Sleep apnea (SA) is a harmful respiratory disorder that has caused widespread concern around the world. Considering that electrocardiogram (ECG)-based SA diagnostic methods were effective and human-friendly, many machine learning or deep learning methods based on ECG have been proposed by prior works. However, these methods are based on feature engineering or supervised and semisupervised learning techniques, and the feature sets are always incomplete, subjective, and highly dependent on labeled data. In addition, some related studies ignored the data imbalance problem which leads to poor performance of classifier on minority classes. In this study, an SA detection model based on frequential stacked sparse auto-encoder (FSSAE) and time-dependent cost-sensitive (TDCS) classification model was proposed. The FSSAE extracts feature set automatically with unsupervised learning technique, and the TDCS classification model is proposed by combining the hidden Markov model (HMM) and the MetaCost algorithm to improve the performance of the classifier by considering temporal dependence and the imbalance problem. In the test set, the result of per-segment classification achieved 85.1%, 86.2%, and 84.4% for accuracy, sensitivity, and specificity, respectively, proving that our method is helpful for SA detection.

In this paper, novel heart rate variability (HRV) indices were extracted for the autonomic nervous system (ANS) activity assessment in congestive heart failure (CHF). It has been reported that CHF is a chronic cardiovascular syndrome along with ANS dysfunction, and HRV is a useful tool for ANS assessment. The multi-frequency components Entropy (MFC-En), which is obtained by the Hilbert-Huang transform and the entropy algorithm, was proposed as novel HRV indices for analyzing ANS with CHF. This paper included 24-h HRV signals of 98 subjects collected with Holter (54 healthy, 12 low-risk CHF, and 32 high-risk CHF subjects). The MFC-En indices successfully showed a statistical significance between the control and CHF groups (p < 0.001). The CHF classification accuracy of the MFC-En was 86.7%, while the ratio of the low- and high-frequency power was only 79.6%. Moreover, statistical significances were found among the control, low-risk CHF, and high-risk CHF groups (p < 0.01). Therefore, the MFC-En is a useful tool for CHF assessment that revealed the ANS of CHF patient is more activated by measuring the complexity of the rhythms changes of the ANS throughout the day.

Obstructive sleep apnea (OSA) is a common sleep disorder that is often associated with reduced heart rate variability (HRV), thus reflecting modulation of the autonomic system. Sliding trend fuzzy approximate entropy (SlTr-fApEn), which is based on the empirical mode decomposition (EMD) method, has been proposed as a novel index for analyzing HRV with OSA. This study included 60 electrocardiogram recordings from the PhysioNet database (40 OSA recordings and 20 healthy recordings) with apnea or no apnea in 5-minute segments. HRV indices obtained by sliding trend analysis were compared to those obtained by time-frequency domain analysis. Among all indices, the ratio of low-frequency power and high-frequency power (LF/HF) and sliding trend indices could significantly differentiate OSA recordings from normal recordings (p < 0.05). The OSA screening accuracy of SlTr-fApEn (85%) was higher than that of LF/HF (80%). Disease state analysis showed significant differences in SlTr-fApEn among the control group, normal OSA group, and apnea OSA group (p < 0.05). Therefore, SlTr-fApEn can reflect the complexity of autonomic changes during a short time period.

Congestive heart failure (CHF) is a cardiovascular disease associated with autonomic dysfunction, where sympathovagal imbalance was reported in many studies using heart rate variability (HRV). To learn more about the dynamic interaction in the autonomic nervous system (ANS), we explored the directed interaction between the sympathetic nervous system (SNS) and the parasympathetic nervous system (PNS) with the help of transfer entropy (TE). This article included 24-h RR interval signals of 54 healthy subjects (31 males and 23 females, 61.38 ± 11.63 years old) and 44 CHF subjects (8 males and 2 females, 19 subjects’ gender were unknown, 55.51 ± 11.44 years old, 4 in class I, 8 in class II and 32 in class III~IV, according to the New York Heart Association Function Classification), obtained from the PhysioNet database and then segmented into 5-min non-overlapping epochs using cubic spline interpolation. For each segment in the normal group and CHF group, frequency-domain features included low-frequency (LF) power, high-frequency (HF) power and LF/HF ratio were extracted as classical estimators of autonomic activity. In the nonlinear domain, TE between LF and HF were calculated to quantify the information exchanging between SNS and PNS. Compared with the normal group, an extreme decrease in LF/HF ratio (p = 0.000) and extreme increases in both TE(LF→HF) (p = 0.000) and TE(HF→LF) (p = 0.000) in the CHF group were observed. Moreover, both in normal and CHF groups, TE(LF→HF) was a lot greater than TE(HF→LF) (p = 0.000), revealing that TE was able to distinguish the difference in the amount of directed information transfer among ANS. Extracted features were further applied in discriminating CHF using IBM SPSS Statistics discriminant analysis. The combination of the LF/HF ratio, TE(LF→HF) and TE(HF→LF) reached the highest screening accuracy (83.7%). Our results suggested that TE could serve as a complement to traditional index LF/HF in CHF screening.

Obstructive sleep apnea (OSA) is the most common sleep-related breathing disorder that potentially threatened people's cardiovascular system. As an alternative to polysomnography for OSA detection, ECG-based methods have been developed for several years. However, previous work is focused on feature engineering, which is highly dependent on the prior knowledge of human experts and maybe subjective. Moreover, feature engineering also highlights the prominent shortcoming of current learning algorithms that the features are unable to extracted and organized from the data. In this study, we proposed a method to detect OSA based on deep neural network and Hidden Markov model (HMM) using single-lead ECG signal. The method utilized sparse auto-encoder to learn features, which belongs to unsupervised learning that only requires unlabeled ECG signals. Two types classifiers (SVM and ANN) are used to classify the features extracted from the sparse auto-encoder. Considering the temporal dependency, HMM was adopted to improve the classification accuracy. Finally, a decision fusion method is adopted to improve the classification performance. About 85% classification accuracy is achieved in the per-segment OSA detection, and the sensitivity is up to 88.9%. Based on the results of per-segment OSA detection, we perfectly separate the OSA recording from normal with accuracy of 100%. Experimental results demonstrated that our proposed method is reliable for OSA detection.

Obstructive sleep apnea (OSA) is a common sleep disorder that often associates with reduced heart rate variability (HRV) indicating autonomic dysfunction. HRV is mainly composed of high frequency components attributed to parasympathetic activity and low frequency components attributed to sympathetic activity. Although, time domain and frequency domain features of HRV have been used to sleep studies, the complex interaction between nonlinear independent frequency components with OSA is less known. This study included 30 electrocardiogram recordings (20 OSA patient recording and 10 healthy subjects) with apnea or normal label in 1-min segment. All segments were divided into three groups: N-N group (normal segments of normal subjects), P-N group (normal segments of OSA subjects) and P-OSA group (apnea segments of OSA subjects). Frequency domain indices and interaction indices were extracted from segmented RR intervals. Frequency domain indices included nuLF, nuHF, and LF/HF ratio; interaction indices included mutual information (MI) and transfer entropy (TE (H→L) and TE (L→H)). Our results demonstrated that LF/HF ratio was significant higher in P-OSA group than N-N group and P-N group. MI was significantly larger in P-OSA group than P-N group. TE (H→L) and TE (L→H) showed a significant decrease in P-OSA group, compared to P-N group and N-N group. TE (H→L) were significantly negative correlation with LF/HF ratio in P-N group (r = −0.789, p = 0.000) and P-OSA group (r = −0.661, p = 0.002). Our results indicated that MI and TE is powerful tools to evaluate sympathovagal modulation in OSA. Moreover, sympathovagal modulation is more imbalance in OSA patients while suffering from apnea event compared to free event.

Sleep is a complex phenomenon involved important modifications throughout life and changes of autonomic cardiovascular control.In present study, we aimed to analyze the influence of sleep stages on cardiovascular activity.Firstly, heart rate variability (HRV), pulse rate variability (PRV) and pulse transit time (PTT) were derived from ECG and PPG signals in 10 heathy volunteers during overnight polysomnography.Then, we extracted linear and nonlinear features of HRV, PRV and PTT.Linear features were consisted of mean intervals, the standard deviation of all normal interval (SDNN) and coefficient of variation (CV), and nonlinear features including approximate entropy (ApEn), sample entropy (SampEn), and fuzzy entropy (FuzzyEn).The results indicated that the SDNN and CV were significantly larger in rapid eye movement (REM) sleep compared to slow wave sleep (SWS) (p<0.05) in HRV, PRV and PTT.ApEn was significantly larger in REM compared to SWS (p<0.05) in HRV, PRV.In conclusion, the cardiovascular fluctuations and complexes were decreased when the sleep state progressed from REM to SWS, and the results of linear and nonlinear feathers extracted by PRV were consistent with HRV.