基于唾液炎症因子水平预测牙周状况的机器学习模型的建立

Establishment of machine learning models for the diagnosis of gingivitis and periodontitis based on the level of inflammatory factors in saliva

目的建立基于唾液炎症因子水平的牙龈炎、牙周炎诊断机器学习模型。方法Luminex液相芯片检测牙周健康对照组(41人)、牙龈炎患者组(60人)和牙周炎患者组(54人)唾液中22种炎症因子水平,Spearman分析选择P<0.05的炎症因子建立六种机器学习模型,比较它们在区分牙周健康者、牙龈炎患者和牙周炎患者中的诊断性能。结果支持向量机(SVM)、PSO改良支持向量机(PSO-SVM)、GA改良支持向量机(GA-SVM)模型准确率为100%;深度学习(BP)和GA改良深度学习(GA-BP)模型准确率为87.10%,Fisher分类判别模型(LDA)准确率为83.87%。结论SVM模型准确率和运行时间最佳,可判断牙周健康状况,机器学习模型可能成为牙龈炎和牙周炎诊断新方法。

Objective To establish machine learning models for the diagnosis of gingivitis and periodontitis based on the levels of inflammatory factors in saliva.Methods Twenty-two inflammatory factors levels in saliva of periodontally healthy controls(n=41),patients with chronic gingivitis(n=60)and patients with chronic periodontitis(n=54)were detected by Luminex technology kit.Spearman analysis determined that those with P<0.05 were used to establish six machine learning classification models and their diagnostic performance in distinguishing periodontally healthy subjects,patients with chronic gingivitis and patients with chronic periodontitis were compared.Results The accuracy of support vector machine(SVM)model,PSO modified support vector machine(PSO-SVM)model and GA modified support vector machine(GA-SVM)model were 100%,the accuracy of deep learning(BP)model and GA modified deep learning(GA-BP)model were 87.10%,and the accuracy of Fisher classification discriminant model(LDA)was 83.87%.Conclusions The SVM model has the best accuracy and running time and can be used to predict periodontal health status.The machine learning classification model may become a new method for gingivitis and periodontitis diagnosis in the future.