BACKGROUND:In recent years,deep learning technologies have been increasingly applied in the field of oral medicine,enhancing the efficiency and accuracy of oral imaging analysis and promoting the rapid development of intelligent oral medicine. OBJECTIVE:To elaborate the current research status,advantages,and limitations of deep learning based on oral imaging in the diagnosis and treatment decision-making of oral diseases,as well as future prospects,exploring new directions for the transformation of oral medicine under the backdrop of deep learning technology. METHODS:PubMed was searched for literature related to deep learning in oral medical imaging published from January 2017 to January 2024 with the search terms"deep learning,artificial intelligence,stomatology,oral medical imaging."According to the inclusion criteria,80 papers were finally included for review. RESULTS AND CONCLUSION:(1)Classic deep learning models include artificial neural networks,convolutional neural networks,recurrent neural networks,and generative adversarial networks.Scholars have used these models in competitive or cooperative forms to achieve more efficient interpretation of oral medical images.(2)In the field of oral medicine,the diagnosis of diseases and the formulation of treatment plans largely depend on the interpretation of medical imaging data.Deep learning technology,with its strong image processing capabilities,aids in the diagnosis of diseases such as dental caries,periapical periodontitis,vertical root fractures,periodontal disease,and jaw cysts,as well as preoperative assessments for procedures such as third molar extraction and cervical lymph node dissection,helping clinicians improve the accuracy and efficiency of decision-making.(3)Although deep learning is promising as an important auxiliary tool for the diagnosis and treatment of oral diseases,it still has certain limitations in model technology,safety ethics,and legal regulation.Future research should focus on demonstrating the scalability,robustness,and clinical practicality of deep learning,and finding the best way to integrate automated deep learning decision support systems into routine clinical workflows.

Objective : To explore the relationship between single nucleotide polymorphisms ( SNPs) in D-loop region of mitochondrial DNA ( mtDNA) and mtDNA copy number and the risk of dermatomyositis ( DM) ，and its in- fluencing factors． Methods : 74 patients with DM and 92 healthy controls were included in the study． Genomic DNA was extracted from peripheral blood and the target fragment of mtDNA D-loop region was amplified by PCR technique，and the products were subsequently sequenced．Serum levels of ROS were assessed using a high-sensi- tivity reactive oxygen species detection kit．The expression levels of cytokines，interleukin ( IL) -5，IL-13，inter- feron-γ ( IFN-γ) ，IL-2，IL-6，IL-10，tumor necrosis factor-α ( TNF-α) and IL-4 were measured using Flow Fluo- rescence Immunmicrobeads Assay．Wilcoxon rank-sum test was used to assess the potential correlation between cy- tokines and SNPs associated with DM risk．The relative copy number of mtDNA was measured using quantitative re- al-time polymerase chain reaction ( qPCR) analysis． Results : Two SNPs ( 16304T / C，16519T / C) were found to be associated with the risk of developing DM，and alleles 16304C ( χ2 = 4. 937，P = 0. 026) and 16519C ( χ2 = 4. 405，P = 0. 036) in the mitochondrial D-loop region were confirmed to be associated with DM development risk． The DM risk-associated allele 16304C was significantly associated with lower IL-4 expression ( P = 0. 016) ．The mtDNA copy number was significantly higher in DM patients than in controls ( P ＜0. 001) ． Conclusion Mitochondrial D-loop SNPs can be potential biomarkers for DM risk，and SNPs may be involved in DM by influencing cytokines．DM shows high expression of mtDNA copy number，and the increase in mtDNA copy number may lead to mitochondrial dysfunction，which triggers the pathogenesis of DM．