ABSTRACT
Sleep apnea (SA) detection method based on traditional machine learning needs a lot of efforts in feature engineering and classifier design. We constructed a one-dimensional convolutional neural network (CNN) model, which consists in four convolution layers, four pooling layers, two full connection layers and one classification layer. The automatic feature extraction and classification were realized by the structure of the proposed CNN model. The model was verified by the whole night single-channel sleep electrocardiogram (ECG) signals of 70 subjects from the Apnea-ECG dataset. Our results showed that the accuracy of per-segment SA detection was ranged from 80.1% to 88.0%, using the input signals of single-channel ECG signal, RR interval (RRI) sequence, R peak sequence and RRI sequence + R peak sequence respectively. These results indicated that the proposed CNN model was effective and can automatically extract and classify features from the original single-channel ECG signal or its derived signal RRI and R peak sequence. When the input signals were RRI sequence + R peak sequence, the CNN model achieved the best performance. The accuracy, sensitivity and specificity of per-segment SA detection were 88.0%, 85.1% and 89.9%, respectively. And the accuracy of per-recording SA diagnosis was 100%. These findings indicated that the proposed method can effectively improve the accuracy and robustness of SA detection and outperform the methods reported in recent years. The proposed CNN model can be applied to portable screening diagnosis equipment for SA with remote server.
Subject(s)
Humans , Electrocardiography , Machine Learning , Neural Networks, Computer , Sensitivity and Specificity , Sleep Apnea Syndromes/diagnosisABSTRACT
Objectives: To plot and analyze ECG three-dimensional (3D) RR scatter plots in patients with various diseases in order to find the correlation between the graphic features and various diseases. Methods: According to ECG data provided by Physio Bank website, 3D RR scatter plots were constructed and analyzed by means of Matlab platform programming and clustering algorithm in following clinical settings: normal sinus rhythm,?arrhythmia, atrial fibrillation (AF), sudden cardiac death and chronic heart failure (CHF). Results: 3D RR scatter plots of normal sinus rhythm, arrhythmia, AF, sudden cardiac death and CHF were converged into several categories, all had the features of ultimate, stability and attraction. Significant differences on the number of clusters, clustering degrees and distribution ranges were observed among subjects with normal sinus rhythm, arrhythmia, AF, sudden cardiac death and CHF. Conclusions: 3D RR scatter plots had diversified distributions in different diseases, it can be divided into multiple categories with multi-distribution features. 3D RR scatter plots can clearly define the overlapping parts of 2D scatter diagram and are helpful to distinguish different diseases in a more accurate mode, future studies are warranted to explore the underlying clinical implications of the multiple parameters derived from the 3D RR scatter plots.
ABSTRACT
Objective: To explore the advantage of RR-Lorenz plot (RR-LP) in analyzing the patients of sinus rhythm with long RR interval. Methods: A total of 308 RR-LP patients with long RR interval were retrospectively studied. The patients were divided into 7 groups according to the type of long RR intervals. ① Sinus bradycardia with arrhythmia group, n=63,②Repeated transient sinus arrest group, n=16, ③Ⅱ° sino-atrial block group, n=14, ④Ⅱ° atrial ventricular block (Ⅱ° AVB) group, n=47, ⑤ Un-passed atrial premature beats (APB) group, n=28, ⑥ Atrial premature beats group, n=72 and ⑦ Premature ventricular beats group, n=68. We analyzed the patients of RR interval greater than 1500 ms with ambulatory electrocardiogram record. Results: ①RR-LP of sinus bradycardia with arrhythmia group showed a single distributing area with 1500 ms for the origin of transverse and longitudinal axis with B line slope at 1, tilt angle of 45°.②RR-LP of repeated transient sinus arrest,Ⅱ° sino-atrial block,Ⅱ° AVB and APB groups showed special four distributing areas with B line slope at (0.51 ± 0.01), tilt angle of (23.04 ± 0.50) °, B line slope at 0.6, tilt angle of (27°), B line slope at (0.57 ± 0.21), tilt angle of (25.69 ± 9.59)° and B line slope at (0.50 ± 0.01), tilt angle of (22.59 ± 0.54) ° respectively.③RR-LP of premature beats groups showed special four regional distributing areas, B line slope for atrial premature beats was at (0.38 ± 0.12), tilt angle of (17.06 ± 5.22) ° and B line slope for ventricular premature beats was at (0.07 ± 0.05), tilt angle of (3.02 ± 2.39) °. Conclusion: RR-LP in patients of sinus rhythm with long RR interval had speciifc morphology and distribution features, the local abnormality could be found in a plane via all RR intervals which provided a differential diagnosis for repeated occurrence of short RR interval.
ABSTRACT
Background: Altered thyroid functions are associated with variation in autonomic regulation of cardiovascular activity. Cardiac Autonomic Nervous Activity (CANA) can be assessed quantitatively by analysis of Heart Rate Variability (HRV). Objective: To observe the relationship between CANA with altered TSH and FT4. Methods: This cross sectional study was carried out in the Department of Physiology, BSMMU, Dhaka between1st July 2007 and 30th June 2008 on 60 patients with excess thyroid hormone (group B, aged 30-50 years). Based on treatment, 30 untreated newly diagnosed patients were designated as group B1 and 30 patients under treatment with antithyroid drugs for at least 2 months were included into group B2 in order to observe the effect of treatment. All these patients were selected from the Out Patient Department of Endocrinology wing of Department of Medicine, BSMMU, Dhaka. Sociodemographically matched 20 apparently healthy euthyroid persons were selected for comparison (group A). To confirm thyroid status, serum TSH and serum FT4 levels were measured by AxSym system and some of the spectral HRV parameters i.e.mean R-R interval, mean heart rate, variance, LF n.u, HF n.u and LF/HF ratio were assessed by recordings of ECG for 5 minute (short term) with a polyrite. For statistical analysis Pearson’s correlation coefficient (r) test was used. Results: With serum TSH level, the LF n.u. power and LF/HF ratio showed significant (p<0.05) positive correlations but HF n.u. power showed significant (p<0.05) negative correlation in group B1. But these three parameters showed non significant correlations with TSH in the other two groups (A, B2). Similarly with serum TSH level, variance and mean R-R interval showed negative and mean HR showed positive correlation in group B1. In group A, all these parameters were positively correlated whereas in groupB2, RR interval and variance were positively and mean HR was negatively correlated. All these correlations were statistically non significant. With serum FT4 levels, mean R-R and HF n.u. were negatively and mean heart rate, LFnu, LF/HF were positively correlated in all three groups but variance showed positive in group A and negative correlation in B1 and B2. All these correlations were statistically non significant. Conclusion: From this study it can be concluded that changes in autonomic nervous regulation are related to altered serum level of TSH and FT4 in hyperthyroids.
ABSTRACT
PURPOSE: The purposes of this study were to determine short- and long-term fractal correlation behavior of heart rates during daily activity in patients with neurocardiogenic syncope. METHODS: Twenty five patients with histories of neurocardiogenic syncope episodes were included. Their analogue 24h ambulatory Holter electrocardiograms were analyzed. The tape was digitized and the digitized electrocardiograms were partioned into sections of one hour. Then their RR intervals were measured and 20,000 points of RRI were used. To quantify the fractal correlation behavior, we employed the detrended fluctuation analysis, and short-term(n16, alpha2) fractal scaling exponents were calculated. RESULTS: When compared to control, 24-hour average values of alpha1 and all alpha1 values at quarters of each day were significantly higher in patients with syncope. On the contrary, their 24-hour average value of alpha2 and all alpha2 values at quarters of each day were lower in patients with syncope. However, statistical significances were found in 24-hour average value of alpha2 and in alpha2 value at MN-6AM. CONCLUSION: In the syncope patients with neurocardiogenic syncope, short-term fractal scaling exponents of RR interval was significantly high throughout the day. Therefore, their RR intervals were smoother in the short term scale and had a tendency to continue in the same direction of increase or decrease, which may contribute to persistent decrease in heart rate during a syncopal attack.
Subject(s)
Child , Humans , Electrocardiography , Fractals , Heart Rate , Heart , Syncope , Syncope, VasovagalABSTRACT
Purpose: The purposes of this study were to compare the magnitude and phase between the RR interval and QT interval variability in the frequency domain. METHODS: Twenty four, 12-13 year old healthy males were randomly selected. At resting state and for 5 minutes, ECGs were obtained, and they were digitized to 1000Hz. After measurement of RR interval, QT interval variability was measured using template matching strategy. After normalization of the RR and QT interval time series, power spectral and cross spectral analysis were performed. From each of the time series, low- (0.04-0.15 hertz) and high- (0.15-0.4 hertz) frequency power were measured. From the phase spectrum, the phases and time lags between the two time series at each of the two frequency range were calculated. RESULTS: The average of RR interval and QT interval was 616.0+/-71.0, 364.0+/-47.0 msec, respectively. Their normalized low- and high- frequency power was 4.4+/-7.9 NU(normalized unit), 0.1+/-0.1 NU(p<0.005), and 11.0+/-30.0 NU, 0.3+/-0.3(NU, p<0.005), respectively. The phase differences and resulting time lags between the two interval were -0.5+/-0.4 pi radian(-0.9 seconds) and -0.2+/-0.3 pi radian(-0.4 seconds) in the low- and high-frequency range, respectively. CONCLUSION: During resting state, when compared to RR interval, QT interval oscillates in significantly lower amplitude in both low- and high- frequency ranges. However, the oscillations precede those of the RR interval 0.9 seconds and 0.4 seconds, respectively.