Deep learning in ultrasound(US) imaging aims to construct foundational models that accurately reflect the modality's unique characteristics. Nevertheless, the limited datasets and narrow task types have restricted this field in recent years. To address these challenges, we introduce US-MTD120K, a multi-task ultrasound dataset with 120,354 real-world two-dimensional images. This dataset covers three standard plane recognition and two diagnostic tasks in ultrasound imaging, providing a rich basis for model training and evaluation. We detail the data collection, distribution, and labelling processes, ensuring a thorough understanding of the dataset's structure. Furthermore, we conduct extensive benchmark tests on 27 state-of-the-art methods from both supervised and self-supervised learning(SSL) perspectives. In the realm of supervised learning, we analyze the sensitivity of two main feature computation methods to ultrasound images at the representational level, highlighting that models which judiciously constrain global feature computation could potentially serve as a viable analytical approach for US image analysis. In the context of self-supervised learning, we delved into the modelling process of self-supervised learning models for medical images and proposed an improvement strategy, named MoCo-US, a solution that addresses the excessive reliance on pretext task design from the input side. It achieves competitive performance with minimal pretext task design and enhances other SSL methods simply. The dataset and the code will be available at https://github.com/JsongZhang/CDOA-for-UMTD .

Ultrasound imaging is widely used in early pregnancy to screen for fetal brain anomalies. However, the accuracy of diagnosis can be influenced by various factors, including the sonographer’s experience and environmental conditions. To address these limitations, advanced methods are needed to enhance the efficiency and reliability of fetal anomaly screening. In this study, we propose a novel approach based on a Fetal Brain Structures Detection Network (FBStrNet) for identifying key anatomical structures in fetal brain ultrasound images. Specifically, FBStrNet builds on the YOLOv5 baseline model, incorporating a lightweight backbone to reduce model parameters, replacing the loss function, and utilizing a decoupled detection header to improve accuracy. Additionally, our method integrates prior clinical knowledge to minimize false detection rates. Experimental results demonstrate that FBStrNet outperforms state-of-the-art methods, achieving real-time detection of fetal brain anatomical structures with an inference time of just 11.5 ms. This capability enables sonographers to efficiently visualize critical anatomical features, thereby improving diagnostic precision and streamlining clinical workflows.

Background: Ovarian cancer is the deadliest female reproductive malignancy. Accurate preoperative differentiation of benign and malignant ovarian masses is critical for appropriate treatment. O-RADS category 4 lesions present a wide range of malignant risk, challenging radiologists. Ultrasonic images are the primary focus of current deep learning models, with no consideration for clinical data. Methods: We proposed OTC-NET, a model that uses multimodal data for classification, which combines ultrasound images and clinical information to improve the classification ability of O-RADS 4 ovarian masses. Results: OTC-NET outperforms seven deep learning models and three radiologists of varying experience, with AUC significantly higher than junior (p < 0.001), intermediate (p < 0.01), and senior (p < 0.05) radiologists. Additionally, OTC-NET–assisted diagnosis notably improves AUC and accuracy of junior and senior radiologists (p < 0.05). Conclusions: These results indicate that OTC-NET provides superior diagnostic accuracy and has strong potential for clinical application.

This study addresses the challenge of the difficult identification of organ structures in the standard sections of fetal abdominal ultrasounds. A deep learning-based multi-task model named SEG-FAUSP was developed to segment the core anatomical structures of seven key fetal abdominal ultrasound sections. We collected fetal abdominal ultrasound images from pregnant women in the mid-pregnancy period (18–24 weeks) using various mainstream ultrasound devices, and professional physicians annotated key anatomical structures (e.g., umbilical veins, gastric bubbles, spine) in the images. Based on an improved deep learning framework, the model accurately segments and locates the target organ structures through a parallel dual-branch semantic segmentation network, which avoids the over-reliance on large-scale pre-trained data in traditional methods. Experimental results show that the model achieves excellent performance in anatomical structure segmentation, with the intersection over union of the bladder and gastric bubble both reaching above 0.84; its segmentation accuracy for complex structures such as the inferior vena cava is also significantly superior to the baseline model. As an end-to-end model, it simplifies the clinical interpretation process of fetal abdominal ultrasound, reduces the risk of missed diagnoses caused by unclear organ identification, provides an efficient auxiliary tool for prenatal screening in grassroots medical institutions, and is of great significance for improving the quality of newborns.

To establish reference ranges for fetal gastric parameters via two-dimensional (2D) and three-dimensional (3D) ultrasound virtual organ computer-aided analysis (VOCAL) technology and assess fetal gastric development and function in normal and hyperglycemia in pregnancy (HIP) cases. A prospective cross-sectional study was conducted with 457 normal fetuses (11–40 gestational weeks) via 2D and 3D ultrasound VOCAL techniques to measure gastric parameters (maximum length, transverse diameter, anteroposterior diameter, circumference, area, and volume) and calculate the gastric circumference-to-abdominal circumference (GC/AC), gastric area-to-abdominal area (GA/AA), and gastric emptying rate (GER). Nomograms were developed according to gestational age (GA). Intra- and inter-observer reliability were assessed in 50 randomly selected cases. Additionally, 126 HIP cases (90 well-controlled, 36 poorly-controlled) were analyzed via these nomograms. Comprehensive nomograms for 2D and 3D gastric parameters were established. The gastric dimensions (maximum length, transverse diameter, anteroposterior diameter, circumference, area, and volume) and GER increased with increasing GA (p < 0.05), whereas the GC/AC and GA/AA ratios were not correlated (p > 0.05). The measurements revealed high intra- and inter-observer reliability. Compared with both the well-controlled HIP and normal groups, the poorly-controlled HIP group had significantly greater gastric parameters and lower GERs (p < 0.05), with no significant differences between the well-controlled HIP and normal groups (p > 0.05). Both HIP groups had lower GC/AC and GA/AA ratios than that in the normal group (p < 0.05), with significant differences between the poorly-controlled and well-controlled HIP groups (p < 0.05). Fetal gastric parameter nomograms using 2D and 3D ultrasound VOCAL technology provide valuable tools for assessing fetal gastric development and function in normal and HIP pregnancies, highlighting the impact of glycemic control.

Background: This study aims to evaluate novel immune-related biomarkers for distinguishing borderline ovarian tumors (BOTs) from Benign ovarian tumors (BeOTs), addressing the diagnostic challenges posed by their intermediate biological behavior between benign and malignant neoplasms.Methods: We obtained the microarray expression profiles from the datasets (GSE4122 + GSE6822 + GSE36668) in the Gene Expression Omnibus (GEO) database and integrated them with the immune-related genes in the ImmPort database. Differentially immune-related genes (DIRGs) underwent Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. Protein-protein interaction (PPI) network was built to explore the connection. Candidate biomarkers were identified using the Least absolute shrinkage and selection operator (LASSO) and support vector machine-recursive feature elimination (SVM-RFE), with their diagnostic ability evaluated using Receiver operating characteristic (ROC) curves. A nomogram was constructed to predict BOTs. To validate the diagnostic potential and expression profiles, immunohistochemistry (IHC) analysis was performed in conjunction with the evaluation of an independent test group. We characterized the infiltration profiles of 22 immune cell types in BOTs through the CIBERSORT algorithm.Results: We identified 26 DIRGs between BOTs and BeOTs. These DIRGs were primarily associated with the positive regulation of transferase activity, the positive regulation of epithelial cell proliferation, and the positive regulation of the MAPK cascade. KEGG analysis indicated enrichment of Rap1 and PI3K-Akt signaling pathways. FGFR3, GNAI1, NR3C1, and PDGFA were found to have potential diagnostic value for BOTs (AUC_FGFR3 = 0.883, AUC_GNAI1 = 0.789, AUC_NR3C1 = 0.760, AUC_PDGFA = 0.783) and further validated in the test group (AUC_FGFR3 = 0.917, AUC_GNAI1 = 0.900, AUC_NR3C1 = 0.867, AUC_PDGFA = 0.833). Low expression of NR3C1 and GNAI1 and high expression of FGFR3 and PDGFA are associated with the development of BOTs. In addition, NR3C1 negatively correlated with CD4 memory resting T cells, as well as positively correlated with T cells gamma delta (P < 0.05).Conclusion: Our study findings suggested that NR3C1 may serve as an immune-related diagnostic biomarker for BOTs, offering a novel perspective for investigating the development and diagnosis of BOTs.