基于卷积神经网络的睡眠呼吸暂停综合征严重程度4分类预测  

Classification prediction of sleep apnea syndrome severity based on convolutional neural network

作　　者：邹汉荣 何汉武[1] ZOU Hanrong;HE Hanwu(College of Mechanical and Electrical Engineering,Guangdong University of Technology,Guangzhou,Guangdong 510006,China)

机构地区：[1]广东工业大学机电工程学院,广东广州510006

出　　处：《中国医学工程》2022年第8期1-6,共6页China Medical Engineering

摘　　要：目的睡眠呼吸暂停综合征(SAS)在我国是一种常见病,但由于医院设备的不完善,该病的确诊率只有不到10%。本文提出了一种仅使用整段血氧饱和度信号并结合人体基本体征数据,通过卷积神经网络(CNN)方法快速预测SAS严重程度,以提高SAS确诊率。方法利用某医院提供的睡眠监测数据和医学诊断报告,构建深度卷积神经网络。输入被试的夜晚监测的整段血氧饱和度信号和基本的体征数据,如年龄、性别、身高、体重和体质量指数(BMI),深度卷积神经网络根据被试的信号数据和体征数据预测SAS严重程度。结果实验的结果表明,该方法通过比较预测结果与医学诊断报告结果,预测SAS严重程度总体准确率达到了86.39%,正常类别的准确率为90.82%,轻度类型的准确率为76.29%,中度类型的准确率为75.47%,重度类型的准确率为98.89%。结论本文提出的方法可以实现SAS的快速预测。该方法能够自动提取血氧饱和度特征和人体体征特征,不需要依赖专家知识。与常见的方法不同,不需要对血氧饱和度信号进行分段,使用的是夜晚睡眠7 h的整段血氧饱和度信号。因为只需要血氧饱和度信号和基本体征数据,更适合医疗条件有限的二三线城市医院,更加方便、快捷,具有更好的泛用性,能够更好地提高SAS的确诊率。【Objective】Sleep apnea syndrome(SAS)is a common disease in China,but due to inadequate hospital equipment,the diagnosis rate of the disease is less than 10%.In this paper,a convolutional neural network method is proposed to rapidly predict the severity of SAS using only the whole blood oxygen saturation signal combined with the data of basic human signs,so as to improve the diagnosis rate of SAS.【Methods】Based on the sleep monitoring data and medical diagnosis report provided by a hospital,deep convolutional neural network is constructed.The whole blood oxygen saturation signal and basic physical sign data,such as age,sex,height,weight and BMI,were input into the deep convolutional neural network to predict the severity of SAS according to the signal data and physical sign data of the subjects.【Results】Experimental results showed that by comparing the prediction results with the results of medical diagnosis report,the overall accuracy of predicting the severity of SAS is 86.39%,the accuracy of normal type is 90.82%,the accuracy of mild type is 76.29%,the accuracy of moderate type is 75.47%,and the accuracy of severe type is 98.89%.【Conclusion】The method proposed in this paper can realize SAS fast prediction.This method can automatically extract the characteristics of blood oxygen saturation and human body signs without relying on expert knowledge.Different from the common method,there is no need to segment the blood oxygen saturation signal,and the whole blood oxygen saturation signal of 7 h sleep at night is used.Because only blood oxygen saturation signal and basic sign data are required,it is more suitable for hospitals in second and third-tier cities with limited medical conditions.It is more convenient,fast and has better universality,and can better improve the diagnosis rate of SAS.

关 键 词：深度学习 卷积神经网络 睡眠呼吸暂停综合征 计算机辅助诊断 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] R766[自动化与计算机技术—控制科学与工程]