基于编码器和注意力机制的睡眠呼吸障碍多分类方法OA
Research on Multi-Class Classification Method for Sleep-Disordered Breathing Based on Encoder and Attention Mechanism
睡眠呼吸暂停(Sleep Apnea,SA)是一种常见的睡眠障碍.传统多导睡眠图(Polysomnography,PSG)虽然是诊断SA的黄金标准,但耗时且昂贵.为解决该问题,文中提出了一种基于心电图(Electrocardiogram,ECG)和血氧饱和度(Blood Oxygen Saturation,SpO2)信号的新型检测方法.分析了 ECG和SpO2特征,利用多尺度卷积神经网络模型(Multi-Scale Convolutional Neural Network,MSCNN)结合 Encoder-SE(Squeeze-and-Excitation)网络模型进行特征训练与分类.MSCNN通过获取不同时间长度的ECG和SpO2特征量来增强对分析信号的分析效果.Encoder-SE网络进一步提升了特征的表达能力,通过SE模块自适应地调整特征重要性,确保模型关注关键特征.实验结果表明,所提方法的平均准确率为93.29％,为SA的临床诊断和治疗提供了新思路与有效工具.
SA(Sleep Apnea)is a common sleep disorder.Although the traditional PSG(Polysomnography)is the gold standard for diagnosing SA,it is time-consuming and expensive.To address this issue,this study proposes a novel detection method based on ECG(Electrocardiogram)and SpO2(Blood oxygen saturation)signals.The features of ECG and SpO2 are analyzed,and a MSCNN(Multi-Scale Convolutional Neural Network)model combined with an Encoder-SE(Squeeze-and-Excitation)network model is utilized for feature training and classification.MSCNN enhances the analysis effect of the ana-lyzed signals by obtaining ECG and SpO2 feature quantities of different time lengths.The Encoder-SE network further im-proves the feature representation ability.Through the SE module,it adaptively adjusts the importance of features to ensure that the model focuses on key features.The experimental results show that the proposed method has an average accuracy of 93.29％,providing new ideas and effective tools for the clinical diagnosis and treatment of SA.
楼利军;何晓玉;蒋明峰
浙江理工大学信息科学与工程学院,浙江 杭州 310018浙江理工大学计算机科学与技术学院(人工智能学院),浙江 杭州 310018浙江理工大学计算机科学与技术学院(人工智能学院),浙江 杭州 310018
信息技术与安全科学
睡眠呼吸暂停综合征深度学习心电图外周血氧饱和度多模态卷积神经网络特征融合编码器注意力机制
sleep apnea syndromedeep learningelectrocardiogramperipheral blood oxygen saturationmultimodal convolutional neural networkfeature fusionencoderattention mechanism
《电子科技》 2026 (1)
73-80,8
国家重点研发计划(2023YFE0205600)National Key R&D Program of China(2023YFE0205600)
评论