A continuous temporal modeling framework based on single-lead ECG for sleep apnea detection
LIU Yang
ZHOU Ruiqi
FANG Zhen
CHEN Xianxiang
Abstract：To address the limitation of existing deep learning methods that often treat electrocardiogram(ECG)signals as isolated segments and ignore temporal context information,we proposed a continuous temporal modeling(CTM)framework based on single-lead ECG.In the feature extraction phase,a convolutional neural network(CNN)was employed to capture local morphological patterns.Furthermore,a spectral channel attention(SCA)module based on the fast Fourier transform(FFT)was introduced to adaptively recal-ibrate channel weights,thereby enhancing features in key frequency bands.In the temporal modeling phase,a bidirectional long short-term memory(BiLSTM)network was adopted to aggregate continuous feature sequences.This approach explicitly modeled the physio-logical evolution surrounding apnea events,utilizing temporal context to correct classification biases resulting from a reliance solely on local features.Experimental results on the public PhysioNet Apnea-ECG dataset demonstrated that the accuracy,sensitivity and speci-ficity of the model was 90.12%,86.82%and 92.18%,respectively.Comparative experiments indicated that incorporating this CTM framework into various backbone models yielded an average performance improvement of approximately 7%over non-continuous base-lines,fully validating the effectiveness of CTM in capturing long-term dependencies.
Keywords：Electrocardiogram signalSleep apnea detectionContinuous temporal modelingAttention mechanismDeep learn-ing
Publication Date：2026-04-30
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：5( 93-97 )