BACKGROUND:Autologous bone,allogeneic bone or artificial bone has been used to promote bone defect repair in the clinic,but the rate of non-healing is still high.The key is to ignore the importance of periosteum in the bone healing process.In the early stage of the project,the project team constructed an electrospinning membrane loaded with vascular endothelial growth factor to highly simulate the intramembranous osteogenesis of natural periosteum at the bone defect site,which promoted bone regeneration to a certain extent.However,the injured area often faces the dilemma of severe inflammatory response mediated by macrophages and lack of seed cells,resulting in the risk of inactivation or diffusion of delivered biological factors.Therefore,it is necessary to further optimize and coordinate the immune regulation and angiogenesis functions of biomimetic periosteum to promote bone repair.OBJECTIVE:To investigate the physicochemical properties of stem cell-engineered bionic periosteum and its role in regulating the inflammatory microenvironment to promote bone repair.METHODS:By combining L-polylactic acid-based microsol electrospinning,type Ⅰ collagen self-assembly and gel stem cell transplantation technology,a bionic periosteum(M@C-B)was constructed,in which the core layer loaded with vascular endothelial growth factor and the shell layer delivered bone marrow mesenchymal stem cells to regulate the immune microenvironment of bone defects.The physicochemical properties of the periosteum were characterized by scanning electron microscopy,transmission electron microscopy,and Fourier transform infrared spectroscopy.A co-culture system was established between the bionic periosteum and macrophages,bone marrow mesenchymal stem cells and human umbilical vein endothelial cells to explore immune regulation and in vitro osteogenic and angiogenic abilities.Finally,the osteogenic properties of the stem cell engineered bionic periosteum were further verified in a rat femoral condyle defect model.RESULTS AND CONCLUSION:(1)Transmission electron microscopy results showed that the micro-sol electrospinning(MS)formed a distinct core-shell structure.Scanning electron microscopy indicated that after the assembly of the collagen-l artificial periosteum(M@C)on the surface of the vascular endothelial growth factor-loaded micro-sol,a distinct"spider web-like"fibrous structure was deposited.Infrared spectroscopy further confirmed the successful self-assembly of collagen-l.Release experiments demonstrated that the M@C group mitigated the burst release phenomenon compared to the MS group,maintaining internal vascular endothelial growth factor activity and sustained release.(2)Live/dead cell staining and CCK-8 assay showed that bone marrow mesenchymal stem cells proliferated well and survived on three types of artificial periosteum:MS,purely aligned poly(L-lactic acid)(PLLA)surface self-assembled collagen-l artificial periosteum(PLLA@C),and vascular endothelial growth factor-loaded micro-sol fiber surface self-assembled collagen-l-bone marrow mesenchymal stem cells artificial periosteum(M@C-B).Among them,the M@C-B group had the highest number of live cells and the fastest proliferation rate.(3)Alkaline phosphatase staining,alizarin red staining,and osteopontin immunofluorescence staining showed that the PLLA@C and M@C-B groups significantly promoted osteogenic differentiation of bone marrow mesenchymal stem cells.Angiogenesis experiments demonstrated that the vascular endothelial growth factor-loaded groups(MS and M@C-B)had longer blood vessel lengths and more reticular vascular-like structures with more cross-linked nodes,with the M@C-B group being the most prominent.(4)Immunofluorescence and flow cytometry showed that artificial periosteum in the M@C-B group significantly inhibited the pro-inflammatory macrophage phenotype and promoted the polarization of macrophages towards the anti-inflammatory M2 phenotype.(5)In vivo studies further confirmed that the M@C-B group showed superior bone mineral density,trabecular thickness,relative bone volume,and trabecular spacing compared to other groups.(6)These results indicate that bone marrow mesenchymal stem cell-engineered artificial periosteum,through the rapid regulation of the bone defect immune microenvironment by the collagen-l-bone marrow mesenchymal stem cells outer phase and the sustained release of vascular endothelial growth factor by the micro-sol electrospinning core-shell structure of the inner phase,synergistically promotes bone healing.

BACKGROUND:Autologous bone,allogeneic bone or artificial bone has been used to promote bone defect repair in the clinic,but the rate of non-healing is still high.The key is to ignore the importance of periosteum in the bone healing process.In the early stage of the project,the project team constructed an electrospinning membrane loaded with vascular endothelial growth factor to highly simulate the intramembranous osteogenesis of natural periosteum at the bone defect site,which promoted bone regeneration to a certain extent.However,the injured area often faces the dilemma of severe inflammatory response mediated by macrophages and lack of seed cells,resulting in the risk of inactivation or diffusion of delivered biological factors.Therefore,it is necessary to further optimize and coordinate the immune regulation and angiogenesis functions of biomimetic periosteum to promote bone repair.OBJECTIVE:To investigate the physicochemical properties of stem cell-engineered bionic periosteum and its role in regulating the inflammatory microenvironment to promote bone repair.METHODS:By combining L-polylactic acid-based microsol electrospinning,type Ⅰ collagen self-assembly and gel stem cell transplantation technology,a bionic periosteum(M@C-B)was constructed,in which the core layer loaded with vascular endothelial growth factor and the shell layer delivered bone marrow mesenchymal stem cells to regulate the immune microenvironment of bone defects.The physicochemical properties of the periosteum were characterized by scanning electron microscopy,transmission electron microscopy,and Fourier transform infrared spectroscopy.A co-culture system was established between the bionic periosteum and macrophages,bone marrow mesenchymal stem cells and human umbilical vein endothelial cells to explore immune regulation and in vitro osteogenic and angiogenic abilities.Finally,the osteogenic properties of the stem cell engineered bionic periosteum were further verified in a rat femoral condyle defect model.RESULTS AND CONCLUSION:(1)Transmission electron microscopy results showed that the micro-sol electrospinning(MS)formed a distinct core-shell structure.Scanning electron microscopy indicated that after the assembly of the collagen-l artificial periosteum(M@C)on the surface of the vascular endothelial growth factor-loaded micro-sol,a distinct"spider web-like"fibrous structure was deposited.Infrared spectroscopy further confirmed the successful self-assembly of collagen-l.Release experiments demonstrated that the M@C group mitigated the burst release phenomenon compared to the MS group,maintaining internal vascular endothelial growth factor activity and sustained release.(2)Live/dead cell staining and CCK-8 assay showed that bone marrow mesenchymal stem cells proliferated well and survived on three types of artificial periosteum:MS,purely aligned poly(L-lactic acid)(PLLA)surface self-assembled collagen-l artificial periosteum(PLLA@C),and vascular endothelial growth factor-loaded micro-sol fiber surface self-assembled collagen-l-bone marrow mesenchymal stem cells artificial periosteum(M@C-B).Among them,the M@C-B group had the highest number of live cells and the fastest proliferation rate.(3)Alkaline phosphatase staining,alizarin red staining,and osteopontin immunofluorescence staining showed that the PLLA@C and M@C-B groups significantly promoted osteogenic differentiation of bone marrow mesenchymal stem cells.Angiogenesis experiments demonstrated that the vascular endothelial growth factor-loaded groups(MS and M@C-B)had longer blood vessel lengths and more reticular vascular-like structures with more cross-linked nodes,with the M@C-B group being the most prominent.(4)Immunofluorescence and flow cytometry showed that artificial periosteum in the M@C-B group significantly inhibited the pro-inflammatory macrophage phenotype and promoted the polarization of macrophages towards the anti-inflammatory M2 phenotype.(5)In vivo studies further confirmed that the M@C-B group showed superior bone mineral density,trabecular thickness,relative bone volume,and trabecular spacing compared to other groups.(6)These results indicate that bone marrow mesenchymal stem cell-engineered artificial periosteum,through the rapid regulation of the bone defect immune microenvironment by the collagen-l-bone marrow mesenchymal stem cells outer phase and the sustained release of vascular endothelial growth factor by the micro-sol electrospinning core-shell structure of the inner phase,synergistically promotes bone healing.

ObjectiveTo investigate whether Huanglian Jiedutang can inhibit neutrophil infiltration in the brains of middle cerebral artery occlusion （MCAO） mice by regulating the expression of neutrophil-related chemokines in exosomes， thereby achieving therapeutic effects. MethodsA total of 130 male specific pathogen-free （SPF） C57BL/6J mice were randomly divided into four groups： Sham-operated group， MCAO model group， Huanglian Jiedutang group （6 g·kg^-1）， and Ginaton group （21.6 mg·kg^-1）， with 10 mice in the Ginaton group and 40 mice in each of the remaining three groups. Mice in the Huanglian Jiedutang group and the Ginaton group were administered the corresponding drugs by oral gavage once daily at a volume of 0.15 mL·（10 g）^-1 for 7 consecutive days， while the sham-operated and model groups received an equal volume of saline via the same route. After 7 days， MCAO surgery was performed. The distal and proximal ends of the right common carotid artery （CCA） were ligated， a small incision was made between the two ligatures， and a silicone rubber-coated monofilament with a rounded tip was inserted into the lumen to occlude the CCA. The filament was left in place for 1 h to establish a focal cerebral ischemia model. At 24 h after modeling， mice were evaluated. Neurological function was assessed using the Longa score. Cerebral infarct volume was measured by 2，3，5-triphenyltetrazolium chloride （TTC） staining. Cerebral blood flow was observed by laser speckle imaging. Hematoxylin and eosin （HE） staining and Nissl staining were used to observe pathological changes in brain tissues. Exosomes were isolated from mouse plasma and brain tissues by ultracentrifugation and molecular size exclusion and identified by electron microscopy， particle size analysis， and protein blotting. Long-chain RNA libraries of exosomes were constructed and sequenced. Real-time quantitative reverse transcription polymerase chain reaction （Real-time PCR） was used to detect the mRNA expression of inflammatory factors and neutrophil-related chemokines in exosomes from plasma and brain tissues of each group. Enzyme-linked immunosorbent assay （ELISA） was used to detect the protein expression of inflammatory factors and neutrophil-related chemokines in exosomes from brain tissues of each group. Immunohistochemistry was used to detect the expression of the neutrophil-specific protein myeloperoxidase （MPO） in the brains of mice in each group. ResultsCompared with the sham-operated group， the model group showed decreased neurological function scores （P<0.01）， obvious cerebral infarction （P<0.01）， reduced cerebral blood flow （P<0.01）， neuronal necrosis in the brain， and decreased numbers of Nissl bodies （P<0.01）. The mRNA expression levels of IL-1β， MPO， CXCL1， CXCL2， CXCL3， CXCL10， CCL2， and CCL3 in exosomes from plasma and brain tissues were significantly increased （P<0.05， P<0.01）. The protein expression levels of IL-1β， MPO， CXCL2， and CXCL10 in exosomes from brain tissues were increased （P<0.05， P<0.01）， and MPO-positive rates and mean optical density values in brain tissues were elevated （P<0.01）. Compared with the model group， the Huanglian Jiedutang group and the Ginaton group showed increased neurological function scores （P<0.05）， reduced cerebral infarct volume （P<0.01）， restored cerebral blood flow （P<0.01）， reduced necrotic cells in the brain， and increased numbers of Nissl bodies （P<0.01）. In the Huanglian Jiedutang group， the mRNA expression levels of IL-1β， MPO， CXCL1， CXCL2， CXCL3， CXCL10， CCL2， and CCL3 in exosomes from plasma and brain tissues were decreased （P<0.05， P<0.01）. The protein expression levels of IL-1β， MPO， CXCL2， and CXCL10 in exosomes from brain tissues were reduced （P<0.05， P<0.01）， and MPO-positive rates and mean optical density values in brain tissues were decreased （P<0.01）. ConclusionHuanglian Jiedutang can effectively regulate the expression of neutrophil-related chemokines in exosomes from plasma and brain tissues of MCAO mice， thereby reducing neutrophil infiltration in the brain and achieving therapeutic effects.

The Pharmacovigilance Guidelines for Clinical Application of Traditional Chinese Medicine Injections （hereinafter referred to as the Guidelines） were released by the China Association of Chinese Medicine， with the standard number T/CACM 1563.4—2024. It is the first specialized guideline in China on the approach to pharmacovigilance activities for the clinical application of traditional Chinese medicine injections （TCMIs）. The Guidelines were jointly developed by the Institute of Basic Research in Clinical Medicine， China Academy of Chinese Medical Sciences， along with 30 experts in TCM pharmacovigilance， clinical practice （TCM， as well as integrated traditional Chinese and Western medicine），and evidence-based medicine from across the country. This publication filled the gap in standard documents in this field， both domestically and internationally. The Guidelines were formulated according to GB/T1.1—2020 Directives for standardization—Part 1： Rules for the structure and drafting of standardizing documents， the WHO Handbook for Guideline Development，and other methodological norms. Based on international norms，national laws and regulations，and scientific research results in the field of pharmacovigilance， methods adopted included expert interviews，literature research，nominal group technique， and Delphi method. Then， key points for pharmacovigilance for TCM injections were summarized and clarified in the four critical sections of "monitoring"，"identification"，"assessment"，and "control". The development process of the Guidelines included project initiation， international registration， expert interviews， literature search， and evaluation. Based on the research results of these steps，a draft was formed and revised through multiple rounds of in-group expert discussion and peer evaluations by 56 external experts. After revisions by the working group based on the feedback， the final version was formed. The Guidelines came into effect on January 8，2024，providing suggestions and reference norms for pharmacovigilance in the clinical application of TCMIs. To further promote the application and popularization of the Guidelines and help pharmacovigilance personnel better understand the development process，this study elucidates the background，methodological framework，and key development steps of the Guidelines.

AIM:To observe the morphological and structural changes of foveal cone photoreceptors in patients with age-related macular degeneration(ARMD)using adaptive optics scanning laser ophthalmoscopy(AOSLO)and to evaluate its application value in ARMD.METHODS:This was a retrospective cross-sectional study. Patients with ARMD who visited the Department of Ophthalmology, Army Medical Center of PLA, Army Medical University, and underwent AOSLO examination between September 2025 and October 2025 were enrolled as the experimental group(ARMD group). Age-matched individuals who underwent AOSLO examination during the same period and had either age-related cataract or pseudophakia with a normal macular region were selected as the control group(CON group). The AOSLO device was used to image a 2.4°×2.4° area of the fovea, and parameters including parafoveal cone photoreceptor density(PCPD), average inter-cell spacing, cell dispersion, and cell regularity were analyzed.RESULTS:A total of 53 participants(66 eyes)were included, comprising 24 patients(33 eyes)in the ARMD group [comprising 6 participants(6 eyes)in the intermediate ARMD group and 22 participants(27 eyes)in the late ARMD group(4 participants had one eye in the intermediate group and the other in the late ARMD group)], and 29 participants(33 eyes)in the CON group. The ARMD group included 13 males and 11 females, with a mean age of 69.36±9.79 y. The control group included 17 males and 12 females, with a mean age of 64.64±10.31 y. Compared to the CON group, the ARMD group exhibited significantly lower PCPD(31635±4887 vs 38524±3578 cells/mm2, P<0.01)and cell regularity(95.16%±0.75% vs 96.07%±0.67%, P<0.01), along with significantly greater average inter-cell spacing(4.43±0.26 vs 4.22±0.23 μm, P<0.01)and cell dispersion(20.23%±2.72% vs 16.47%±1.85%, P<0.01). Subgroup analysis within the ARMD group revealed that PCPD was significantly lower in the late ARMD subgroup(30831±4826 cells/mm2)compared to the intermediate ARMD subgroup(35254±3534 cells/mm2, P<0.05).CONCLUSION:Photoreceptor pathology in ARMD patients, as assessed by AOSLO, is characterized by decreased PCPD and cell regularity, as well as increased inter-cell spacing and dispersion. These structural alterations are closely associated with photoreceptor cell lesions. AOSLO, as a non-invasive and quantitative imaging modality, demonstrates promising application prospects in the clinical diagnosis of ARMD.

ObjectiveTo investigate whether Huanglian Jiedutang can inhibit neutrophil infiltration in the brains of middle cerebral artery occlusion （MCAO） mice by regulating the expression of neutrophil-related chemokines in exosomes， thereby achieving therapeutic effects. MethodsA total of 130 male specific pathogen-free （SPF） C57BL/6J mice were randomly divided into four groups： Sham-operated group， MCAO model group， Huanglian Jiedutang group （6 g·kg^-1）， and Ginaton group （21.6 mg·kg^-1）， with 10 mice in the Ginaton group and 40 mice in each of the remaining three groups. Mice in the Huanglian Jiedutang group and the Ginaton group were administered the corresponding drugs by oral gavage once daily at a volume of 0.15 mL·（10 g）^-1 for 7 consecutive days， while the sham-operated and model groups received an equal volume of saline via the same route. After 7 days， MCAO surgery was performed. The distal and proximal ends of the right common carotid artery （CCA） were ligated， a small incision was made between the two ligatures， and a silicone rubber-coated monofilament with a rounded tip was inserted into the lumen to occlude the CCA. The filament was left in place for 1 h to establish a focal cerebral ischemia model. At 24 h after modeling， mice were evaluated. Neurological function was assessed using the Longa score. Cerebral infarct volume was measured by 2，3，5-triphenyltetrazolium chloride （TTC） staining. Cerebral blood flow was observed by laser speckle imaging. Hematoxylin and eosin （HE） staining and Nissl staining were used to observe pathological changes in brain tissues. Exosomes were isolated from mouse plasma and brain tissues by ultracentrifugation and molecular size exclusion and identified by electron microscopy， particle size analysis， and protein blotting. Long-chain RNA libraries of exosomes were constructed and sequenced. Real-time quantitative reverse transcription polymerase chain reaction （Real-time PCR） was used to detect the mRNA expression of inflammatory factors and neutrophil-related chemokines in exosomes from plasma and brain tissues of each group. Enzyme-linked immunosorbent assay （ELISA） was used to detect the protein expression of inflammatory factors and neutrophil-related chemokines in exosomes from brain tissues of each group. Immunohistochemistry was used to detect the expression of the neutrophil-specific protein myeloperoxidase （MPO） in the brains of mice in each group. ResultsCompared with the sham-operated group， the model group showed decreased neurological function scores （P<0.01）， obvious cerebral infarction （P<0.01）， reduced cerebral blood flow （P<0.01）， neuronal necrosis in the brain， and decreased numbers of Nissl bodies （P<0.01）. The mRNA expression levels of IL-1β， MPO， CXCL1， CXCL2， CXCL3， CXCL10， CCL2， and CCL3 in exosomes from plasma and brain tissues were significantly increased （P<0.05， P<0.01）. The protein expression levels of IL-1β， MPO， CXCL2， and CXCL10 in exosomes from brain tissues were increased （P<0.05， P<0.01）， and MPO-positive rates and mean optical density values in brain tissues were elevated （P<0.01）. Compared with the model group， the Huanglian Jiedutang group and the Ginaton group showed increased neurological function scores （P<0.05）， reduced cerebral infarct volume （P<0.01）， restored cerebral blood flow （P<0.01）， reduced necrotic cells in the brain， and increased numbers of Nissl bodies （P<0.01）. In the Huanglian Jiedutang group， the mRNA expression levels of IL-1β， MPO， CXCL1， CXCL2， CXCL3， CXCL10， CCL2， and CCL3 in exosomes from plasma and brain tissues were decreased （P<0.05， P<0.01）. The protein expression levels of IL-1β， MPO， CXCL2， and CXCL10 in exosomes from brain tissues were reduced （P<0.05， P<0.01）， and MPO-positive rates and mean optical density values in brain tissues were decreased （P<0.01）. ConclusionHuanglian Jiedutang can effectively regulate the expression of neutrophil-related chemokines in exosomes from plasma and brain tissues of MCAO mice， thereby reducing neutrophil infiltration in the brain and achieving therapeutic effects.

The Pharmacovigilance Guidelines for Clinical Application of Traditional Chinese Medicine Injections （hereinafter referred to as the Guidelines） were released by the China Association of Chinese Medicine， with the standard number T/CACM 1563.4—2024. It is the first specialized guideline in China on the approach to pharmacovigilance activities for the clinical application of traditional Chinese medicine injections （TCMIs）. The Guidelines were jointly developed by the Institute of Basic Research in Clinical Medicine， China Academy of Chinese Medical Sciences， along with 30 experts in TCM pharmacovigilance， clinical practice （TCM， as well as integrated traditional Chinese and Western medicine），and evidence-based medicine from across the country. This publication filled the gap in standard documents in this field， both domestically and internationally. The Guidelines were formulated according to GB/T1.1—2020 Directives for standardization—Part 1： Rules for the structure and drafting of standardizing documents， the WHO Handbook for Guideline Development，and other methodological norms. Based on international norms，national laws and regulations，and scientific research results in the field of pharmacovigilance， methods adopted included expert interviews，literature research，nominal group technique， and Delphi method. Then， key points for pharmacovigilance for TCM injections were summarized and clarified in the four critical sections of "monitoring"，"identification"，"assessment"，and "control". The development process of the Guidelines included project initiation， international registration， expert interviews， literature search， and evaluation. Based on the research results of these steps，a draft was formed and revised through multiple rounds of in-group expert discussion and peer evaluations by 56 external experts. After revisions by the working group based on the feedback， the final version was formed. The Guidelines came into effect on January 8，2024，providing suggestions and reference norms for pharmacovigilance in the clinical application of TCMIs. To further promote the application and popularization of the Guidelines and help pharmacovigilance personnel better understand the development process，this study elucidates the background，methodological framework，and key development steps of the Guidelines.

Objective: To investigate the current status of emotional and behavioral problems among preschool children in Hefei, and to explore their association with the family rearing environment, providing scientific evidence for improving the family rearing environment and promoting children s mental health. Methods: From March to May 2023, a stratified cluster sampling method was applied to select 1 893 preschool children from five kindergartens in Hefei. Data on children s emotional and behavioral problems and family rearing environment were collected through electronic questionnaires. The Strengths and Difficulties Questionnaire (Parent Version) was employed to assess children s emotional and behavioral problems. Binary Logistic regression was conducted to analyze their association with family rearing environment. Results: Among preschool children, the prevalence rates of emotional symptoms, conducts, hyperactivity/inattention, peer relationship difficulties, and prosocial behavior were 9.7% (184 cases), 37.7% (713 cases), 16.7% (316 cases), 48.5% (918 cases), and 17.5% (332 cases), respectively. The abnormal detection rate of total difficulties score was 18.3% (346 cases). Binary Logistic regression analysis showed that higher maternal education level (postgraduate degree), parental daily interaction time ≥1 h, daily outdoor activity time for children ≥2 h, consistent family rearing attitudes, and nightly sleep duration for children ≥8 h were associated with lower abnormal risks of total difficulties score in preschool children [ OR(95%CI )=0.48(0.24-0.95), 0.66(0.49-0.90), 0.65(0.48-0.87), 0.39(0.29-0.52), 0.71(0.51-0.99)，all P < 0.05 ]. Conclusion The abnormal detection rates of emotional and behavioral problems among preschool children in Hefei are at moderate levels, and family rearing environment is related to children s emotional and behavioral problems.

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.