在维普、万方、知网、Embase、PubMed和Web of Science数据库中检索2019～2023年间有关机器学习方法预测抑郁症发病风险的文献，系统性地总结这些算法的特点、研究领域、模型效能和当前应用所面临的问题和挑战。研究共纳入92篇文献，结果显示，机器学习预测抑郁症发病风险的模型效果较好，最佳预测模型的AUC值为0。6030～0。9976。未来应当建立多中心、前瞻性的融合多模态的动态预测模型，为抑郁症的临床诊断提供更可靠的依据。
Review on machine learning methods in predicting the risk of depression
The articles on machine learning methods for predicting the risk of depression between 2019 and 2023 are retrieved from 6 databases(VIP,WANFANG,CNKI,Embase,PubMed and Web of Science).The review systematically summarized the algorithm characteristics,research fields,model performance,and current problems and challenges.A total of 92 articles are includes.The analysis results show that the machine learning models for predicting the risk of depression perform well,with the AUC values of the best prediction models ranging from 0.603 0 to 0.997 6.In the future,there should be a construction of multicenter prospective dynamic prediction models that use a multi-modal fusion approach to provide a more reliable basis for the clinical diagnosis of depression.
depressionmachine learningdeep learningnatural language processingprediction modelreview
NETL
NSTL
万方数据
