In audiovisual emotion recognition, representational learning is a research direction receiving considerable attention, and the key lies in constructing effective affective representations with both consistency and variability. However, there are still many challenges to accurately realize affective representations. For this reason, in this paper we proposed a cross-modal audiovisual recognition model based on a multi-head cross-attention mechanism. The model achieved fused feature and modality alignment through a multi-head cross-attention architecture, and adopted a segmented training strategy to cope with the modality missing problem. In addition, a unimodal auxiliary loss task was designed and shared parameters were used in order to preserve the independent information of each modality. Ultimately, the model achieved macro and micro F1 scores of 84.5% and 88.2%, respectively, on the crowdsourced annotated multimodal emotion dataset of actor performances (CREMA-D). The model in this paper can effectively capture intra- and inter-modal feature representations of audio and video modalities, and successfully solves the unity problem of the unimodal and multimodal emotion recognition frameworks, which provides a brand-new solution to the audiovisual emotion recognition.
Emotions ; Humans ; Attention ; Algorithms

Automatic classification of medical questions is of great significance in improving the quality and efficiency of online medical services, and belongs to the task of intent recognition. Joint entity recognition and intent recognition perform better than single task models. Currently, most publicly available medical text intent recognition datasets lack entity annotation, and manual annotation of these entities requires a lot of time and manpower. To solve this problem, this paper proposes a medical text classification model, bidirectional encoder representation based on transformer-recurrent convolutional neural network-entity-label-semantics (BRELS), which integrates medical entity label semantics. This model firstly utilizes an adaptive fusion mechanism to absorb prior knowledge of medical entity labels, achieving local feature enhancement. Then in global feature extraction, a lightweight recurrent convolutional neural network (LRCNN) is used to suppress parameter growth while preserving the original semantics of the text. The ablation and comparison experiments are conducted on three public medical text intent recognition datasets to validate the performance of the model. The results show that F1 score reaches 87.34%, 81.71%, and 77.74% on each dataset, respectively. The results show that the BRELS model can effectively identify and understand medical terminology, thereby effectively identifying users' intentions, which can improve the quality and efficiency of online medical services.
Semantics ; Neural Networks, Computer ; Humans ; Natural Language Processing

Spinal cord injury represents a severe central nervous system trauma characterized by prolonged treatment duration, limited neural regeneration, and delayed functional recovery, greatly affecting patients′ quality of life. The impaired neural tissue struggles to recover effectively due to both the hostile microenvironment and its own compromised state. Current clinical interventions, including early reduction, laminectomy decompression, and intravenous or intrathecal methylprednisolone administration, fail to simultaneously modulate the microenvironment and improve the neural status. Zinc, a trace element abundant in the central nervous system, plays a critical role in gene expression regulation, synaptic plasticity, and neuronal activity. Clinical evidences have indicated that lower serum zinc concentration in patients with spinal cord injury correlates with poorer outcomes and animal experiments have also demonstrated the neuroprotective effects of zinc. In fact, zinc supplementation therapy has not yet been developed into a mature clinical protocol. Besides, related animal studies still lack comprehensive understanding. To this end, the authors reviewed the biological characteristics of zinc, its administration routes, neuroprotective effects and mechanisms in the management of spinal cord injury, aiming to provide references for future basic research and clinical practice.

Objective:To analyze the clinical and endoscopic features of elderly patients with early gastric neuroendocrine neoplasms(g-NEN)to provide insights for clinical diagnosis and treatment.Methods:Seven cases of early-stage g-NEN in elderly patients admitted to Beijing Hospital from May 2020 to October 2024 were reported.Information including patient demographics, medical history, laboratory findings, endoscopic features, and pathological results was summarized an analyzed.Meanwhile, a literature review was conducted on g-NEN cases reported domestically and internationally from the establishment of databases to April 2025, with a comparative analysis of baseline characteristics, background mucosal conditions, and pathological diagnoses.Results:A total of 7 elderly patients with early g-NEN were included, including 3 males and 4 females, with an average age of 67 years at diagnosis.Endoscopic and pathological features: endoscopic lesions were mainly located on the greater curvature of the stomach, mostly slightly elevated, with 4/7(57.1%)being single lesions and the others accompanied by multiple polyps.The size of gastric neuroendocrine tumor(g-NET) lesions ranged from 1 mm to 19 mm, with a median Kyoto gastritis score of 3.In terms of pathological grading, there were 6 cases of G1 and 1 case of gastric neuroendocrine carcinoma(g-NEC). All 7 patients underwent endoscopic submucosal dissection(ESD). The follow-up period after surgery ranged from 3 months to 4 years, and no tumor was found at the vertical and horizontal margins of all lesions on pathology.A literature review of 32 reported cases showed that 40.6%(13/32)of cases exhibited background mucosal atrophy or intestinal metaplasia, with positive rates of chromogranin A(CgA) and synapsin(Syn)reaching 95.8%(23/24) and 100.0%(25/25), respectively.Notably, elderly patients with g-NEC demonstrated aggressive invasiveness, with a metastasis rate of 62.5%(5/8).Conclusions:Elderly patients with g-NEN exhibit high heterogeneity and significant differences in invasiveness, necessitating heightened clinical vigilance.Precise management should integrate background mucosal status, Kyoto gastritis score, pathological biomarkers, and imaging techniques.Blue laser imaging with magnifying endoscopy(BLI-ME)and endoscopic ultrasonography(EUS)can further analyze the characteristics of these tumors, and endoscopic submucosal dissection(ESD)remains the primary treatment modality for early-stage g-NEN.

Objective:To validate the accuracy of three-dimensional anatomical models reconstructed from cone beam CT (CBCT) using micro-CT as the gold standard, and to evaluate the feasibility of performing anatomical analyses on such models.Methods:A total of 13 isolated deciduous teeth with intact roots were collected, including 5 anterior teeth and 8 molars, with a total of 34 root canals. The teeth were extracted from children aged 3-9 years who visited Peking University Hospital of Stomatology from January 2019 to April 2022 due to trauma or periapical disease, and were then scanned by micro-CT (with a voxel size of 0.018 mm) and CBCT (with a voxel size of 0.125 mm), respectively. Using a threshold-based semi-automated region segmentation method, anatomical models of these isolated teeth were reconstructed from the two CTs. Subsequently, the two CT reconstructed models were registered based on the iterative closest point algorithm, followed by deviation analysis. The key anatomical parameters were measured on the micro-CT and CBCT models, respectively, and the differences were calculated.Results:The CBCT reconstruction models were relatively accurate in the hard tissue morphology, and 97.1% (33/34) of the root canals were identified accurately. When it comes to the deviation analysis, the average distance between the matched points on the CBCT reconstruction models and the micro-CT models was (0.01±0.03) mm for the hard tissue, and (0.00±0.03) mm for the pulp chambers and canals, which did not affect clinical observation. The hard tissue and root canal length of CBCT models were both smaller than those of micro-CT models ( P<0.05), with a 95% limits of agreement of (-0.70, 0.14) mm for the hard tissue, and a 95% limits of agreement of (-1.93, 1.00) mm for the pulp chambers and canals. The impact of these differences on clinical operations was all within the acceptable range. Conclusions:Using micro-CT as a validate standard, CBCT with a voxel size of 0.125 mm was proved to be an effective tool for the reconstruction of deciduous teeth. Therefore, the reconstructed models were appropriate for studying deciduous teeth anatomy.

Middle compartment defects,a common subtype of pelvic floor dysfunction(PFD),are primarily characterized by the prolapse of the uterus or vaginal vault.Magnetic resonance imaging(MRI)has emerged as a valuable diagnostic tool for PFD,offering superior soft tissue resolution while eliminating exposure to ionizing radiation.This review comprehensive summarizes current applications of MRI in the diagnosis and treatment of PFD,covering measurement methods,manifestations of three-level structural defects,postoperative efficacy evaluation,vaginal axial assessment,and evaluation of the mesh status.The authors suggest that MRI enables precise preoperative evaluation of three-level defects,thereby facilitating the development of personalized treatment plans.Additionally,MRI provides an accurate postoperative assessment of surgical outcomes and mesh status,offering a new basis for postoperative assessment.MRI demonstrates unique value in the diagnosis and treatment of middle compartment defects.

Objective To evaluate the perioperative safety and mid-term outcomes of laparoscopic sacrocolpopexy(LSC)and laparoscopic pectopexy(LP)for pelvic organ prolapse(POP).Methods A retrospective analysis was conducted on 274 POP patients,including 178 who underwent LSC and 96 who underwent LP,between August 2017 and January 2023.The extent of prolapse and anatomical restoration were assessed preoperatively and postoperatively using the Pelvic Organ Prolapse Quantification(POP-Q)system.Quality of life outcomes were evaluated with validated questionnaires,including Pelvic Floor Distress Inventory-short form 20(PFDI-20),Pelvic Floor Impact Questionnaire-short form 7(PFIQ-7),and Pelvic Organ Prolapse/Urinary Incontinence Sexual Questionnaire-12(PISQ-12).Postoperative patient satisfaction was assessed during follow-ups.Postoperative anatomical restoration,perioperative status,and postoperative complications,recurrence,and quality of life were compared between the two groups.Multivariate logistic regression was performed to identify postoperative risk factors for recurrence.Results The operative time in the LSC group was significantly shorter than that in the LP group(P<0.05).Intraoperative blood loss was higher in the LSC group compared to that in the LP group(P<0.05).The LSC group also exhibited higher rates of de novo stress urinary incontinence and constipation(P<0.05).The mean follow-up duration was(35.91±16.90)months.The positions of the indicator points(Aa,Ba,C,Ap,and Bp)in the POP-Q classification after the operation were all better than those before the operation.The PFDI-20 score,PFIQ-7 score,and PISQ-12 score all improved compared to those before the operation(P<0.05).Comparison of preoperative and postoperative PFDI-20,PFIQ-7,and PISQ-12 scores showed no intergroup differences.Compared with the LP group,the LSC group had the lower preoperative POP-Q measurements at points Aa and Ba(P<0.05),but superior postoperative measurements for all the indicator points(Aa,Ba,C,Ap,and Bp)(P<0.05).Recurrence occurred in 28 cases in the LP group and 4 cases in the LSC group,with the LP group presenting a significantly higher anatomical recurrence rate than the LSC group did(31.46%[28/89]vs.2.41%[4/166],P<0.05).The subjective cure rate(100%)and objective cure rate(97.59%)in the LSC group were superior to those in the LP group(88.76%and 68.54%,respectively;P<0.05).The results of the multivariate logistic regression analysis showed that,after adjusting for the confounding factors,including age,gravidity,parity,body mass index,and duration of POP,the risk of recurrence after LSC surgery was 0.044 times that after LP(odds ratio[OR],0.044;95%CI,0.015-0.133;P<0.001).Conclusion Mid-term outcomes of LP with partial cervical preservation appear inferior to those of LSC,with LSC demonstrating superior anatomical restoration and lower rates of anatomical recurrence.However,improvements in sexual function and quality of life are comparable between the two procedures.Further evaluation with larger sample sizes and longer follow-up is warranted to better characterize long-term outcomes.

ObjectiveTo investigate the possible mechanism of Shenfuhuang Formula （参附黄配方） in prevention and treatment of epsis-associated encephalopathy from the perspective of brain genomics. MethodsC57BL/6 mice were randomly divided into sham surgery group， sepsis group， and Shenfuhuang group， with 20 mice in each group. The sepsis group and Shenfuhuang group were induced to develop sepsis by cecal ligation and puncture （CLP） procedure. At 4 hours after modelling， Shenfuhuang group were gavaged with 2.5 g/（kg·d） of Shenfuhuang Formula， 0.5 ml each time， at 12 hours intervals， for a total of 4 times after modelling. Sepsis group and sham surgery group were given 0.5 ml of purified water orally. At 48 hours after modeling， the transcriptome sequencing was used to explore the differential gene expression in the effects of Shenfuhuang Formula on the brain regions of septic mice， and real-time PCR and ELISA were later used to further validate the differential gene and proteins expression. ResultsA total of 4605 genes were differentially expressed in Shenfuhuang group compared with sepsis group， of which 2353 genes were up-regulated and 2252 genes were down-regulated. According to the results of previous publications， six key genes were screened， including serine/threonine-protein kinase （Nek1）， myelin-associated glycoprotein （Mag）， endothelial cell-specific tyrosine kinase receptor （Tek）， a disintegrin and metalloproteinase with thrombospondin motifs 20 （Adamts20）， lymphocyte antigen 86 （Ly86）， and E3 ubiquitin-protein ligase （Traip）. Further genetic and protein validation revealed that， compared to the sham surgery group， the mRNA levels and corresponding protein levels of Nek1， Mag， Tek， Adamts20， Ly86， and Traip in the brain tissue of septic mice significantly reduced （P＜0.05）. In comparison to the sepsis group， Shenfuhuang group showed significantly increased mRNA levels and corresponding protein levels of Nek1， Mag， Tek， Adamts20， Ly86， and Traip （P＜0.05）. ConclusionThe potential therapeutic targets of Shenfuhuang Formula for treating sepsis-associated encephalopathy may be related to the Nek1， Mag， Tek， Adamts20， Ly86， and Traip genes and their encoded proteins.

ObjectiveTo investigate the whole genomic characteristics and phylogenetic relationships of clinical isolates of enteroaggregative Escherichia coli (EAEC) in diarrhea outpatients in Pudong New Area, Shanghai. MethodsBased on the diarrheal disease surveillance network in Pudong New Area, Shanghai, whole-genome sequencing was performed on a total of 55 EAEC strains isolated from fecal samples of the diarrhea outpatients from January 2015 to December 2019. The genome analyses based on raw sequencing data encompassed genome size, coding genes, dispersed repeat sequences, genomic islands, and protein coding regions, and pan-genome analyses were conducted simultaneously. Contigs sequences assays were performed to analyze molecular characteristics including serotypes, antibiotic resistance genes, and virulence factors. The phylogenetic clusters and multilocus sequence typing (MLST) were identified, and a phylogenetic tree was constructed. ResultsEAEC exhibited an open pan-genome. The predominant serotype of EAEC in diarrhea outpatients in Pudong New Area was O130:H27, and the carriage rate of β-lactam resistance genes was the highest (67.27%, 37/55). A total of 29 virulence factors and 106 virulence genes were identified, phylogenic group B1 was the predominant group, and clonal group CC31 was the dominant clonal group. The strain distribution was highly heterogeneous. ConclusionThe genomic characteristics of EAEC displayed significant strain polymorphism. It is necessary to develop effective strategies for differential diagnosis and improve detection capabilities for infection with EAEC of different serotypes and genotypes.

Objective To explore the value of improved ResNet18 lightweight deep learning(DL)models for automatically detecting gouty arthritis(GA)based on ultrasonogram of the first metatarsophalangeal joint(MTP1).Methods A total of 2 401 ultrasonograms obtained from 260 patients with suspected gout who underwent MTP1 ultrasound examination were included and divided into training set(1 910 ultrasonograms from 209 cases)and test set(491 ultrasonograms from 51 cases)at the ratio of 4∶1.GA lesions on ultrasonograms were manually labeled.After preprocessing,ResNet18 lightweight network was used to construct DL models for identifying the ultrasonogram category was normal or abnormal(with any manifestation of GA).Five-fold cross-validation method was adopted to evaluate the efficacy of the DL models constructed with 2,3,4 or 6 residual blocks,i.e.model 1,2,3 and 4,respectively,and the computational cost and the amount of parameters of each model were recorded.The efficacy of the models were verified using test set,and the best DL model was screened.Results The computational cost of model 1,2,3 and 4 was 7 558.27,2 963.73,4 012.33 and 6 093.39 M,respectively,while the amount of parameters was 4.61,4.91,4.91 and 5.28 M,respectively.Model 2 had the least computational cost with parameters only slightly more than model 1.In test set,no significant difference of accuracy nor the area under the curve was found among 4 models(all P＞0.05).The sensitivity of model 2 was higher than that of model 3,while its specificity was lower only than that of model 3(both P＜0.05),hence model 2 was the best DL model.Conclusion Improved ResNet18 lightweight DL models could be used for automatically detecting GA based on ultrasonogram of MTP1,among which model 2 was the best one.