ObjectiveDiscriminating atypical hepatocellular carcinoma (HCC) from other malignancies in liver nodules classified as Liver Imaging Reporting and Data System category M (LR-M) remains a significant diagnostic challenge on conventional ultrasound examination. The LR-M category, originally intended to capture non-HCC malignancies, paradoxically contains up to 63% of atypical HCCs that deviate from classic enhancement patterns, leading to potential misdiagnosis and suboptimal treatment planning. While deep learning has shown promise in HCC diagnosis, most existing models rely exclusively on single-modality ultrasound, overlooking the diagnostic benefits of integrating complementary information from multiple imaging sources. To address this gap, we propose a novel attention-weighted tri-modal ultrasound network (TUS-Net) that integrates contrast-enhanced ultrasound (CEUS), B-mode ultrasound (BUS), and time-intensity curves (TICs) to improve diagnostic accuracy for these clinically challenging lesions. MethodsOur framework incorporates a three-dimensional convolutional neural network (C3D) backbone to extract spatiotemporal features from CEUS videos, capturing dynamic vascular patterns critical for lesion characterization. To effectively fuse complementary modalities, we introduce a dual-channel feature fusion module (DCFFM) that adaptively combines features from CEUS and BUS through channel-wise attention mechanisms, allowing the model to dynamically weigh the contribution of each modality based on diagnostic relevance. Additionally, we propose a temporal intensity feature fusion module (TIFFM) that leverages quantitative hemodynamic information from TICs to guide the model’s attention toward diagnostically critical temporal phases, such as arterial wash-in and portal venous washout. The model is further enhanced by automated lesion localization using YOLOX and class activation mapping for interpretability, ensuring that predictions align with clinically meaningful imaging features. ResultsEvaluated on a tri-modal ultrasound dataset comprising 161 patients with pathologically confirmed LR-M nodules (131 atypical HCC and 30 non-HCC malignancies), our model achieved an accuracy of 86.83%, a sensitivity of 92.50%, a specificity of 75.50%, and an AUC of 89.32% in screening atypical HCC. Compared to single-modality baselines, TUS-Net demonstrated superior specificity, a clinically critical metric given the higher risk associated with misclassifying non-HCC malignancies. Ablation studies confirmed the contribution of each module, with the full model outperforming both standard C3D and 3D ResNet backbones integrated with attention mechanisms. A reader study involving junior and senior radiologists further validated the clinical utility of AI assistance, showing consistent improvements in specificity and inter-reader consistency, particularly for less experienced clinicians. ConclusionThese results surpass existing benchmark models and demonstrate the potential of our approach to enhance diagnostic precision in clinically specific cases. By intelligently fusing multi-modal ultrasound data with attention-guided mechanisms, TUS-Net offers a reliable and interpretable tool that holds promise for improving the non-invasive diagnosis of atypical HCC in challenging LR-M liver nodules.

ObjectiveDiscriminating atypical hepatocellular carcinoma (HCC) from other malignancies in liver nodules classified as Liver Imaging Reporting and Data System category M (LR-M) remains a significant diagnostic challenge on conventional ultrasound examination. The LR-M category, originally intended to capture non-HCC malignancies, paradoxically contains up to 63% of atypical HCCs that deviate from classic enhancement patterns, leading to potential misdiagnosis and suboptimal treatment planning. While deep learning has shown promise in HCC diagnosis, most existing models rely exclusively on single-modality ultrasound, overlooking the diagnostic benefits of integrating complementary information from multiple imaging sources. To address this gap, we propose a novel attention-weighted tri-modal ultrasound network (TUS-Net) that integrates contrast-enhanced ultrasound (CEUS), B-mode ultrasound (BUS), and time-intensity curves (TICs) to improve diagnostic accuracy for these clinically challenging lesions. MethodsOur framework incorporates a three-dimensional convolutional neural network (C3D) backbone to extract spatiotemporal features from CEUS videos, capturing dynamic vascular patterns critical for lesion characterization. To effectively fuse complementary modalities, we introduce a dual-channel feature fusion module (DCFFM) that adaptively combines features from CEUS and BUS through channel-wise attention mechanisms, allowing the model to dynamically weigh the contribution of each modality based on diagnostic relevance. Additionally, we propose a temporal intensity feature fusion module (TIFFM) that leverages quantitative hemodynamic information from TICs to guide the model’s attention toward diagnostically critical temporal phases, such as arterial wash-in and portal venous washout. The model is further enhanced by automated lesion localization using YOLOX and class activation mapping for interpretability, ensuring that predictions align with clinically meaningful imaging features. ResultsEvaluated on a tri-modal ultrasound dataset comprising 161 patients with pathologically confirmed LR-M nodules (131 atypical HCC and 30 non-HCC malignancies), our model achieved an accuracy of 86.83%, a sensitivity of 92.50%, a specificity of 75.50%, and an AUC of 89.32% in screening atypical HCC. Compared to single-modality baselines, TUS-Net demonstrated superior specificity, a clinically critical metric given the higher risk associated with misclassifying non-HCC malignancies. Ablation studies confirmed the contribution of each module, with the full model outperforming both standard C3D and 3D ResNet backbones integrated with attention mechanisms. A reader study involving junior and senior radiologists further validated the clinical utility of AI assistance, showing consistent improvements in specificity and inter-reader consistency, particularly for less experienced clinicians. ConclusionThese results surpass existing benchmark models and demonstrate the potential of our approach to enhance diagnostic precision in clinically specific cases. By intelligently fusing multi-modal ultrasound data with attention-guided mechanisms, TUS-Net offers a reliable and interpretable tool that holds promise for improving the non-invasive diagnosis of atypical HCC in challenging LR-M liver nodules.

Purpose: The aim of this study was to investigate the feasibility of an intelligent handheld ultrasound (US) device for assisting non-expert general practitioners (GPs) in detecting carotid plaques (CPs) in community populations. Methods: This prospective parallel controlled trial recruited 111 consecutive community residents. All of them underwent examinations by non-expert GPs and specialist doctors using handheld US devices (setting A, setting B, and setting C). The results of setting C with specialist doctors were considered the gold standard. Carotid intima-media thickness (CIMT) and the features of CPs were measured and recorded. The diagnostic performance of GPs in distinguishing CPs was evaluated using a receiver operating characteristic curve. Inter-observer agreement was compared using the intragroup correlation coefficient (ICC). Questionnaires were completed to evaluate clinical benefits. Results: Among the 111 community residents, 80, 96, and 112 CPs were detected in settings A, B, and C, respectively. Setting B exhibited better diagnostic performance than setting A for detecting CPs (area under the curve, 0.856 vs. 0.749; P<0.01). Setting B had better consistency with setting C than setting A in CIMT measurement and the assessment of CPs (ICC, 0.731 to 0.923). Moreover, measurements in setting B required less time than the other two settings (44.59 seconds vs. 108.87 seconds vs. 126.13 seconds, both P<0.01). Conclusion Using an intelligent handheld US device, GPs can perform CP screening and achieve a diagnostic capability comparable to that of specialist doctors.

Purpose: The aim of this study was to investigate the feasibility of an intelligent handheld ultrasound (US) device for assisting non-expert general practitioners (GPs) in detecting carotid plaques (CPs) in community populations. Methods: This prospective parallel controlled trial recruited 111 consecutive community residents. All of them underwent examinations by non-expert GPs and specialist doctors using handheld US devices (setting A, setting B, and setting C). The results of setting C with specialist doctors were considered the gold standard. Carotid intima-media thickness (CIMT) and the features of CPs were measured and recorded. The diagnostic performance of GPs in distinguishing CPs was evaluated using a receiver operating characteristic curve. Inter-observer agreement was compared using the intragroup correlation coefficient (ICC). Questionnaires were completed to evaluate clinical benefits. Results: Among the 111 community residents, 80, 96, and 112 CPs were detected in settings A, B, and C, respectively. Setting B exhibited better diagnostic performance than setting A for detecting CPs (area under the curve, 0.856 vs. 0.749; P<0.01). Setting B had better consistency with setting C than setting A in CIMT measurement and the assessment of CPs (ICC, 0.731 to 0.923). Moreover, measurements in setting B required less time than the other two settings (44.59 seconds vs. 108.87 seconds vs. 126.13 seconds, both P<0.01). Conclusion Using an intelligent handheld US device, GPs can perform CP screening and achieve a diagnostic capability comparable to that of specialist doctors.

Purpose: The aim of this study was to investigate the feasibility of an intelligent handheld ultrasound (US) device for assisting non-expert general practitioners (GPs) in detecting carotid plaques (CPs) in community populations. Methods: This prospective parallel controlled trial recruited 111 consecutive community residents. All of them underwent examinations by non-expert GPs and specialist doctors using handheld US devices (setting A, setting B, and setting C). The results of setting C with specialist doctors were considered the gold standard. Carotid intima-media thickness (CIMT) and the features of CPs were measured and recorded. The diagnostic performance of GPs in distinguishing CPs was evaluated using a receiver operating characteristic curve. Inter-observer agreement was compared using the intragroup correlation coefficient (ICC). Questionnaires were completed to evaluate clinical benefits. Results: Among the 111 community residents, 80, 96, and 112 CPs were detected in settings A, B, and C, respectively. Setting B exhibited better diagnostic performance than setting A for detecting CPs (area under the curve, 0.856 vs. 0.749; P<0.01). Setting B had better consistency with setting C than setting A in CIMT measurement and the assessment of CPs (ICC, 0.731 to 0.923). Moreover, measurements in setting B required less time than the other two settings (44.59 seconds vs. 108.87 seconds vs. 126.13 seconds, both P<0.01). Conclusion Using an intelligent handheld US device, GPs can perform CP screening and achieve a diagnostic capability comparable to that of specialist doctors.

Objective To explore a deep learning-based automatic bone age estimation model for elbow joint X-ray images of Chinese Han adolescents and children and evaluate its performance.Methods A total of 943(517 males and 426 females)elbow joint frontal view X-ray images of Chinese Han ado-lescents and children aged 6.00 to<16.00 years were collected from East,South,Central and North-west China.Three experimental schemes were adopted for bone age estimation.Scheme 1:Directly in-put preprocessed images into the regression model;Scheme 2:Train a segmentation network using"key elbow joint bone annotations"as labels,then input segmented images into the regression model;Scheme 3:Train a segmentation network using"full elbow joint bone annotations"as labels,then in-put segmented images into the regression model.For segmentation,the optimal model was selected from U-Net,UNet++and TransUNet.For regression,VGG16,VGG19,InceptionV2,InceptionV3,ResNet34,ResNet50,ResNet101 and DenseNet121 models were selected for bone age estimation.The dataset was randomly split into 80%(754 samples)for training and validation for model fitting and hyperparameter tuning,and 20%(189 samples)as an internal test set to test the performance of the trained model.An additional 104 elbow joint X-ray images from the same demographic and age group were col-lected and used as an external test set.Model performance was evaluated by comparing the mean ab-solute error(MAE),root mean square error(RMSE),accuracies within±0.7 years(P±0.7 years)and±1.0 years(P±1.0 years)between the estimated age and the actual age,and by drawing radar charts,scat-ter plots,and heatmaps.Results When segmented with Scheme 3,the UNet++model achieved good segmentation performance with a segmentation loss of 0.000 4 and an accuracy of 93.8%at a learning rate of 0.000 1.In the internal test set,the DenseNet121 model with Scheme 3 yielded the best results with MAE,P±0.7 years and P±1.0 years being 0.83 years,70.03%,and 84.30%,respectively.In the external test set,the DenseNet121 model with Scheme 3 also performed best,with an average MAE of 0.89 years and an average RMSE of 1.00 years.Conclusion When performing automatic bone age estima-tion using elbow joint X-ray images in Chinese Han adolescents and children,it is recommended to use the UNet++model for segmentation.The DenseNet121 model with Scheme 3 achieves optimal per-formance.Using segmentation networks,especially that trained with annotation areas encompassing the full elbow joint including the distal humerus,proximal radius,and proximal ulna,can improve the ac-curacy of bone age estimation based on elbow joint X-ray images.

Objective To explore a deep learning network model suitable for bone age estimation using shoulder joint X-ray images in Chinese Han adolescents.Methods A retrospective collection of 1 286 shoulder joint X-ray images of Chinese Han adolescents aged 12.0 to<18.0 years(708 males and 578 females)was conducted.Using random sampling,approximately 80%of the samples(1 032 cases)were selected as the training and validation sets for model learning,selection and optimization,and the other 20%samples(254 cases)were used as the test set to evaluate the model's generalization ability.The original single-channel shoulder joint X-ray images and dual-channel inputs combining original images with segmentation labels(manually annotated shoulder joint regions multiplied pixel-by-pixel with original images,followed by segmentation via the U-Net++network to retain only key shoulder joint region information)were respectively input into four network models,namely VGG16,ResNet18,ResNet50 and DenseNet121 for bone age estimation.Additionally,manual bone age estimation was con-ducted on the test set data,and the results were compared with the four network models.The mean absolute error(MAE),root mean square error(RMSE),coefficient of determination(R2),and Pear-son correlation coefficient(PCC)were used as main evaluation indicators.Results In the test set,the bone age estimation results of the four models with dual-channel input of shoulder joint X-ray images outperformed those with single-channel input in all four evaluation indicators.Among them,DenseNet121 with dual-channel input achieved best results with MAE of 0.54 years,RMSE of 0.82 years,R2 of 0.76,and PCC(r)of 0.88.Manual estimation yielded an MAE of 0.82 years,ranking second only to dual-channel DenseNet121.Conclusion The DenseNet121 model with dual-channel input combined with original images and segmentation labels is superior to manual evaluation results,and can effectively estimate the bone age of Chinese Han adolescents.

ObjectiveTo study the regulatory effect of the partial sequence within the 3’ untranslated region （3’UTR） of slit guidance ligand 1 （Slit1） （Slit1-3’UTR） on the fibrotic phenotypes of cardiac fibroblasts （CFs） and its potential mechanism. MethodsThe adenovirus vector was used to overexpress the 1526nt sequence of Slit1-3’UTR in ICR neonatal mouse CFs （mCFs）. The expression of fibrosis-related genes in mCFs， such as collagen type 1 alpha1（COL1A1）， collagen type 3 alpha3 （COL3A1） and alpha smooth muscle actin （α-SMA） were detected by Western blot assay. The effect of Slit1-3’UTR 1526nt on the proliferation and migration of mCFs was assessed by EdU staining and Trans-well assays. Angiotensin Ⅱ （Ang Ⅱ） was used to treat mCFs， and the impact of Slit1-3’UTR 1526nt on the fibrotic phenotypes of Ang Ⅱ-induced mCFs was evaluated. After overexpression of Slit1-3’UTR 1526nt， miR-34a-5p mimic was transfected into mCFs， followed by actinomycin D treatment to detect the mRNA stability of Slit1-3’UTR 1526nt， and the levels of miR-34a-5p and its target gene SIRT1（si-SIRT1） in mCFs were determined. The effects of miR-34a-5p and small interfering RNA targeting SIRT1 on the Slit1-3’UTR 1526nt-mediated regulation of fibrotic phenotypes were also determined. ResultsAdenovirus-mediated overexpression of Slit 1-3’UTR 1526nt was achieved in mCFs. Overexpression of Slit 1-3’UTR 1526nt markedly inhibited the expression of the fibrosis-related genes， proliferation and migration of mCFs and fibrotic phenotypes of Ang Ⅱ. The results of actinomycin D assay showed that miR-34a-5p inhibited the stability of Slit1-3’UTR 1526nt in mCFs， while the level of miR-34a-5p was reduced in mCFs with overexpression of Slit1-3’UTR 1526nt. Transfection of miR-34a-5p promoted the fibrotic phenotypes， and reversed the inhibitory effect of Slit1-3’UTR 1526nt on the fibrotic phenotypes of mCFs. Overexpression of Slit1-3’UTR 1526nt significantly increased the level of miR-34a-5p target gene SIRT1 in mCFs. Transfection of miR-34a-5p and si-SIRT1 consistently reversed the inhibitory effects of Slit1-3’UTR 1526nt on the fibrotic phenotypes of mCFs. ConclusionSlit1-3’UTR1526nt inhibits the fibrotic phenotypes of mCFs by binding to miR-34a-5p and increasing the expression of its target gene of SIRT1.

ObjectivePrimary liver cancer, predominantly hepatocellular carcinoma (HCC), is a significant global health issue, ranking as the sixth most diagnosed cancer and the third leading cause of cancer-related mortality. Accurate and early diagnosis of HCC is crucial for effective treatment, as HCC and non-HCC malignancies like intrahepatic cholangiocarcinoma (ICC) exhibit different prognoses and treatment responses. Traditional diagnostic methods, including liver biopsy and contrast-enhanced ultrasound (CEUS), face limitations in applicability and objectivity. The primary objective of this study was to develop an advanced, light-weighted classification network capable of distinguishing HCC from other non-HCC malignancies by leveraging the automatic analysis of brightness changes in CEUS images. The ultimate goal was to create a user-friendly and cost-efficient computer-aided diagnostic tool that could assist radiologists in making more accurate and efficient clinical decisions. MethodsThis retrospective study encompassed a total of 161 patients, comprising 131 diagnosed with HCC and 30 with non-HCC malignancies. To achieve accurate tumor detection, the YOLOX network was employed to identify the region of interest (ROI) on both B-mode ultrasound and CEUS images. A custom-developed algorithm was then utilized to extract brightness change curves from the tumor and adjacent liver parenchyma regions within the CEUS images. These curves provided critical data for the subsequent analysis and classification process. To analyze the extracted brightness change curves and classify the malignancies, we developed and compared several models. These included one-dimensional convolutional neural networks (1D-ResNet, 1D-ConvNeXt, and 1D-CNN), as well as traditional machine-learning methods such as support vector machine (SVM), ensemble learning (EL), k-nearest neighbor (KNN), and decision tree (DT). The diagnostic performance of each method in distinguishing HCC from non-HCC malignancies was rigorously evaluated using four key metrics: area under the receiver operating characteristic (AUC), accuracy (ACC), sensitivity (SE), and specificity (SP). ResultsThe evaluation of the machine-learning methods revealed AUC values of 0.70 for SVM, 0.56 for ensemble learning, 0.63 for KNN, and 0.72 for the decision tree. These results indicated moderate to fair performance in classifying the malignancies based on the brightness change curves. In contrast, the deep learning models demonstrated significantly higher AUCs, with 1D-ResNet achieving an AUC of 0.72, 1D-ConvNeXt reaching 0.82, and 1D-CNN obtaining the highest AUC of 0.84. Moreover, under the five-fold cross-validation scheme, the 1D-CNN model outperformed other models in both accuracy and specificity. Specifically, it achieved accuracy improvements of 3.8% to 10.0% and specificity enhancements of 6.6% to 43.3% over competing approaches. The superior performance of the 1D-CNN model highlighted its potential as a powerful tool for accurate classification. ConclusionThe 1D-CNN model proved to be the most effective in differentiating HCC from non-HCC malignancies, surpassing both traditional machine-learning methods and other deep learning models. This study successfully developed a user-friendly and cost-efficient computer-aided diagnostic solution that would significantly enhances radiologists’ diagnostic capabilities. By improving the accuracy and efficiency of clinical decision-making, this tool has the potential to positively impact patient care and outcomes. Future work may focus on further refining the model and exploring its integration with multimodal ultrasound data to maximize its accuracy and applicability.

Objective To comparatively analyze the clinical characteristics of primary bilateral macronodular adrenal hyperplasia(PBMAH)and adrenal cortisol-producing Adenoma(CPA),and enhance the understanding of two diseases.Methods The clinical data of 85 PBMAH patients(PBMAH group)and 195 CPA patients(CPA group)diagnosed at Department of Endocrinology,the First Medical Center of Chinese PLA General Hospital,from September 2014 to August 2024 were retrospectively analyzed.The demographic characteristics,comorbidities,biochemical indicators,adrenocorticotropic hormone-cortisol(ACTH-F)levels,and adrenal imaging features and treatment conditions were compared between the two groups.Results(1)General characteristics:Compared with CPA group,PBMAH group had older age at diagnosis and a higher proportion of male patients.(2)Clinical characteristics:Compared with CPA group,PBMAH group had a longer disease duration,a higher proportion of subclinical Cushing's syndrome(CS),and a higher proportion of hypertension,impaired glucose tolerance/diabetes,bone mass reduction or osteoporosis,with higher serum potassium levels,and the differences were statistically significant(P<0.01).(3)Hormone levels:Both PBMAH and CPA groups showed ACTH-F rhythm disorder,significantly increased cortisol levels and suppressed ACTH.Compared with PBMAH group,CPA group had stronger autonomous cortisol secretion ability,manifested by increased midnight serum cortisol(F0:00),16:00 serum cortisol(F16:00),24-hour urinary free cortisol(24 h UFC)levels and lower 8:00 serum ACTH(ACTH8:00)and 16:00 serum ACTH(ACTH16:00)(P<0.01).After low-dose dexamethasone suppression test(LDDST),CPA group showed lower suppression rates of ACTH and cortisol,and higher proportions of paradoxical elevation in serum cortisol and 24 h UFC compared with PBMAH(P<0.01).Conclusions PBMAH has a longer disease course and higher proportions of comorbid metabolic disorders than CPA,mostly manifested as subclinical Cushing's syndrome.CPA has stronger autonomous cortisol secretion ability,with cortisol less likely to be suppressed after LDDST and more obvious paradoxical elevation of cortisol and 24 h UFC.