tissue through immunohistochemistry) as well as its effects on the IL-1/3/IL1R1/MyD88 signaling pathway in prostate tissue (by detecting pathway protein expression through Western blot and measuring inflammatory factor concentration using enzyme-linked immunosorbent assay). Results: To identify the optimal parameter of LIPUS for CP/CPPS, we treated model rats with LIPUS at intensities of 0.5, 1.0 or 1.5 W/cm2 for 2 or 4 weeks, and evaluated the therapeutic effects using multiple endpoints. Among these intensities, LIPUS at 1.0 W/cm2 energy/4 weeks exhibited the most prominent efficacy: compared with the model group, it significantly alleviated pelvic pain and reduced prostate inflammation scores. Further mechanistic studies demonstrated that, relative to the model group, LIPUS at 1.0 W/cm2 energy/4 weeks could regulate Th17/ Treg balance and downregulate the expression of related proteins, including IL-1/3, IL1R1 and MyD88. Conclusion: LIPUS, when administered at 1.0 W/cm2 for 4 weeks, showed superior efficacy in reducing pain in CP/ CPPS rats. These findings suggest that LIPUS may offer a promising therapeutic approach for CP/CPPS and provide insights into its underlying mechanism of action.

Background: Endometrial endometrioid carcinoma (EEC) tumor grade is a critical prognostic factor, but its accurate preoperative non-invasive assessment remains challenging due to the limitations of conventional imaging and biopsy. Transvaginal ultrasound (TVUS) is the primary imaging modality but offers limited quantitative insights for grading. Deep learning radiomics (DLR), which combines the strengths of deep learning (DL) for automatic feature extraction and radiomics for quantifying tumor heterogeneity, holds promise for uncovering prognostic information from routine ultrasound images. This study aimed to develop and validate a DLR model based on preoperative TVUS images for the non-invasive differentiation of EEC tumor grades. Methods: A total of 297 EEC cases with confirmed histological grades, including grade 1 (G1), grade 2 (G2), and grade 3 (G3), were selected from 1,258 endometrial cancer patients who underwent hysterectomy across eight centers. Radiomics features were extracted from TVUS images, and a radiomics model was constructed using the extreme gradient boosting (XGBoost) algorithm. Simultaneously, DL features were extracted using ResNet-50 to establish a DL model. A combined DLR model was then developed by integrating both feature sets, employing five-fold cross-validation for internal validation. An external testing cohort comprising 129 cases with corresponding grading data was collected from three independent centers. The performance of the three models in identifying EEC differentiation grade was compared using receiver operating characteristic (ROC) curve analysis to evaluate their diagnostic accuracy. Results: In differentiating EEC grades, the DLR model outperformed both the single radiomics and DL models. In the identification of G3 and G1/G2, the AUC of the DLR model was 0.871 and 0.843 in the training cohort and the external testing cohort, respectively. The AUC of the identification of G2 and G1 was 0.856 and 0.816 in the training cohort and the external testing cohort, respectively. Decision curve analysis confirmed the clinical utility of the DLR model. Conclusions: The DLR model based on TVUS images shows potential value for the non-invasive differentiation of EEC tumor grading and provides a useful supplement for non-invasive clinical staging of endometrial carcinoma prior to surgery.

The aim of this research was to develop a nomogram that integrates ultrasomics features and clinical factors to non-invasively predict preoperative lymph-vascular space invasion (LVSI) in patients with cervical cancer (CC). A total of 217 patients from three hospitals were retrospectively analyzed (the training set, n = 122; the test set, n = 53; and the validation set, n = 42). Tumor segmentation of the ultrasound(US) images was performed manually, then extracting a multitude of ultrasomics features from the segmented regions of interest (ROIs). After identifying the most significant ultrasomics features via a series of analyses and algorithms, five machine learning (ML) classification algorithms were utilized to develop and compare the ultrasomics models. Besides, we obtained clinically independent predictors for the diagnosis of LVSI and established the clinical model by univariate and multivariate analyses. Next, we compared the predictive capabilities of the clinical, ultrasomics, and combined models in forecasting LVSI in CC. Artificial neural networks (ANN) emerged as the top performer among the five ML classification algorithms. International Federation of Gynecology and Obstetrics (FIGO) staging for CC served as the independent predictor of LVSI. The nomogram, incorporating ultrasomics features and FIGO staging, demonstrated the highest diagnostic performance, with area under the curve (AUC) (95

Objective:To investigate qualitative and quantitative microvascular biomarkers derived from ultrasound localization microscopy (ULM) for differentiating benign and malignant breast lesions, and to evaluate their added value in improving BI-RADS-based diagnosis. Methods:In this prospective study, 295 patients with 301 breast lesions classified as BI-RADS category 4-5 on conventional ultrasound (US) underwent contrast-enhanced ultrasound and ULM prior to surgery or biopsy. Qualitative vascular biomarkers and quantitative parameters were derived from ULM images. Diagnostic performance was evaluated using receiver operating characteristic (ROC) analysis. BI-RADS categories were further re-evaluated by integrating ULM-derived biomarkers. Results:ULM qualitative vascular biomarkers classified microvascular morphology into five patterns: avascular, dot-strip, circular, chaotic, and radial. Benign lesions predominantly exhibited avascular, dot-strip, and circular patterns (5.6%, 23.4%, and 48.6%, respectively), whereas malignant lesions were mainly characterized by chaotic and radial patterns (56.7% and 28.4%, respectively; p < 0.001). The qualitative assessment achieved an AUC of 0.825 (95% CI: 0.780-0.871). All ULM quantitative biomarkers showed significant differences between benign and malignant lesions (all p < 0.05), among which maximum diameter demonstrated the highest diagnostic value (AUC = 0.898, 95% CI: 0.861-0.935). A combined quantitative biomarker (CQB) model further improved performance (AUC = 0.920). Integration of ULM qualitative and quantitative biomarkers with BI-RADS further improved diagnostic accuracy (AUC = 0.976; sensitivity 0.964; specificity 0.916) and may improve risk stratification beyond BI-RADS. Conclusion:ULM-derived qualitative and quantitative microvascular biomarkers provide complementary information beyond conventional ultrasound and may improve diagnostic performance and risk stratification in BI-RADS 4-5 breast lesions, but their role in safely reducing unnecessary biopsies requires further clinical validation.

BackgroundAccurately distinguishing benign from malignant indeterminate thyroid nodules is essential to avoid unnecessary diagnostic surgeries. This study aims to develop an online nomogram model that enables precise preoperative assessment of malignancy risk in indeterminate thyroid nodules (ITNs) following fine-needle aspiration, thereby helping to reduce unnecessary thyroidectomies.MethodsPatients with thyroid nodules were recruited from five centers and divided into retrospective training, independent testing, and prospective validation cohorts. Radiomics features and deep learning features were extracted from both B-mode ultrasound (BMUS) and strain elastography ultrasound (SEUS) images for each patient. Malignancy-associated features were selected to construct BMUS and SEUS signature scores, respectively. Multivariate regression analysis was performed on all variables to develop and visualize a comprehensive model for diagnosing benign and malignant ITNs. The model was further evaluated for discrimination, calibration, and clinical usefulness.ResultsMultimodal imaging features, genetic testing, and elastography levels were identified as key biological markers for diagnosing ITNs. The nomogram model built with these variables demonstrated strong performance. The area under the receiver operating characteristic curve was 0.907 (95% CI: 0.877-0.931) in the training set, 0.885 (95% CI: 0.821-0.932) in the external test set, and 0.860 (95% CI: 0.762-0.929) in the prospective validation set. Compared to clinical and individual scoring models, the nomogram demonstrated superior performance and calibration.ConclusionsThe proposed nomogram accurately diagnoses ITNs and holds promise for reducing unnecessary diagnostic thyroidectomies in clinical practice.