基于机器学习的儿童阻塞性睡眠呼吸暂停诊断方法

Diagnostic Method for Pediatric Obstructive Sleep Apnea based on Machine Learning

目的建立一个机器学习模型,利用夜间心率和血氧数据来诊断儿童阻塞性睡眠呼吸暂停(OSA)。方法使用3150名疑似患有OSA的儿童的多导睡眠图数据构建训练集和测试集,利用年龄、性别、体重指数、3%氧减指数、最低血氧饱和度指数和平均血氧饱和度指数作为诊断特征,使用机器学习算法CatBoost建立诊断模型。结果诊断模型在呼吸暂停低通气指数≥5和呼吸暂停低通气指数≥10的分类任务中,分类准确率分别为85.67%和89.81%,曲线下面积AUC值分别为0.84和0.87,表现优于线性判别分析模型。结论本研究提供了一种利用机器学习模型诊断儿童OSA的方法。这种方法减少了诊断所需的信号数量,并展现出较高的可靠性。

Objective This study aimed to develop a machine learning model using nighttime heart rate and blood oxygen data to diagnose pediatric obstructive sleep apnea(OSA).Methods A training set and a test set were constructed using polysomnography(PSG)data from 3150 children suspected of having OSA.Age,gender,body mass index,3%oxygen desaturation index,minimum SpO2,and average SpO2 were utilized as diagnostic features.The CatBoost machine learning algorithm was employed to build the diagnostic model.Results The diagnostic model achieved classification accuracies of 85.67%and 89.81%for tasks involving apnea-hypopnea index(AHI)≥5 and AHI≥10,respectively.The corresponding AUC values were 0.84 and 0.87,outperforming the linear discriminant analysis model.Conclusion This study provided a method for diagnosing pediatric OSA using a machine learning model.This approach reduced the number of required signals for diagnosis and demonstrated higher reliability.