PDF(3722 KB)
Bimodal sleep staging study based on U2-Net and CBAM fusion attention
ZHAO Qian, LI Jin, FENG Feilong, QIANG Ning, HU Jing
PDF(3722 KB)
Bimodal sleep staging study based on U2-Net and CBAM fusion attention
Aiming at the difficulties present in current automatic sleep staging methods, a method for automatic sleep staging of EEG and ECG dual modal signals by combining U2-Net and CBAM fusion attention is proposed.Firstly, the EEG-ECG signals in the MIT-BIH public dataset used in this paper are preprocessed. Then, the U2-Net network with multi-scale feature extraction module is used to extract waveform features in EEG and ECG in parallel. Secondly, CBAM fusion attention is used to assign weights to all features. Finally, the Softmax activation function is used to classify sleep periods into six. The results show that when sleep staging is performed based on U2-Net and CBAM fusion attention models, the overall accuracy of hexaclassification using ECG single-modal signals is 80.2%, and the F1 score is 75.3%. The overall accuracy of six classifications using EEG single-modal signals was 85.8%, and the F1 score was 81.7%;The overall accuracy of the six classifications using EEG-ECG dual-modal signals was 90.4%, and the F1 score was 85.6%. This shows that the bimodal sleep staging model proposed in this paper is feasible and effective, and provides a new idea for automatic sleep staging.
automatic sleep staging / EEG-ECG dual-modal signal / U2-Net network / CBAM fuses attention {{custom_keyword}} /
Tab.1 Sample sizes of different sleep stages used in this study表1 本文使用的不同睡眠阶段样本数量 |
| W | S1 | S2 | S3 | S4 | REM | 合计 |
|---|---|---|---|---|---|---|
| 3 087 | 1 808 | 3 871 | 483 | 181 | 697 | 10 127 |
Tab.2 The number of samples in the training and test sets during six classifications表2 睡眠六分类时的训练集和测试集样本数量 |
| 分类数据 | 训练集(80%) | 测试集(20%) |
|---|---|---|
| W | 2 467 | 620 |
| S1 | 1 446 | 362 |
| S2 | 3 097 | 774 |
| S3 | 386 | 97 |
| S4 | 145 | 36 |
| REM | 558 | 139 |
Tab.3 Six-classifications sleep staging results of ECG, EEG, EEG-ECG表3 ECG、EEG、EEG-ECG的六分类睡眠分期结果 单位:% |
| 信号类型 | 总体结果 | 每一睡眠周期的F1值 | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| F1值 | 准确率 | W | S1 | S2 | S3 | S4 | REM | ||||
| ECG | 75.3 | 80.2 | 83.7 | 35.2 | 87.3 | 82.8 | 75.4 | 87.1 | |||
| EEG | 81.7 | 85.8 | 90.3 | 64.1 | 87.4 | 80.5 | 79.8 | 88.1 | |||
| EEG-ECG | 85.6 | 90.4 | 94.4 | 70.3 | 91.7 | 86.8 | 81.5 | 88.7 | |||
Fig.11 Confusion matrix when using single-mode ECG signals for six classifications图11 使用单模态ECG信号进行六分类时的混淆矩阵 |
Fig.12 Confusion matrix when using single-mode EEG signals for six classifications图12 使用单模态EEG信号进行六分类时的混淆矩阵 |
Tab.4 Comparison of sleep staging work with other methods表4 与其他方法的睡眠分期工作对比 |
| 研究者 | 信号类型 | 分类类别 | 分类模型 | 准确率/% |
|---|---|---|---|---|
| Rossow等[17] | EEG | Wake,NREM,REM | Hidden Markov model | 60.14 |
| Lesmana等[18] | HRV | Wake,Light sleep,Deep sleep,REM | Extreme learning machine | 71.51 |
| Utomo等[19] | ECG | Wake,Light sleep,Deep sleep,REM | Weighted extreme Machine learning | 73.09 |
| Tripathy等[20] | ECG-EEG | Wake,Light sleep, Deep sleep,REM | Deep neural network | 73.7 |
| Song等[21] | EEG | Light sleep,Deep sleep | Multivariate discriminant analysis | 79.4 |
| Zhao等[22] | ECG-EEG | Wake,Light sleep,Deep sleep,REM | Deep neural network | 80.4 |
| 本文模型 | EEG-ECG | W,S1,S2,S3,S4,REM | U2-Net+CBAM | 90.4 |
| [1] |
Although short sleep duration is related to chronic conditions, such as hypertension, diabetes and obesity, the association with stroke is less well known. Using 2006-2011 National Health Interview Surveys, we assessed the association between self-reported duration of sleep and prevalence of stroke stratifying by age and sex. Of the 154 599 participants aged 18 years or older, 29.2%, 61.8% and 9.0% reported they sleep ≤6, 7-8 and ≥9 h per day, respectively. Corresponding age-standardized prevalence of stroke were 2.78%, 1.99% and 5.21% (P < 0.001). Logistic regression models showed a higher prevalence of stroke among those who slept ≤6 or ≥9 h a day compared with those who slept 7-8 h, after adjusting for sociodemographic, behavioural and health characteristics. Further stratifying by age and sex showed that the association of duration of sleep and stroke differed among different age or sex groups. Among young adults (18-44 years), a higher prevalence of stroke was found among women with short sleep. Higher prevalence of stroke was found among middle-aged men and women reporting short or long sleep duration. Among older adults (≥65 years), higher prevalence of stroke was found only among those who slept ≥9 h. In this national sample of adults, the association between duration of sleep and stroke varied by sex and age. Although there was an association of short sleep duration with stroke, we also observed the association of long sleep duration with stroke, especially among those aged 65 years or older.Published 2014. This article is a U.S. Government work and is in the public domain in the USA.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [2] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [3] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [4] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [5] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [6] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [7] |
牛艳玲, 刘健, 朱一荻, 等. 基于心率变异性信号的睡眠呼吸暂停检测方法[J]. 陕西师范大学学报(自然科学版), 2020, 48(6):26-32.
提出了一种仅使用心率变异性信号来自动检测睡眠呼吸暂停的方法。首先，从心电信号中提取心率变异性信号，并进行异常值处理和分段；其次，利用基于方差分析（analysis of variance，ANOVA)）和最大相关-最小冗余（max-relevance and min-redundancy，mRMR）的两阶段特征选择策略获得特征向量;最后，采用五折交叉验证训练随机森林分类器。结果表明：本算法使用9个特征将每分钟“epoch”信号分类为呼吸暂停或正常，在测试集中获得的睡眠呼吸暂停识别的平均准确率为80%，Kappa系数0.61；该方法具有无创且低成本的特性，有利于便携式睡眠检测设备的硬件实现。
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [8] |
郭艳平, 刘聪, 侯凤贞, 等. 基于单通道脑电信号的睡眠自动分期研究[J]. 陕西师范大学学报(自然科学版), 2020, 48(6):18-25.
睡眠分期是睡眠评估的基础，在睡眠紊乱症的早期诊断和干预中起着重要的作用。本文利用集合经验模态分解对单通道脑电信号进行预处理，联合使用从分解得到的固有模态信号中提取的线性和非线性动力学等多元特性，构建了机器学习模型的输入特征空间，并最终训练出可行的睡眠自动分期模型。通过对111个健康受试者整夜睡眠数据的分期实验发现，使用本文提出的特征构建策略，能在多种经典的机器学习算法（反向传播神经网络、支持向量机、随机森林和极端梯度提升）中获得具有实用价值的睡眠自动分期模型。其中，基于极端梯度提升算法的模型在对睡眠状态进行4种分期和5种分期的任务中，准确率分别为81.0%和79.7%。
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [9] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [10] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [11] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [12] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [13] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [14] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [15] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [16] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [17] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [18] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [19] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [20] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [21] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| [22] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
| {{custom_ref.label}} |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
PDF(3722 KB)
Fig.1 Bimodal sleep staging network model based on U2-Net and CBAM fusion attentionFig.2 Comparison of EEG signal before and after denoisingFig.3 Comparison of ECG signal before and after denoising
Tab.1 Sample sizes of different sleep stages used in this studyTab.2 The number of samples in the training and test sets during six classificationsFig.4 Overall structure diagram of U2-Net networkFig.5 Structural diagram of MSEFig.6 Multimodal feature fusion operation diagramFig.7 Structural diagram of CBAMFig.8 The classifier diagramFig.9 Loss function curve and accuracy curve of the training setFig.10 Loss function curve and accuracy curve of the verification setTab.3 Six-classifications sleep staging results of ECG, EEG, EEG-ECGFig.11 Confusion matrix when using single-mode ECG signals for six classificationsFig.12 Confusion matrix when using single-mode EEG signals for six classificationsFig.13 Confusion matrix when using dual-modal EEG-ECG signals for six classificationsFig.14 Comparison of module ablation experimental resultsFig.15 Comparison results of loss functionsTab.4 Comparison of sleep staging work with other methods/
| 〈 |
|
〉 |
