Electrocardiogram analysis using a combination of statistical, geometric, and nonlinear heart rate variability features

Sudden cardiac death (SCD) prediction based on nonlinear heart rate variability features and SCD index

Detection of congestive heart failure from short-term heart rate variability segments using hybrid feature selection approach

Heart rate variability (HRV) analysis from 24-hour ambulatory ECG has been widely used in risk stratification of patients after myocardial infarction (MI). The accuracy of HRV assessment is known to potentially vary when different commercial systems are used. However, the consistency of HRV measurements has never been fully investigated. Twenty-six post-MI patients (mean age 59 +/- 8 years, 22 men) were studied, of whom 13 succumbed to sudden cardiac death (SCD) within 1 year and 13 remained alive for at least 3 years (MI survivors). Each patient had a 24-hour Holter ECG recorded before hospital discharge. HRV analysis was performed four times from the same recordings using three different Holter tape analysis systems (Marquette, Reynolds, and CardioData) by four independent operators (CardioData system was used twice, once in the United Kingdom and once in the United States). Mean normal-to-normal RR intervals (mNN) and 3 HRV parameters (SDNN, RMSSD, and HRV triangular index [HRVi]) were derived from each recording. The consistency of mNN and HRV measurements was evaluated by coefficient of variance (CV) and by the Bland-Altman method. The results demonstrated that (1) all indices measured by different systems were statistically similar (P = NS) except the measurement of RMSSD (P = 0.01), (2) the measurements of mNN were highly reproducible with a maximum mean difference of 1.8 +/- 13.8 ms and maximum limits of agreement from -14.6 to +15.6 ms. The maximum mean differences were--1.8 +/- 1.4 unit and 4.4 +/- 9.6 ms for HRVi and SDNN, respectively, and RMSSD was less reproducible with a maximum mean difference of--11.1 +/- 11.5 ms, and limits of agreement from -16.2 to +9.6 ms; and (3) the consistency of mNN (CV 0.9% +/- 0.9%) was significantly higher than that of HRVi, SDNN, and RMSSD (P < 0.0001). The consistency of HRVi was similar to that of SDNN (4.8% +/- 2.1% vs 5.7% +/- 4.8%, P = 0.4), and the consistency of RMSSD (26.6% +/- 13.3%) was significantly lower than that of the other measurements (P < 0.00001). In conclusion, the measurements of mNN by different analytical systems are the most consistent among the parameters studied. The global 24-hour measurements of HRV (SDNN and HRVi) are highly reproducible, whereas the measurement of short-term HRV components (RMSSD) is significantly less reproducible.

The paper is devoted to the application of machine learning methods for epileptic seizures prediction based on features of electroencephalogram signal (EEG) and heart rate variability (HRV) parameters derived from simultaneously registered electrocardiogram signal (ECG). To study the possibilities of machine learning to predict epileptic seizures based on EEG and HRV analysis, open databases were used which contain annotations indicating the presence or absence of epileptic seizures at each minute of time. The goals of this work are prediction of epileptic seizures in terms of ictal, pre-ictal and interictal states recognition using features of brain electrical activity and HRV, and the choice of classification methods that provide the highest accuracy for this task. HRV parameters in time and frequency domains, as well as parameters of EEG signals based on energy of EEG rhythms, Hearst index, Higuchi fractal dimension and different estimations of entropy for EEG signals were used to achieve these goals. Using different sets of features, the accuracy of classifiers based on k-nearest neighbors method, decision trees, random forest classifier, multi-layer perceptron, support vector machines, and extreme gradient boosting classifier was determined.

In present study, we proposed not only a novel methodology useful in developing the various features of heart rate variability (HRV), but also a suitable prediction model to enhance the reliability of medical examinations and treatments for coronary artery disease. In order to develop the various features of HRV, we analyzed HRV for three recumbent postures. The interaction effects between the recumbent postures and groups of normal people and heart patients were observed based on linear and nonlinear features of HRV. Forty-three control subjects and 64 patients with coronary artery disease participated in this study. In order to extract various features, we tested five classification methods and evaluated performance of classifiers. As a result, SVM and CMAR (gave about 72---88% goodness of accuracy) outperformed the other classifiers.

Introduction Parkinsonian syndromes of different nature feature autonomic nervous system dysfunction (ANSd) which may herald or follow motor symptoms (MS) in Parkinson disease (PD), but anticipate by years and progress more rapidly in multiple system atrophy (MSA). Abnormal Heart Rate Variability (HRV) has been reported in both PD and MSA. Purpose This study aimed to evaluating if HRV analysis (HRVa) could contribute to differentiate PD from MSA and to guide early and more appropriate treatment and prognostic assessment. Methods 66 consecutive parkinsonian patients (pts) (48 males, 18 females) (mean age 63.8±10 y) were enrolled between 2010 and 2020. Initial clinical diagnosis was MSA (19), PD (20), PD with autonomic impairment (PD-AI) (10), Lewy body disease (2), supranuclear palsy (3), undefined Parkinsonism (UP) (12). The severity of MS was quantified with UPDRS and Hoehn/Yahr scales. HRV was analyzed with linear (TD and FD) and non-linear (NL) methods (Kubios Premium 3.4.1), during daily activity and NREM sleep (2-minutes intervals) and during REM sleep (1-minute Interval). 44/66 underwent also ASN functional evaluation with the Ewing protocol. Data of 52 age-matched healthy subjects (HS) were used for comparison. For statistical analysis, SPSS software (version 21) was used. Discriminant Analysis (DA) was applied to identify the more relevant parameters differentiating PD from MSA and classifying “undefined” cases. Prolonged follow-up (f-up) provided clinical diagnostic certainty, as goldstandard for validation of HRVa classification accuracy. Results UPDRS and Hoehn/Yahr scores were higher in MSA than in PD (p&amp;lt;0.0001). Along 10-years of f-up, 16/66 pts died (56,2% MSA), 5/66 had new diagnostic definition, and 15/66 were lost to f-up. Most of HRV parameters of 34 pts with certain diagnosis (17 MSA and 17 PD) whose f-up was completed, were significantly different from those of HS. VLFpower and recurrence plot parameters calculated during active wakefulness were different (p&amp;lt;0.05) between MSA and PD, and correctly classified them with 88,2% predictive accuracy (PA) at DA, according to the formula: F1 = (vlfpw%ar* − 0,087) + (rprec* − 0,319) + (rpshen*10.81) + (−20,53) Comparative classification accuracy of HRVa, Ewing score, UPDRS and Hoehn/Yahr scales is shown in Table 1. Attempting prognostic risk assessment by applying F1 to HRV data of pts with PD-AI and of pts with UP, those classified as MSA according to F1 showed the worst clinical outcome during the f-up. Conclusions This study confirms that a higher degree of prevailing sympathetic ASNd in MSA compared to PD can be identified by a combination of Linear and NL HRV parameters, and correctly differentiate with high (88,2%) predictive accuracy all pts with certain diagnosis verified during prolonged f-up. Moreover, HRVa might be useful for earlier prognostic stratification of pts with parkinsonian symptoms of uncertain nature and to improve clinical and pharmacological management. Funding Acknowledgement Type of funding sources: None.

Physiological stress is human body's response to a challenge by multiple systems in the body, in which ANS (Autonomic Nervous System) plays key roles. This paper proposes a stress level evaluation method based on time-frequency domain analysis of HRV (Heart Rate Variability) features combining statistical optimization. The most reliable and efficient indicator to ANS is HRV features obtained from time and frequency domain analysis of R-R intervals recorded during the modified Stroop test. A BP Neural network is utilized to train HRV features selected by t-test and one-way Anova test and classify new samples into four stress level categories i.e. relaxed state, low stress, medium stress and high stress. Weighing factors in classification and number of neurons in hidden layer are considered for optimized network. The proposed method has been validated by experimental results.

The current study deals with the investigation of the effect of long-term endurance training on the autonomic nervous system of healthy adults. ECG was recorded for 5 min under resting condition in a sitting position using an ECG acquisition device for 25 swimmers and 25 age-matched sedentary controls. Heart Rate Variability (HRV) parameters of the volunteers were used for statistical analysis and classification using binary classification trees and artificial neural networks. The LF/HF ratio for swimmers and sedentary controls was found to be 0.89 ± 0.32 and 0.94 ± 0.46, respectively. This may be attributed to the vagal dominance due to endurance training in the swimmers. Statistical ECG signal processing and db06 wavelet based processing were performed to understand the effect of swimming on the cardiac health. The signal classification results indicated that both the HRV and the processed ECG signal features may be used for the classification of the swimmers and the sedentary controls using CART, Boosted tree, Random Forest and neural network algorithms.

Depressed heart rate variability (HRV) after a myocardial infarction is associated with increased mortality. This is thought to be due to reduced parasympathetic activity and heightened sympathetic activity. Aging is associated with depressed HRV, but little is known of the affect of aging on parasympathetic activity. This study examined 56 healthy subjects (age range 40-102 years; 39 women). None had a history of heart disease or were on medication that would affect cardiac function. All had normal resting ECGs, normal heart size on chest X ray, and normal electrolytes. In all subjects, 24-hour Holter recordings were performed and used to measure HRV. In particular, the study examined the affect of age on HRV triangular index, which gives an estimate of overall HRV, and on RMSSD (square root of the mean squared differences of successive normal-to-normal RR intervals), which gives an estimate of short-term components of HRV and is thought to reflect the overall extent of vagal modulations of heart rates. Both these parameters were compared in patients younger and older than 70 years. Each recording lasted at least 17 hours; the majority of recordings were longer than 20 hours. There was a significant decrease in HRV triangular index with age (r = -0.4, P < 0.05) and no significant change in RMSSD with age(r = -0.08, P = NS). There was a significant difference in HRV index in those > 70 years compared with those < 70 years (38.0 +/- 9.3 vs 31.0 +/- 11, respectively, P < 0.02). There was no significant difference in RMSSD between the two age groups (26.7 +/- 8.2 ms vs 28.4 +/- 11.3 ms, respectively, P = NS). Thus, the study concludes that aging reduces the global measure of HRV and may reflect reduced responsiveness of autonomic activity to external environmental stimuli with age. However, the time-domain short-term components of HRV are not affected by age and, therefore, the fast and presumably vagal modulations of heart rate appear to be maintained.

The measurement of ECG-derived heart rate variability (HRV) provides early diagnosis and prevention of cardiac events such as coronary artery disease (CAD) and myocardial infarction (MI). The disease severity causes ECG shape deformation, leading to difficulty in R-wave detection required for heart rate variability (HRV) analysis. Therefore, the study aimed to optimize R-wave threshold filters at 50%, 60%, 70%, and 80% within the framework of HRV analysis for better diagnosis of MI and CAD. Lead-II ECG was recorded from CAD (n=15), MI (n=15), and Control (n=10) group participants using an MP45 amplifier (Biopac Systems Inc., USA). The ECG morphologic features were extracted to analyze the change in R-wave amplitude. The R-R interval time series was extracted at different threshold filters (50%, 60%, 70%, and 80%) and used for heart rate variability (HRV) analysis. The HRV features extracted using Kubios V2.0 in time, frequency, and nonlinear domains at different threshold filters were compared. The obtained findings revealed that R-wave amplitude was suppressed under the CAD and MI groups in comparison to the control subjects. The observed results also suggested 70% threshold filters for optimum detection of R-wave under CAD and MI. Further, the time and frequency domains presented the best result in comparison to the nonlinear HRV domain. The present study suggested that HRV analysis may be made more predictive for various cardiac diseases by optimizing R-wave threshold filters.

An intrusion detection system (IDS) is an important protection instrument for detecting complex network attacks. Various machine learning (ML) or deep learning (DL) algorithms have been proposed for implementing anomaly-based IDS (AIDS). Our review of the AIDS literature identifies some issues in related work, including the randomness of the selected algorithms, parameters, and testing criteria, the application of old datasets, or shallow analyses and validation of the results. This paper comprehensively reviews previous studies on AIDS by using a set of criteria with different datasets and types of attacks to set benchmarking outcomes that can reveal the suitable AIDS algorithms, parameters, and testing criteria. Specifically, this paper applies 10 popular supervised and unsupervised ML algorithms for identifying effective and efficient ML-AIDS of networks and computers. These supervised ML algorithms include the artificial neural network (ANN), decision tree (DT), k-nearest neighbor (k-NN), naive Bayes (NB), random forest (RF), support vector machine (SVM), and convolutional neural network (CNN) algorithms, whereas the unsupervised ML algorithms include the expectation-maximization (EM), k-means, and self-organizing maps (SOM) algorithms. Several models of these algorithms are introduced, and the turning and training parameters of each algorithm are examined to achieve an optimal classifier evaluation. Unlike previous studies, this study evaluates the performance of AIDS by measuring the true positive and negative rates, accuracy, precision, recall, and F-Score of 31 ML-AIDS models. The training and testing time for ML-AIDS models are also considered in measuring their performance efficiency given that time complexity is an important factor in AIDSs. The ML-AIDS models are tested by using a recent and highly unbalanced multiclass CICIDS2017 dataset that involves real-world network attacks. In general, the k-NN-AIDS, DT-AIDS, and NB-AIDS models obtain the best results and show a greater capability in detecting web attacks compared with other models that demonstrate irregular and inferior results.

Diagnosis of major depressive disorder by combining multimodal information from heart rate dynamics and serum proteomics using machine-learning algorithm

Heart rate variability (HRV) analysis, which is providing information about the autonomic nervous system with noninvasive method, performed with using digitized electrocardiogram records. In this study, whether HRV analysis can be done from heart sounds (which are recorded with electronic stethoscopes) has been investigated. For this purpose, an algorithm has been developed for detecting the consecutive heart time period from simultaneously recorded heart sounds and ECG. Some standard HRV parameters are calculated and compared from the HRV derived from ECG and heart sound signals. The experimental results show that, both HRV parameters derived from the heart sounds while the resting position and the ECG are very close. In conclusion, they can be used in HRV analysis.

Heart rate variability (HRV) analysis has increasingly become a promising marker for the assessment of the autonomic nervous system. The easy derivation of the HRV has determined its popularity, being successfully used in many research and clinical studies. However, the conventional HRV analysis is performed on 5 min ECG recordings which in e-health monitoring might be unsuitable, due to real-time requirements. Thus, the aim of this study is to evaluate the association between the raw ECG heartbeats and the HRV features to further reduce the number of heart beats required for the HRV estimation enabling real time monitoring. We propose a deep learning based system, specifically a recurrent neural network for the inference of two time domain HRV features: AVNN (the average of all the NN intervals) and IHR (instantaneous heart rate). The obtained results suggest that both AVNN and IHR can be accurately inferred from a shorter ECG interval of about 1 min, with a mean error of <5% of the computed HRV features.

Oxidative stress has been implicated in over 100 disorders in recent years; however, the situation in restless legs syndrome (RLS) has not been studied yet. Fifty patients with RLS not medicated for RLS and 50 sex- and age-matched, healthy controls and controls with no pathology except mild iron deficiency or iron deficiency anaemia were enrolled. Patients with secondary RLS other than iron deficiency were excluded. Total oxidant status (TOS), total antioxidant status (TAS), oxidative stress index (OSI), arylesterase (ARE), paraoxonase (PON), stimulated paraoxonase (stim-PON), lipid hydroperoxides (LOOHs), acetyl cholinesterase (AChE) and butyryl cholinesterase (BuChE) were measured. Heart rate variability (HRV) analysis was performed. TOS, ARE and AChE were increased (P = 0·018, P < 0·001 and P < 0·001, respectively), whereas LOOHs were decreased (P < 0·001) in RLS group. TAS, OSI, PON and stim-PON were comparable. Erythrocyte sedimentation rate (ESR) and mean platelet volume (MPV) were increased (P = 0·021 and P = 0·037, respectively) in RLS group. HRV triangular index (HRVi) was lower (P = 0·012) in RLS group. Other HRV parameters were similar. Increased AChE and decreased LOOHs, which were influenced by increased PON1, were considered as indicators of efforts towards the protection of dopaminergic activity in central nervous system in RLS group. Increased ESR, MPV and low HRVi indicate elevated sympathetic activity in RLS group. Elevated sympathetic activity might be beneficial in relieving RLS symptoms, also causing increases in TOS. The evidence we found regarding oxidative stress and autonomic nervous system might be seminal in RLS treatment.