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.

ObjectiveTo investigate the effect of Sanet Vision Integrator (SVI)-based visuomotor integration training on the severity of neglect symptoms, upper limb motor function and activities of daily living (ADL) in patients with post-stroke unilateral spatial neglect (USN). MethodsFrom July, 2023 to September, 2025, 41 patients with post-stroke USN in Beijing Bo'ai Hospital were enrolled and randomly assigned to control group (n = 20) and experimental group (n = 21). Both groups received conventional rehabilitation and visual scanning training, while the experimental group received supplementary SVI-based visuomotor integration training, including saccadic, rotation and hand-eye coordination training, for eight weeks. They were assessed with Chinese Behavioral Inattention Test-Hong Kong version (CBIT-HK), Fugl-Meyer Assessment-Upper Extremities (FMA-UE) and modified Barthel Index (MBI) before and after treatment. ResultsAfter treatment, all the scores significantly improved in both groups (|t| > 4.004, P < 0.001), and the total scores of CBIT-HK and the core visual search sub-items (star cancellation, line crossing and letter cancellation), FMA-UE and MBI were better in the experimental group than in the control group (t > 2.096, P < 0.05). ConclusionSVI-based visuomotor training can effectively alleviate the USN symptoms, broaden the span of visual attention, facilitate the functional recovery of the affected upper extremity, and improve ADL in patients after stroke.

Immobilized enzyme-based enzyme electrode biosensors, characterized by high sensitivity and efficiency, strong specificity, and compact size, demonstrate broad application prospects in life science research, disease diagnosis and monitoring, etc. Immobilization of enzyme is a critical step in determining the performance (stability, sensitivity, and reproducibility) of the biosensors. Random immobilization (physical adsorption, covalent cross-linking, etc.) can easily bring about problems, such as decreased enzyme activity and relatively unstable immobilization. Whereas, directional immobilization utilizing amino acid residue mutation, affinity peptide fusion, or nucleotide-specific binding to restrict the orientation of the enzymes provides new possibilities to solve the problems caused by random immobilization. In this paper, the principles, advantages and disadvantages and the application progress of enzyme electrode biosensors of different directional immobilization strategies for enzyme molecular sensing elements by specific amino acids (lysine, histidine, cysteine, unnatural amino acid) with functional groups introduced based on site-specific mutation, affinity peptides (gold binding peptides, carbon binding peptides, carbohydrate binding domains) fused through genetic engineering, and specific binding between nucleotides and target enzymes (proteins) were reviewed, and the application fields, advantages and limitations of various immobilized enzyme interface characterization techniques were discussed, hoping to provide theoretical and technical guidance for the creation of high-performance enzyme sensing elements and the manufacture of enzyme electrode sensors.

Objective: In in vitro systems for differentiating and expanding natural killer (NK) cells, feeder cells provide essential cell-cell contact and paracrine signals that drive precursor proliferation and terminal maturation. However, existing xenogeneic feeder cells or tumor-derived genetically modified feeder cells pose risks of residual immunogenicity and malignant transformation, limiting clinical use. This study aims to develop a humanized mesenchymal-like stromal cell (hematopoietic organoid-derived human mesenchymal stromal cells, HO-hMSCs) derived from iPSC-based hematopoietic organoids, and elucidate its mechanisms of NK-supportive activity to enable a safe, efficient platform for clinical-grade NK cell production. Methods: Human induced pluripotent stem cells (iPSCs) were differentiated into hematopoietic organoids, from which HO-hMSCs were isolated. Flow-cytometric phenotyping and bulk RNA-sequencing were performed to compare HO-hMSCs with umbilical cord-derived MSCs (UC-hMSCs). The effect of HO-hMSCs on NK cell differentiation efficiency (CD3 CD56 ) and effector maturation (CD16 expression) were assessed by co-culture experiments, using UC-hMSCs as control. Results: 1) Hematopoietic organoid induction and NK differentiation: iPSCs were induced to form hematopoietic organoids using cytokine cocktails, which further differentiated into high-purity CD45 CD56 NK cells [(82.8%±12.07)% efficiency on day 21]. 2) HO-hMSC characteristics: HO-hMSCs exhibited upregulated expression of Notch pathway ligands (DLL4, JAG1, 4.06-8.04-fold), homeobox genes (HOXA3, HOXA5, log FC=1.28 and 1.44), and key regulators of NK development (GATA3, BCL11A) and cytokine receptors (IL7R, IL27RA, 6.76 to 13.34-fold increase). 3) Functional validation: Compared to UC-hMSCs, HO-hMSCs co-culture significantly enhanced NK cell proportion by 30.5% (P<0.05) and increased CD16 positivity (+20.5%). Conclusion: This study for the first time reveals that human hematopoietic organoid-derived HO-hMSCs possess potent hematopoietic niche-supportive activity. It provides a humanized, feeder-free platform for robust clinical-grade NK cell production and expands the translational utility of organoid technologies in cell therapy.

Objective To investigate the association between birth weight and dementia risk and the mediating roles of chronic diseases,and to assess potential biological pathways underlying the birth weight-associated dementia risk based on large-scale proteomics.Methods We used data from 279 743 participants aged 40 to 69 years enrolled in the UK Biobank.Birth weight was categorized into low birth weight(≤2 500 g),normal birth weight(2 500-3 999 g),and macrosomia(≥4 000 g).Multivariable Cox proportional hazards regression models were used to assess the associations between birth weight categories and all-cause dementia and its subtypes(Alzheimer's disease and vascular dementia).Proteomics analyses were conducted to identify proteins and the potential pathways involved.Results Low birth weight was associated with higher risks for all-cause dementia and its subtypes.The hazard ratios were 1.18(95%CI,1.08-1.30)for all-cause dementia,1.14(95%CI,1.00-1.31)for Alzheimer's disease,and 1.22(95%CI,1.01-1.48)for vascular dementia.A non-linear relationship was observed between birth weight and dementia risk(P for nonlinearity<0.001).Certain cardiometabolic diseases in middle-aged adults,such as diabetes,stroke,hypertension,and dyslipidemia,played a significant mediating role in the relationship between low birth weight and dementia risk,with the mediation proportion being 6.3%to 15.8%.Proteomic analyses identified 21 proteins linked to both low birth weight and all-cause dementia risk,which were significantly enriched in the pathways for viral protein interaction with cytokines and cytokine receptors,adipocytokine signaling,and cytokine-cytokine receptor interaction.Conclusion Low birth weight is positively associated with dementia risk.Cardiometabolic diseases in middle-aged adults may mediate the relationship between low birth weight and dementia risk.A number of proteins and the associated pathways underscore the relationship between low birth weight and dementia risk.

This study aims to explore the mechanism of lung and gut protection by Tanreqing Capsules on the mice infected with influenza virus based on "the lung-gut axis". A total of 110 C57BL/6J mice were randomized into control group, model group, oseltamivir group, and low-and high-dose Tanreqing Capsules groups. Ten mice in each group underwent body weight protection experiments, and the remaining 12 mice underwent experiments for mechanism exploration. Mice were infected with influenza virus A/Puerto Rico/08/1934(PR8) via nasal inhalation for the modeling. The lung tissue was collected on day 3 after gavage, and the lung tissue, colon tissue, and feces were collected on day 7 after gavage for subsequent testing. The results showed that Tanreqing Capsules alleviated the body weight reduction and increased the survival rate caused by PR8 infection. Compared with model group, Tanreqing Capsules can alleviate the lung injury by reducing the lung index, alleviating inflammation and edema in the lung tissue, down-regulating viral gene expression at the late stage of infection, reducing the percentage of neutrophils, and increasing the percentage of T cells. Tanreqing Capsules relieved the gut injury by restoring the colon length, increasing intestinal lumen mucin secretion, alleviating intestinal inflammation, and reducing goblet cell destruction. The gut microbiota analysis showed that Tanreqing Capsules increased species diversity compared with model group. At the phylum level, Tanreqing Capsules significantly increased the abundance of Firmicutes and Actinobacteria, while reducing the abundance of Bacteroidota and Proteobacteria to maintain gut microbiota balance. At the genus level, Tanreqing Capsules significantly increased the abundance of unclassified＿f＿Lachnospiraceae while reducing the abundance of Bacteroides, Eubacterium, and Phocaeicola to maintain gut microbiota balance. In conclusion, Tanreqing Capsules can alleviate mouse lung and gut injury caused by influenza virus infection and restore the balance of gut microbiota. Treating influenza from the lung and gut can provide new ideas for clinical practice.
Animals ; Drugs, Chinese Herbal/administration & dosage* ; Mice ; Lung/metabolism* ; Mice, Inbred C57BL ; Capsules ; Orthomyxoviridae Infections/virology* ; Gastrointestinal Microbiome/drug effects* ; Male ; Humans ; Female ; Influenza A virus/physiology* ; Influenza, Human/virology*

Objective To analyze the changes in optical coherence tomography angiography (OCTA) parameters in patients with essential hypertension，and to explore the effect of blood pressure on OCTA parameters. Methods A total of 164 patients with essential hypertension were selected and divided into controlled blood pressure group (n=92) and uncontrolled blood pressure group (n=72). OCTA examination was performed on the optic disc and macula of all patients, and the right eyes were selected for analysis. Results There were no significant differences in retinal nerve fiber layer (RNFL) thickness, radial peripapillary capillary (RPC) total vascular density, RPC total small vessel density, perifovea superficial capillary plexus (SCP) vascular density, and perifovea deep capillary plexus (DCP) vascular density between the two groups of patients. There were no significant differences in foveal avascular zone (FAZ) area, FAZ diameter, and fovea retinal thickness between the two groups of patients. The density of the parafovea SCP, parafovea DCP, and fractal dimension (FD) in the uncontrolled blood pressure group were significantly lower than those in the controlled blood pressure group (P＜0.05). Multiple linear regression analysis showed that elevation of blood pressure was a independently related factor of reduced parafovea DCP density (P＝0.026), while there was no correlation between the uncontrolled blood pressure and parafovea SCP density and FD level. Conclusions The blood pressure level is correlated with the parafovea DCP density, while has no correlation with other OCTA parameters in hypertension patients.

[Objective] To compare next generation sequencing (NGS) library construction technology between probe hybridization capture and amplicon methods, and analyze the influencing factors of HLA genotyping resolution level and its prospects in clinical applications. [Methods] A total of 207 clinical samples with known typing results and samples from the proficiency testing plan were selected. The conformity rate of HLA genotyping results, allele coverage and typing data analysis indicators were confirmed, and the effects of two library construction methods on the level of HLA genotyping discrimination were compared. [Results] The concordance rate of 207 samples with the feedback results of PT or prior well-characterized HLA genotypes was 100%. Among them, 91 samples were captured using hybridization probe capture method. Compared with the original amplicon method, the hybridization probe capture method can distinguish the alleles of DRB1 and DPB1 that cannot be determined in 13 samples. The allelic imbalance of DRB1, DPA1, and DQB1 loci in 6 samples was resolved. Three samples were found to have missed detection of alleles at the DQA1 and DQB1 loci. [Conclusion] The performance indicators of hybridization probe capture and amplicon performance confirmation meet the requirements of clinical detection of HLA genotyping, which provides an experimental method and basis for clinical application.