ArchiveArtificial intelligence (AI) is systematically reshaping the entire vaccine research and development pipeline through integrating deep learning architectures with multi-omics data. This facilitates a paradigm shift from traditional trial-and-error empirical approaches to a novel rational design framework centered on data-driven and algorithm-generated methodologies. Despite challenges including data heterogeneity, model opacity, algorithmic bias, and lagging regulatory frameworks, AI enables a transformative leap from structural mimicry to functional innovation in vaccine design via deep integration with immunological mechanisms. This review comprehensively analyzes AI applications across key vaccine development stages: target identification, antigen design, adjuvant screening, process optimization, clinical trial design, and vaccine hesitancy mitigation. We particularly summarize recent tools, core architectures, functional features, and representative cases for AI-powered rational vaccine design, providing actionable references for future AI implementation in vaccinology.
To investigate the effects of selective overexpression of hippocampal cannabinoid receptor 1 (CB1R) on newly generated cells and newly matured neurons in the hippocampus of stress-induced rats.
Thirty-two 6-8 week-old male SD rats (body weight 180 to 220 g) were randomly assigned to a Control group (n=10) and a Stress group (n=22). After the Stress group underwent 3 weeks of chronic restraint stress (CRS), 11 rats were randomly selected to form the Stress+CB1R-OE group, receiving hippocampal injections of an adeno-associated virus (AAV) for CB1R overexpression via stereotaxic surgery. The Control group and the remaining Stress group received a null AAV, designated as Control+AAV-NC and Stress+AAV-NC groups, respectively; viral transfection lasted 3 weeks. Body weight was measured every Sunday morning for weeks 1 to 7. Behavioral levels were assessed using the novel environment feeding suppression test. Immunofluorescence staining was used to detect the expression of 5-bromo-2'-deoxyuridine (BrdU), Neuronal Nuclei (NeuN), and CB1R in the hippocampal dentate gyrus (DG) (n=5 per group, 3 slices per rat). A NOVEL confocal microscope was used for scanning. Imaris software calculated the density of BrdU⁺ and BrdU⁺/NeuN⁺ cells per unit volume in the DG, while Image J software measured the mean fluorescence intensity of CB1R and NeuN in the DG. Western blot assay was used to detect hippocampal CB1R protein expression levels.
After 3 weeks of CRS, the body weight of the Stress+AAV-NC group was significantly lower than that of the Control+AAV-NC group (P=0.001). Selective hippocampal CB1R overexpression for 3 weeks partially ameliorated body weight gain in the Stress+CB1R-OE group, but their weight remained significantly lower than the Control+AAV-NC group (P=0.004). The novel environment feeding suppression test showed that compared to the Control+AAV-NC group, the Stress+AAV-NC group had a significantly prolonged feeding latency (P<0.001), with no significant difference in feeding amount within 5 min. After selective CB1R overexpression, the Stress+CB1R-OE group showed a significantly shortened feeding latency (P=0.039), with no significant differences in feeding amount among the three groups. Immunofluorescence results indicated that compared to the Control+AAV-NC group, the Stress+AAV-NC group had a significantly reduced mean fluorescence intensity of CB1R⁺ cells in the DG (P=0.008). After selective CB1R overexpression, the Stress+CB1R-OE group showed a significantly increased mean CB1R fluorescence intensity in the DG compared to the Stress+AAV-NC group (P<0.001). Western blotting results showed that hippocampal CB1R protein expression was significantly lower in the Stress+AAV-NC group than in the Control+AAV-NC group (P=0.001). Compared to the Stress+AAV-NC group, the Stress+CB1R-OE group had a significantly increased hippocampal CB1R expression level (P<0.001). Compared to the Control+AAV-NC group, the Stress+AAV-NC group had reduced densities of BrdU⁺ newborn cells (P=0.003) and BrdU⁺/NeuN⁺ newly matured neurons (P=0.001) in the DG. After selective CB1R overexpression, the Stress+CB1R-OE group showed significantly increased densities of BrdU⁺ cells (P=0.035) and BrdU⁺/NeuN⁺ neurons (P=0.028) in the DG. No significant difference was observed in the mean NeuN fluorescence intensity in the DG among the three groups.
Selective overexpression of hippocampal CB1R can alleviate stress-mediated reduction in hippocampal neurogenesis and subsequently improve behavioral deficits in stress-induced rats.
To construct an engineered milk-derived exosome (mExos) vector for efficient messenger RNA (mRNA) delivery to the respiratory tract by systematically screening cationic modification materials.
mExos were isolated via ultracentrifugation, followed by electroporation-mediated loading of firefly luciferase (Fluc) or enhanced green fluorescent protein (EGFP) mRNA. Chitosan (CS), polyethyleneimine (PEI), poly(beta-amino ester) (PBAE), cationic lipid (DOTAP), and protamine were used to modify mRNA-loaded mExos. Dynamic light scattering, nanoparticle tracking analysis, flow cytometry and luciferase reporter assay were employed to evaluate particle size, zeta potential, colloidal stability, cellular uptake, transfection efficiency, and cytotoxicity. Fifty female BALB/c mice (6 to 8 weeks, about 20 g) received intranasal administration of dye-labeled or Fluc-mRNA-loaded formulations. In vivo distribution and pulmonary transfection kinetics were monitored by IVIS imaging, while histopathology and serum biochemistry of major organs were analyzed at 24 h post-administration.
Cationic-modified mExos were successfully constructed. CS-modified mExos (CS-mExos) demonstrated surface charge reversal while maintaining optimal size and stability (PDI unchanged within 48 h). In vitro, CS-mExos enhanced cellular uptake in 16HBE and A549 cells by 2.56-fold (P<0.001) and 4.56-fold (P<0.001), respectively, and increased Fluc-mRNA transfection efficiency by 28.9-fold (P<0.001) and 26.5-fold (P<0.001) versus unmodified mExos, with >95% cell viability. In vivo imaging revealed CS-mExos specifically accumulated in the lungs with sustained retention >72 h, mediating 9.5-fold higher peak Fluc-mRNA expression at 6 h post-administration. Histopathology and serum biochemistry confirmed no significant organ damage or abnormal hepatic/renal indices.
Systematic screening identified chitosan as the optimal cationic material for engineering milk exosomes. The developed CS-mExos vector enables efficient targeted mRNA delivery to the respiratory tract with enhanced pulmonary transfection and favorable biosafety.
To investigate the effect and mechanism of a novel artemisinin derivative, disuccinate artemisinin (DA), in enhancing bacterial clearance by upregulating antibacterial autophagy in macrophages through targeting host mitochondrial respiration.
The experiment utilized E. colistrains ATCC 35218 and FITC-labeled K12 to infect the mouse macrophage cell line RAW264.7. Phagocytosis of bacteria by macrophages under different time points and doses was assessed via fluorescence imaging to determine the optimal drug dosage and time point. The impact of the drug on antibacterial autophagy in macrophages was evaluated using colony counting, Western blotting and immunofluorescence co-localization assay. Autophagy inhibitors 3-methyladenine (3-MA) and bafilomycin A1 were employed to examine the effect of autophagy suppression on drug efficacy through colony counting and immunofluorescence co-localization. Glucose deprivation was induced using low-glucose medium, and its influence on drug action was observed via fluorescent bacterial counting and immunofluorescence co-localization. Glycolytic intervention was investigated using the glycolytic activator oligomycin A and inhibitor 2-deoxy-D-glucose (2-DG), with effects monitored through fluorescent bacterial counting and immunofluorescence co-localization. Mitochondrial respiration was modulated with the activator carbonyl cyanide 4-trifluoromethoxy phenylhydrazone (FCCP) and inhibitor rotenone, and its impact on drug action was assessed via fluorescent bacterial counting and immunofluorescence co-localization. The effect of the drug on mitochondrial respiration in macrophages was determined by measuring the oxygen consumption rate (OCR) using cellular energy metabolism assays.
DA concentration-dependently enhanced macrophage phagocytosis and clearance of both drug-resistant ATCC35218 and sensitive K12 E. coli (P<0.05). Immunofluorescence and immunoblotting demonstrated DA upregulated antibacterial autophagy (P<0.01), remaining effective when autophagy initiation was blocked (P<0.01) but losing efficacy upon autophagic flux inhibition (P<0.01). Under glucose deprivation, DA retained its autophagy-enhancing effect (P<0.01). Both activation and inhibition of glycolysis limited DA’s efficacy (P<0.01). Maximizing mitochondrial respiration with FCCP abolished DA’s pro-phagocytic and autophagy-enhancing effects (P<0.01), whereas respiratory blockade with rotenone preserved them (P<0.01). Metabolic analysis showed DA restored infection-impaired maximal mitochondrial respiration, spare respiratory capacity, and ATP production (P<0.01).
DA enhances bacterial clearance by upregulating antibacterial autophagy through restoration of mitochondrial respiration in infected macrophages, representing a host-directed anti-infective candidate targeting mitochondrial metabolism.
To investigate the ameliorative effect of human umbilical cord mesenchymal stem cells (hUC-MSCs) on radiation-induced reproductive damage in male mice and its potential mechanisms.
After obtaining informed consent and signed consent forms from healthy full-term pregnant women in the Department of Obstetrics and Gynecology at Xijing Hospital of Air Force Medical University, umbilical cord tissue approximately from 6 to 8 cm in length from healthy full-term newborns was collected under sterile conditions in the operating room for isolating and culturing hUC-MSCs; 45 healthy male C57BL/6 mice aged from 6 to 8 weeks with a body weight of 22±2 g were divided into Control+PBS (n=3), Control+MSCs (n=6), Rad+MSCs (n=6), Control (n=10), Rad (n=10), and Rad+hUC-MSCs (n=10) groups using a random number table method; the Rad, Rad+MSCs, and Rad+hUC-MSCs groups were subjected to whole-body single-dose irradiation using 60Co γ-ray at a dose rate of 1.28 Gy/min and a total dose of 6 Gy; immediately after irradiation, the Rad+MSCs and Rad+hUC-MSCs groups received a single tail vein injection of hUC-MSCs suspension (2×10⁶ cells/200 μL); the distribution of hUC-MSCs in testicular tissue was observed using a multimodal in vivo imaging system and confocal fluorescence microscopy; at 12 weeks post-irradiation, bilateral testes were harvested and weighed to calculate the testicular coefficient; sperm quality was assessed under a microscope; testicular morphology and structure were evaluated using Hematoxylin and Eosin (HE) staining; reactive oxygen species (ROS) and mitochondrial membrane potential (MMP) levels in testicular tissue were measured by flow cytometry; malondialdehyde (MDA), total superoxide dismutase (T-SOD), and catalase (CAT) contents were detected by colorimetry; testicular cell apoptosis was assessed using TUNEL staining; and Western blot was used to detect the expression levels of Parkin RBR E3 ubiquitin-protein ligase (Parkin), PTEN-induced putative kinase 1 (PINK1), Beclin1, OPA1 mitochondrial dynamin-like GTPase (OPA1), dynamin-related protein 1 (Drp1), Bcl2-associated X protein (Bax), and B cell lymphoma-2 (Bcl-2).
At 24 h and 48 h post-irradiation, multimodal in vivo imaging showed no fluorescent signals in the testicular tissue of the Control+PBS group, while fluorescent signals were observed in the Control+MSCs and Rad+MSCs groups; further confocal fluorescence microscopy revealed red fluorescence in the interstitial regions of testicular tissue in the Control+MSCs and Rad+MSCs groups, detecting fluorescence dye labeled MSCs signals; at 12 weeks post-irradiation, compared to the Control group, the Rad group showed significantly decreased sperm count (P<0.05) and motility (P<0.01), significantly increased malformation rate (P<0.001), vacuolization and disordered arrangement of spermatogenic cells, reduced seminiferous tubule diameter and spermatogenic epithelium thickness (P<0.001), elevated testicular tissue ROS levels (P<0.05), significantly decreased MMP levels (P<0.001), increased MDA content (P<0.001), elevated T-SOD activity (P<0.05), decreased CAT activity (P<0.05), increased protein expression of Parkin (P<0.01), PINK1 (P<0.05), Beclin1 (P<0.05), OPA1 (P<0.01), and Drp1 (P<0.05), increased TUNEL-positive cells (P<0.001), and decreased Bcl-2/Bax ratio (P<0.01); compared to the Rad group, the Rad+hUC-MSCs group exhibited significantly increased sperm count and motility (P<0.05), significantly decreased malformation rate (P<0.001), improved vacuolization and arrangement of spermatogenic cells, increased spermatogenic epithelium thickness (P<0.01), reduced testicular tissue ROS levels (P<0.05), significantly increased MMP levels (P<0.01), decreased MDA content (P<0.001), reduced T-SOD activity (P<0.05), increased CAT activity (P<0.05), decreased protein expression of Parkin (P<0.05), PINK1 (P<0.05), Beclin1 (P<0.05), OPA1 (P<0.05), and Drp1 (P<0.05), reduced TUNEL-positive cells (P<0.001), and increased Bcl-2/Bax ratio (P<0.05).
hUC-MSCs may alleviate radiation-induced male reproductive damage by mitigating mitochondrial damage, reducing testicular tissue oxidative stress injury and cell apoptosis.
To detect the infiltration level of CD103+CD39+CD8+ T cell subsets in gastric cancer tissues and analyze their response characteristics, clarifying the clinical relevance of this subset.
Surgical specimens (tumor tissues, adjacent tissues, and peripheral blood) were collected from gastric cancer patients at the Department of General Surgery, Second Affiliated Hospital of Army Medical University between January 2023 and May 2024. Flow cytometry and immunofluorescence staining quantified the proportion and number of CD103+CD39+CD8+ T cells in tumor versus adjacent tissues. Flow cytometry further analyzed CD69, PD-1, and IFN-γ expression in this subset. Peripheral blood mononuclear cells (PBMCs) were isolated and cultured in vitro to assess TGF-β1-induced differentiation. Correlation with disease stage was analyzed by detecting the infrltration level of CD103+CD39+CE8+ T Cell subsets via flow cytometry.
Compared with adjacent tissues, gastric cancer tissues showed significantly increased proportions of CD103+CD39+CD8+ T cells (P<0.05); immunofluorescence confirmed higher numbers in tumor tissues. Relative to CD103+CD39- and CD103-CD39-CD8+ T subsets, CD103+CD39+CD8+ T cells exhibited elevated CD69 and PD-1 expression (P<0.05) but reduced IFN-γ (P<0.05). In vitro TGF-β1 stimulation upregulated co-expression of CD103 and CD39 in CD8+ T cells (P<0.05). Moreover, advanced gastric cancer patients had higher tumor infiltration of this subset than early-stage patients (P<0.05).
CD103+CD39+CD8+ T cell subsets display increased infiltration and functionally impaired responses in gastric cancer, facilitating tumor immune escape.
To evaluate the diagnostic performance of combining sonographic features with BRAFV600E mutation for differentiating benign and malignant Bethesda category Ⅲ thyroid nodules.
This retrospective cohort study included 275 patients (278 nodules) diagnosed as Bethesda category Ⅲ by ultrasound-guided fine needle aspiration cytology (US-FNAC) at First Affiliated Hospital of Army Medical University between January 2020 and December 2024, with final confirmation via surgical pathology or repeated US-FNAC. Nodules were classified as benign or malignant based on pathological results. Diagnostic efficacy of age, sex, 9 sonographic features, ACR TI-RADS score, US-FNAC nuclear morphology, and BRAFV600E mutation was analyzed using multivariate logistic regression.
Significant differences existed in age, nodule margin, aspect ratio, maximum diameter, echogenicity, Adler blood flow grading, ACR TI-RADS classification, US-FNAC nuclear morphology, and BRAFV600E status between benign and malignant nodules (all P<0.05). Independent risk factors for malignancy included age≤52 years (OR=7.444), BRAFV600E mutation (OR=108.218), ground-glass nuclei (OR=5.389), and abnormal cervical lymph nodes (OR=13.351, P<0.05). The combined model demonstrated accuracy was 0.823, sensitivity was 0.863, specificity was 0.925, positive predictive value was 0.977, negative predictive value was 0.649, and AUC was 0.955.
Integration of sonographic features and BRAFV600E mutation provides high diagnostic performance for Bethesda category Ⅲ thyroid nodules.
To investigate the feasibility of reducing radiation dose in cerebral perfusion CT (CTP) while maintaining diagnostic accuracy by optimizing temporal resolution.
This retrospective cohort study included 117 stroke-suspected patients undergoing CTP at the First Affiliated Hospital of Army Medical University from January to September 2025. Three protocols were designed: Protocol 1 (full temporal resolution, 21-phase acquisition, baseline), Protocol 2 (every-other-phase sampling),Protocol 3 (peak-enhancement +1/-1 phase retention with every-other-phase sampling for remaining phases). Radiation reduction percentages relative to Protocol 1 were calculated. Regions of interest (ROI) were placed in ischemic core/penumbra to measure mean±SD of perfusion parameters: time to maximum of the residue function (Tmax), mean transit time(MTT), cerebral blood volume(CBV) and cerebral blood flow(CBF). Inter-protocol differences were assessed, and ischemic penumbra volume discrepancies were also analyzed.
Protocol 2 and 3 reduced radiation dose by 50.0% and 42.9% versus Protocol 1, respectively. For perfusion parameters: Tmax SD differed significantly across protocols [F(1.92,222.42)=11.17, P<0.001, partial η²=0.09]; post hoc tests showed significant mean differences between Protocol 1 vs 2 (P<0.016 7) and Protocol 1 vs 3 (P<0.016 7), but not Protocol 2 vs 3. MTT means showed no significant difference [F(1.88,218.11)=0.47, partial η²=0.004] , though SD differed significantly between Protocol 1 vs 2 and 1 vs 3 (P<0.016 7). CBV SD differed significantly [F(1.98,229.25)=22.97, P<0.001, partial η²=0.17] with no mean difference (Friedman test). CBF means showed no significant difference (Bonferroni-adjusted) but SD differed significantly between Protocol 1 vs 2 and 1 vs 3 (P<0.016 7). Ischemic penumbra volumes differed significantly across all protocols (P<0.016 7).
Temporal resolution optimization significantly reduces CTP radiation dose while minimally impacting core perfusion parameters (MTT, CBV, CBF means). CBF remained stable across reduced-dose protocols and may serve as a reliable diagnostic criterion for infarct core delineation.
To compare the effects of 2 different blood flow restriction (BFR) rehabilitation protocols versus classic high-load eccentric resistance training on the morphology and biomechanical properties of the Achilles tendon in healthy adults, and to provide alternative training strategies for preventing Achilles tendon injuries in military training.
A randomized controlled trial was conducted. Fifty-five healthy adults from a university in Xi’an were enrolled between December 20 and 23, 2024, and randomly assigned to one of three groups using a random number table: a high-load eccentric resistance training group (HL, n=18), a low-load eccentric resistance training with calf-applied BFR group (C-BFR, n=16), and a low-load eccentric resistance training with thigh-applied BFR group (T-BFR, n=19). All groups completed an 8-week training program. Achilles tendon cross-sectional area was assessed using color Doppler ultrasonography at baseline, 6 weeks and 8 weeks. Achilles tendon stiffness and Young’s modulus were evaluated using color Doppler ultrasonography and a wireless microFET3 dynamometer. The maximal voluntary isometric contraction (MVIC) of the medial gastrocnemius was simultaneously measured using the wireless microFET3 and surface electromyography. Countermovement jump (CMJ) height and drop jump (DJ) ground contact time were measured using a three-dimensional force platform.
After 8 weeks of intervention, color Doppler ultrasonography assessment revealed that the cross-sectional area of the Achilles tendon significantly increased in all three groups compared to baseline (P<0.01, η²=0.847), with a greater increase in the C-BFR group than in the T-BFR group. Tendon stiffness significantly increased in all three groups compared to baseline (P<0.01, η²=0.722), and the increase was greater in the C-BFR group than in the HL and T-BFR groups. For Young’s modulus, no statistically significant change was observed in the T-BFR group compared to baseline (P>0.05), while both the C-BFR and HL groups showed significant increases (P<0.05, η²=0.181), with a greater increase in the C-BFR group than in the HL and T-BFR groups. The MVIC of the medial gastrocnemius significantly increased in all three groups compared to baseline (P<0.01, η²=0.775), with a greater increase in the C-BFR group than in the T-BFR group. CMJ height significantly increased in all three groups (P<0.01, η²=0.569), with a greater increase in the C-BFR group than in the T-BFR group. DJ ground contact time significantly shortened in all three groups (P<0.01, η²=0.518), and the shortening was greater in the C-BFR group than in the T-BFR group.
Both C-BFR and T-BFR combined with low-load eccentric resistance training can achieve improvements in Achilles tendon morphology, biomechanical properties, and kinematic performance similar to those achieved with classic high-load eccentric resistance training. Furthermore, compared to the T-BFR protocol, the C-BFR protocol demonstrates superior benefits.
To leverage deep learning technology to assist endoscopists in accurately diagnosing colonoscopy images of Crohn’s disease (CD), ulcerative colitis (UC), and Normal tissue.
A total of 24 492 colonoscopy images from 1 309 subjects were collected from Department of Gastroenterology, Daping Hospital, Army Medical University, and Sir Run Run Shaw Hospital between January 2018 and November 2020. The dataset included 4 729 CD images from 424 cases, 7 074 UC images from 605 cases, and 12 689 Normal images from 280 cases. The images were randomly split by case in a 7∶1∶2 ratio into a training set (915 cases, 17 136 images), an internal test set (131 cases, 2 626 images), and an external validation set (263 cases, 4 730 images). To address feature fusion between CNN and Transformer architectures and correct endoscopic image distortion, we constructed a Ring-Feature Fusion Network (RFFNet) based on ResNeSt50 and MViTv2. The network employs bidirectional channel-spatial attention for cross-stage feature fusion, effectively combining the local feature extraction strengths of CNN with the global modeling capabilities of Transformer. An innovative ring-feature mechanism was introduced to handle barrel distortion and depth-of-field phenomena in colonoscopy images, enhancing the model’s adaptability to the geometric properties of the distal intestinal lumen. RFFNet was compared with five existing deep learning models. Recognition results on the external validation set were visualized using confusion matrices, and performance was evaluated using accuracy, sensitivity, specificity, F1-score, and area under the curve (AUC). Ablation studies were conducted on the external validation set using metrics like accuracy and F1-score to validate the effectiveness and necessity of each improvement in RFFNet. Gradient-weighted class activation mapping (Grad-CAM) was used to generate heatmaps, visually demonstrating the model’s focus areas and improving interpretability.
On the external validation set, RFFNet achieved an overall accuracy of 95.68% (95%CI: 94.14% to 98.58%) and an overall AUC of 0.987 (95%CI: 0.985 to 0.990). For CD, UC, and Normal classes, the AUCs were 0.982 (95%CI: 0.979 to 0.986), 0.981 (95%CI: 0.977 to 0.984), and 0.997 (95%CI: 0.995 to 0.998), respectively; sensitivities were 93.35% (95%CI: 91.67% to 94.71%), 94.00% (95%CI: 92.60% to 95.16%), and 98.56% (95%CI: 97.99% to 98.97%), respectively; specificities were 94.58% (95%CI: 93.81% to 95.27%), 94.20% (95%CI: 93.36% to 94.94%), and 98.61% (95%CI: 98.05% to 99.01%), respectively; F1-scores were 87.81% (95%CI: 86.28% to 89.25%), 90.05% (95%CI: 89.00% to 91.25%), and 98.58% (95%CI: 98.22% to 98.92%), respectively. These results indicate that RFFNet achieved high diagnostic precision. Compared to using CNN or Transformer models alone, RFFNet’s overall accuracy improvement was statistically significant (P<0.05). Ablation studies confirmed that the model, through a dynamic spatial attention mechanism, deeply integrates CNN’s local fine-grained feature extraction with Transformer’s global contextual modeling advantages. The CNN-Transformer architecture improved overall accuracy by 0.11% compared to using CNN alone, respectively. The ring-feature mechanism corrected endoscopic optical distortion and depth-of-field attenuation, enhancing modeling of the distal lumen geometry, and improved RFFNet’s overall accuracy by 0.57%. Grad-CAM heatmaps demonstrated the model’s adaptive ability to capture structural features in colonoscopy images.
RFFNet, through its dual-backbone architecture that deeply couples CNN’s local texture perception with Transformer’s global dependency modeling, and explicit correction of endoscopic distortion via ring-features, enables high-accuracy real-time classification of CD, UC, and normal mucosa.
To develop and validate a deep learning model for real-time identification and localization of 17 common abdominal anomalies on ultrasound images, providing reliable technical support for primary care sonography.
Retrospectively collected 39 530 B-mode ultrasound images from 18 947 patients with abdominal anomalies at the Second Affiliated Hospital of Army Medical University from January 2017 to September 2024). A stratified random sample of 1 891 images formed an independent testing set; the remaining 37 639 images were split 7:1:2 into training (26 258), validation (3 840), and internal test (7 541) sets. After standardized annotation, RetinaNet, Faster R-CNN, RT-DETR, and YOLOv13 were compared on the internal test set using IoU, precision, recall, F1-score, and mAP@50. The optimal model was further evaluated against 6 sonographers (3 novice, 3 intermediate) and novice sonographers assisted by the model on the independent test set.
YOLOv13 achieved superior performance on the internal testing set: mAP@50 0.653 (95%CI:0.608 to 0.669) versus RetinaNet 0.546 (95%CI:0.511 to 0.563), Faster R-CNN 0.578 (95%CI:0.531 to 0.591), RT-DETR 0.627 (95%CI:0.585 to 0.643); precision 0.633 (95%CI:0.624 to 0.643), recall 0.715 (95%CI:0.704 to 0.725), F1-score 0.671 (95%CI:0.663 to 0.681). On the independent testing set (IoU=0.5), YOLOv13’s F1-score (0.624) exceeded all novice sonographers’ mean F1-score (0.484; P<0.001). Model assistance improved novice sonographers’ mean F1-score from 0.484 to 0.644 (33.1% improvement; P<0.001), approaching intermediate-level performance (0.656); mean precision increased 20.6% (0.529 to o 0.638), mean recall increased 42.8% (0.472 to 0.674), and mean interpretation time decreased 24.4% (213.33 min to 161.33 min).
The YOLOv13 model demonstrates optimal performance for detecting 17 abdominal ultrasound anomalies and significantly enhances diagnostic efficiency of novice sonographers, serving as an effective assistive tool for abdominal ultrasound screening in primary care settings.
To investigate changes in the depressive disorder burden in China, global, and across socio-demographic index (SDI) regions during the COVID-19(from 2019 to 2021) compared with 1990 to 2019, and to analyze sex-regional disparities.
Using depressive disorder data from the Global Burden of Disease (GBD) database (from 1990 to 2021) for China, global populations, and 5 SDI regions, we characterized depressive disorder burden distribution during the pandemic via metrics including age-standardized prevalence rate (ASPR), age-standardized disability-adjusted life years (ASR-DALYs), and estimated annual percentage change (EAPC). Joinpoint regression modeled temporal trends, and the Bayesian age-period-cohort (BAPC) model estimated the excess depressive disorder burden attributable to COVID-19 in 2020—2021.
China’s depressive disorder ASPR increased to 2 875.68/100 000 person-years (95%CI: 2 589.96 to 3 203.43) in 2021 from 2 825.00/100 000 (95%CI: 2 550.42 to 3 136.08) in 2019; ASR-DALYs rose to 430.61/100 000 (95%CI: 305.24 to 586.21) from 430.14/100 000 person-years (95%CI: 303.92 to 580.82). Joinpoint analysis identified an inflection point for China’s ASPR in 2019, with a significant upward annual percentage change (APC=0.95, P<0.05) from 2019 to 2021—a smaller increase than global and other SDI regions. No significant inflection was observed for China’s ASR-DALYs, whereas other regions showed clear upward inflection points near 2019 (P<0.05). China exhibited the lowest excess depressive disorder burden due to COVID-19: excess ASPR was 121.67/100 000 and excess ASR-DALYs was 12.25/100 000 person-years in 2021, only 28% and 14% of global levels, respectively. Females had higher ASPR/ASR-DALYs and excess burden than males, with females in high-SDI regions bearing the heaviest burden.
COVID-19 imposed severe excess depressive disorder burdens worldwide. China’s effective pandemic response policies and psychological assistance measures mitigated this burden. Countermeasures Governments should increase investment in mental healthcare resources, enhance public mental health literacy, and ensure rapid implementation of containment measures during public health emergencies—including regular information updates to stabilize public sentiment—alongside community-based psychological support.
To analyze the status and trends of the burden of adverse effects of medical treatment (AEMT) in China from 1990 to 2023 using open data from the Global Burden of Disease (GBD) 2023.
Based on the GBD 2023 data, we extracted the number of incident cases, prevalent cases, disability-adjusted life years (DALYs), and deaths due to AEMT in China from 1990 to 2023. We also extracted age-standardized rates, including the age-standardized incidence rate (ASIR), age-standardized prevalence rate (ASPR), age-standardized DALYs rate (ASDR), and age-standardized mortality rate (ASMR). The annual percent change (APC), average annual percent change (AAPC), and their 95% confidence intervals (95% CI) for ASIR, ASPR, ASDR, and ASMR were calculated to analyze trends in the burden of AEMT in China. Epidemiological characteristics and differences in the AEMT burden among different age and sex groups in China were compared for the years 1990, 2020, and 2023.
From 1990 to 2023, the number of incident cases, prevalent cases, DALYs, and deaths due to AEMT in China changed from 432 784 (95%UI:347 489 to 526 244), 33 087 (95%UI: 24 103 to 42 381), 686 404(95%UI:398 451 to 1 086 894), and 11 426(95%UI:7 233 to 17 456) to 342 349(95%UI: 279 429 to 423 443), 26 165(95%UI: 19 529 to 33 662), 121 245(95%UI: 90 707 to 155 623), and 3 337(95%UI: 2 494 to 4 212) in 2020, and further to 372 069 (95%UI: 303 152 to 464 618), 28 436 (95%UI: 21 029 to 36 643), 107 384(95%UI, 79 527 to 137 417), and 3 128 (95%UI: 2 305 to 3 985) in 2023. From 1990 to 2023, the AAPC values for ASIR, ASPR, ASDR, and ASMR were -0.994 8%(95%CI: -1.199 7% to -0.789 5%), -0.993 3%(95%CI: -1.202 2% to -0.783 9%), -5.968 4%(95%CI: -6.195 9% to -5.740 3%), and -5.432 2% (95%CI: -5.658 8% to -5.205 1%), respectively (P<0.05). In 1990, incidence and mortality rates were higher in younger age groups and among females compared to males, while rates were higher among older males compared to older females. In 2020 and 2023, incidence was higher among females than males in younger age groups and higher among older males than older females; the differences in incidence rates between males and females within the same age group and year narrowed compared to 1990, while mortality among older males remained higher than among females in the same periods.
From 1990 to 2023, key indicators of the AEMT burden in China-incidence, prevalence, DALYs, and mortality-showed an overall declining trend. Among the elderly population, AEMT incidence and mortality increased with age, and elderly males were more likely to die from AEMT than elderly women. Countermeasures Enhancing investment in medical training and resources to improve the capability for diagnosing and treating AEMT, continuously strengthening AEMT prevention and control, and implementing tailored interventions for the elderly, particularly elderly males.