Recent advances in large models demonstrate significant prospects for transforming the field of medical imaging. These models, including large language models, large visual models, and multimodal large models, offer unprecedented capabilities in processing and interpreting complex medical data across various imaging modalities. By leveraging self-supervised pretraining on vast unlabeled datasets, cross-modal representation learning, and domain-specific medical knowledge adaptation through fine-tuning, large models can achieve higher diagnostic accuracy and more efficient workflows for key clinical tasks. This review summarizes the concepts, methods, and progress of large models in medical imaging, highlighting their potential in precision medicine. The article first outlines the integration of multimodal data under large model technologies, approaches for training large models with medical datasets, and the need for robust evaluation metrics. It then explores how large models can revolutionize applications in critical tasks such as image segmentation, disease diagnosis, personalized treatment strategies, and real-time interactive systems, thus pushing the boundaries of traditional imaging analysis. Despite their potential, the practical implementation of large models in medical imaging faces notable challenges, including the scarcity of high-quality medical data, the need for optimized perception of imaging phenotypes, safety considerations, and seamless integration with existing clinical workflows and equipment. As research progresses, the development of more efficient, interpretable, and generalizable models will be critical to ensuring their reliable deployment across diverse clinical environments. This review aims to provide insights into the current state of the field and provide directions for future research to facilitate the broader adoption of large models in clinical practice.
Humans ; Diagnostic Imaging/methods* ; Precision Medicine/methods* ; Image Processing, Computer-Assisted/methods*

Objective To observe the value of conventional ultrasound(US)and contrast-enhanced ultrasound(CEUS)for diagnosing partially cystic papillary thyroid carcinomas(PCPTC).Methods Thirty patients with PCPTC(32 nodules)and 50 patients with benign partially cystic thyroid nodules(PCTN)(52 nodules)were retrospectively enrolled.US and CEUS manifestations were compared between groups,and the efficacy of US and US combined with CEUS for diagnosing PCPTC were analyzed.Results Significant differences of margin,solid part echo,liquid part location,microcalcification and blood flow distribution of the nodules showed with US,as well as enhancement phase,enhancement intensity,enhancement uniformity,peripheral high-enhancement ring and regression mode of lesions shown on CEUS were found between groups(all P＜0.05).The area under the curve(AUC)of US combined with CEUS for diagnosing PCPTC was 0.942,significantly higher than that of US alone(0.750,P＜0.001).Conclusion US had certain value for diagnosing PCPTC.Combining with CEUS could effectively improve diagnostic efficacy of US for PCPTC.

Objective The relationship between asthma and hypertension was explored by using the least absolute shrinkage and selection operator(LASSO)regression and Mendelian randomization(MR).Methods Data from the National Health and Nutrition Examination Survey(NHANES)of the United States from 2001 to 2018 were selected.LASSO regression was used to screen variables,and a nomogram prediction model for hypertension risk was established.The model was verified to evaluate its predictive ability and stability for the risk of hypertension.Finally,two-sample MR analysis was used to verify the causal relationship.Results The study ultimately included 26 456 cases,including 8833 hypertensive patients.LASSO regression screened out 9 variables,including age,educational level,race,marital status,smoking history,asthma,diabetes,cancer and body mass index,and a nomogram model was constructed based on these variables.The calibration curve indicated that there was a good fit between two groups.After adjusting for the measured confounding factors,the inverse variance weighted analysis showed that asthma was associated with an increased risk of hypertension(OR=2.67,95％CI:1.31-5.43,P＜0.01).Conclusion There is a causal relationship between asthma and hypertension,and controlling asthma conditions may help reduce the risk of developing hypertension.

Objective:To construct four recombinant hemagglutinin (HA) antigens from seasonal influenza viruses and evaluate their immunogenicity in mouse models.Methods:HA coding sequences from four seasonal influenza virus strains Wisconsin (H1N1), Darwin (H3N2), Austria (B/Victoria lineage, BV) and Phuket (B/Yamagata lineage, BY) were optimized and synthesized, and then used to construct four recombinant plasmids. Recombinant baculoviruses were obtained through transformation and transfection. The expression of recombinant HA antigens was identified by SDS-PAGE and Western blot. The recombinant HA antigens were purified by nickel column affinity chromatography and intramuscularly administered to BALB/c mice after formulation with Al(OH) 3 or AddaVax adjuvant. Humoral immune responses were assessed by indirect ELISA and hemagglutination inhibition test, while cellular immune responses were evaluated by ELISPOT. Microneutralization test was used to detect the titers of serum antibodies in mice. Statistical analysis was performed using t test or non-parametric rank sum test. Results:PCR amplification and agarose gel electrophoresis confirmed the correct construction of the recombinant bacmids. Western blot showed verified the successful expression of the four recombinant antigens (H1-HA, H3N2-HA, BV-HA, and BY-HA). SDS-PAGE results showed that the purity of all four recombinant HA antigens exceeded 95%. After three-dose immunization, the total IgG levels in mice immunized with the recombinant H1N1-HA, H3N2-HA, or BV-HA formulated with AddaVax adjuvant were higher than those in the corresponding groups immunized with the same recombinant antigen alone (all P<0.05). The secretion levels of IFN-γ, IL-2, and IL-4 in the group receiving the mixture of all four recombinant HA antigens formulated with AddaVax adjuvant were higher than those in the group immunized with a commercial quadrivalent split influenza vaccine (all P<0.01). Results of the microneutralization test showed that the antibody titer in the quadrivalent split influenza vaccine group was 1∶225, whereas the titer in the group immunized with the mixture of four recombinant HA antigens formulated with AddaVax adjuvant could reach up to 1∶1 200. Conclusions:In this study, four recombinant seasonal influenza virus HA antigens are successfully expressed and demonstrated good immunogenicity in mice when formulated AddaVax adjuvant.

Alzheimer's disease(AD), a neurodegenerative disorder associated with aging, is the most prevalent form of dementia.As the aging population continues to expand, AD presents significant health and caregiving challenges for families and society, making it a pressing international public health concern.In recent years, numerous countries have implemented dementia prevention and treatment strategies that emphasize community-based comprehensive approaches.Currently, the community-based AD prevention and treatment model in China is still in the exploratory phase, with community efforts lacking organization.In alignment with China's action plan for advancing dementia prevention and treatment, and to achieve the strategic objective of "healthy aging, " this consensus is based on the principle of three-level prevention and is tailored to the characteristics of AD disease progression.It aims to develop a comprehensive prevention and treatment strategy for AD that is suitable for communities in China, providing technical guidance and support to establish a scientific basis for formulating community AD prevention and treatment models.

Objective:To integrate implicit templates with deep learning techniques, a novel neural network, the tooth-deformable deep implicit network (T-DDIN), was constructed to achieve high-precision shape completion of tooth defects in a personalized manner.Methods:A total of 550 intraoral scan models were collected from patients treated at the Department of Orthodontics and Department of Prosthodontics, West China Hospital of Stomatology, Sichuan University (500 for training and 50 for testing), between March 2022 and March 2024. T-DDIN reconstructed defective tooth morphology using an implicit template and a latent encoding prediction network. During model evaluation, Class Ⅱ cavity defects and occlusal wear defects were simulated in the test set. Morphological restoration was performed using both traditional computer aided design (CAD) methods and the T-DDIN deep learning approach. The two methods were compared based on three-dimensional deviation, occlusal adjustment volumes, cusp angle deviation, and restoration time.Results:The T-DDIN group demonstrated significantly lower three-dimensional deviation for Class Ⅱ cavity defects and occlusal wear restoration [(0.14±0.05) and (0.16±0.09) mm], occlusal adjustment volumes [(0.44±0.03) and (0.49±0.03) mm 3], and difference value of the tooth cusp angles (5.69°±1.90° and 6.04°±0.53°) compared to the traditional CAD group (both P<0.001). No significant differences were observed within the T-DDIN group between the two defect types in terms of three-dimensional deviation ( P=0.098) or occlusal adjustment volume ( P=0.154) or difference value of the tooth cusp angles ( P=0.196). However, in the traditional CAD group, three-dimensional deviation, occlusal adjustment volume and difference value of the tooth cusp angles was significantly higher in occlusal wear restorations than in Class Ⅱ cavity defects restorations ( P<0.001). The T-DDIN group, which involved Class Ⅱ cavity defects and occlusal wear, demonstrated significantly less recovery time of morphology (37.2±7.7) and (39.4±6.2) s compared to the traditional CAD group ( P<0.001). Conclusions:T-DDIN demonstrated superior stability and accuracy in morphological reconstruction for various types of dental defects while significantly reducing restoration time.

Objectives:To evaluate the effect of extended single prone positioning ventilation on survival and weaning rate of acute respiratory distress syndrome (ARDS) patients supported by VV-ECMO.Methods:ARDS patients supported by VV-ECMO admitted to Jinhua Central Hospital, the Fourth Affiliated Hospital of Zhejiang University School of Medicine and the First Hospital of Jiaxing from September 2014 to May 2025 were retrospectively enrolled into the study. The clinical data, ECMO and ventilator related parameters and outcomes of the patients were collected. The patients were divided into the extended prone positioning group and prone positioning group according to whether the duration of prone position ventilation was greater than 24 h. The clinical data of the two groups were compared to explore the effects on 30-day survival in-hospital survival and ECMO withdraw rate of these patients. Multivariate logistic regression analysis was used to explore the relationship between the duration of single prone position ventilation and the success of ECMO weaning, 30-day survival and hospital survival.Results:Total of 163 ARDS patients supported by VV-ECMO receiving prone positioning ventilation were included in study, 64 in extended prone positioning group and 72 in prone positioning group. The 30-day survival (54.7% vs. 52.8%) in-hospital survival (51.6% vs. 48.6%) and ECMO withdraw rate (57.8% vs. 61.1%) between the two groups were not statistically different ( P>0.05) as well as the duration of ECMO support [12(10,15)d vs. 11(10,13)d] the duration of ventilation [16(13,18)d vs. 16(12,18)d] the duration of ICU stay [26(15,32)d vs. 26(19,29)d] and the duration of hospital stay [32(15,42)d vs. 34(28,35)d]. Logistic regression analysis revealed that the duration of each prone position ventilation was not associated with successful weaning ( OR=0.979, 95% CI:0.952-1.006), 30-day survival ( OR=1.015, 95% CI: 0.975-1.056) and hospital survival ( OR=1.014, 95% CI: 0.974-1.055) even after adjusting for the severity of illness, age, and type of pneumonia. Conclusions:For ARDS patients supported by VV-ECMO, extended single prone positioning ventilation for more than 24 hours neither increase 30-day survival in-hospital survival and successful ECMO weaning rate, nor shorten ECMO support duration.

Objective To explore the characteristics of tumor immune microenvironment of pancreatic ductal adenocarcinoma(PDAC)based on the differentially expressed genes associated with macrophages in early and late tumor tissues of PDAC patients,so as to provide new targets for early diagnosis and treatment of PDAC.Methods Three early-stage and 4 late-stage tumor samples were collected from PDAC patients.Tissue transcriptome data were analyzed by single-cell sequencing technology.With The Cancer Genome Atlas(TCGA)database,macrophage related genes differentially expressed in early and late PDAC related to prognosis were obtained.The prognostic risk scoring model was constructed using the least absolute shrinkage and selection operator(LASSO)-Cox method.The risk genes associated with PDAC prognosis were screened using the Gene Expression Profiling Interaction Analysis 2(GEPIA2)online survival analysis tool.Their immune infiltration was analyzed using the cell-type identification by estimating relative subsets of RNA transcripts(CIBERSORT)method,and their expression in different subpopulations of macrophages at different periods was resolved by t-distributed stochastic nearest-neighbour embedding(tSNE)clustering downscaling and the Monocle package.The protein-protein interactions and immunohistochemistry were further analyzed with the help of STRING database and The Human Protein Atlas(HPA)database.Results A total of 48 differentially expressed macrophage-related genes in early and late PDAC were found to be associated with prognosis.Eleven differentially expressed macrophage genes were screened as PDAC prognostic genes,and they were used to construct a prognostic risk scoring model for PDAC.Two prognostic genes with significant risk indicators were screened:2'-5'-oligoadenylate synthetase 2(OAS2)was mainly expressed in secreted phosphoprotein 1(SPP1)-positive macrophages,and assembly factor for spindle microtubules(ASPM)was mainly expressed in proliferation marker protein Ki-67(MKI67)-positive macrophages.There were significant differences in the expression of OAS2 in PDAC macrophages at different stages.Immune infiltration studies showed that OAS2 and ASPM were highly expressed in regulatory T cells and M2-type macrophages(both P＜0.05).Protein-protein interaction network showed the expression relationship between OAS2 and ASPM and other co-expressed proteins.Immunohistochemistry results showed that the expression of OAS2 was higher in tumor tissues than in normal tissues.Conclusion OAS2 and ASPM are both poor prognostic genes that are specifically expressed and exert negative immune effects in SPP1+macrophages and MKI67+macrophages,respectively,promoting the progression of PDAC and ultimately leading to a poor prognosis,and it is expected to provide new targets for the prevention and treatment of PDAC.

Objective:To analyze the effects and mechanisms of ginseng polysaccharide in myocardial fibrosis and endothelial function in coronary heart disease model rats based on mitogen activated extracellular signal regulated kinase (MEK) and extracellular signal regulated kinase (ERK) signaling pathways.Methods:Totally 60 SD rats were divided into control group (10 rats) and model group (50 rats). A coronary heart disease model was established in modeling rats. The successfully modeled rats were divided into model group, drug control group, ginseng polysaccharide high- and low-dosage groups, with 10 rats in each group. The drug control group was orally administered with 2 mg/kg Lipitor solution, while the ginseng polysaccharide low- and high-dosage groups were orally administered with 100 and 200 mg/kg ginseng polysaccharide solution. The control group and model group were orally administered with equal volumes of physiological saline solution once a day for one week. HE staining and Masson staining were used to observe the morphological changes and pathological morphology of myocardial tissue. ELISA was used to detect the levels of NO, eNOS, ET-1, ADMA, TNF-α, CRP, and IL-6. Immunochemiluminescence was used to detect the expression levels of CK-MB and cTnl. Real time fluorescence quantification was used to identify the expression of MEK and ERK. Western blot was used to detect the expression of MEK, ERK, p-MEK, and p-ERK in myocardial tissue.Results:Compared with the model group, the levels of serum NO and eNOS increased in the drug control group and ginseng polysaccharide low- and high-dosage groups ( P<0.05), while the levels of ET-1, ADMA, TNF-α, CRP, IL-6, CK-MB, and cTnl decreased ( P<0.05), and the mRNA and protein expressions of MEK and ERK in myocardial tissue decreased ( P<0.05). Conclusions:Ginseng polysaccharides can improve the degree of myocardial fibrosis in rats, alleviate endothelial function and myocardial injury. Its mechanism may be related to the inhibition of the MEK/ERK signaling pathway.

Objective: To investigate the epidemiological and spatio-temporal characteristics of influenza in Jiaxing City, Zhejiang Province, so as to provide insights into perfecting the prevention and control strategies of influenza. Methods: Data of influenza in Jiaxing City from 2019 to 2023 were collected from the Chinese Disease Prevention and Control Information System. Population data of the same period were collected from the Zhejiang Health Information Network Reporting System. The epidemiological characteristics of influenza were analyzed using descriptive analysis. Vector map information was collected from the Open Street Map, and the spatio-temporal clustering characteristics of influenza were analyzed using spatial autocorrelation and spatio-temporal scanning. Results: A total of 181 501 cases of influenza were reported in Jiaxing City from 2019 to 2023, with an average annual reported incidence of 653.93/105. The majority of cases were aged 5 to <15 years (59 785 cases, 32.94%). The majority of the occupations were students (78 239 cases, 43.11%) and pre-school children (33 715 cases, 18.58%). The county (city, district) with the highest reported incidence was Haining City (1 451.70/105), and the town (street) with the highest reported incidence was Chang'an Town (1 932.78/105). Spatial autocorrelation analysis showed that the incidence of influenza in Jiaxing City from 2019 to 2023 had positive spatial correlations (all Moran's I>0, all P<0.05), with a high-high clustering in the southern region. Spatio-temporal scanning analysis showed that there was a spatio-temporal clustering of influenza in Jiaxing City from 2019 to 2023, with the southern region being the primary-type clustering area and the period between November and January of the following year being the clustering time. Conclusion There was a significant spatio-temporal clustering of influenza in Jiaxing City from 2019 to 2023, with winter being the peak season and the southern region being the primary area.