Convolutional Neural Network for Individual Identification Using Phase Space Reconstruction of Electrocardiogram.

Electrocardiogram (ECG) biometric provides an authentication to identify an individual on the basis of specific cardiac potential measured from a living body. Convolutional neural networks (CNN) outperform traditional ECG biometrics because convolutions can produce discernible features from ECG through machine learning. Phase space reconstruction (PSR), using a time delay technique, is one of the transformations from ECG to a feature map, without the need of exact R-peak alignment. However, the effects of time delay and grid partition on identification performance have not been investigated. In this study, we developed a PSR-based CNN for ECG biometric authentication and examined the aforementioned effects. Based on a population of 115 subjects selected from the PTB Diagnostic ECG Database, a higher identification accuracy was achieved when the time delay was set from 20 to 28 ms, since it produced a well phase-space expansion of P, QRS, and T waves. A higher accuracy was also achieved when a high-density grid partition was used, since it produced a fine-detail phase-space trajectory. The use of a scaled-down network for PSR over a low-density grid with 32 × 32 partitions achieved a comparable accuracy with using a large-scale network for PSR over 256 × 256 partitions, but it had the benefit of reductions in network size and training time by 10 and 5 folds, respectively.

Sleep apnea assessment using declination duration-based global metrics from unobtrusive fiber optic sensors.

OBJECTIVE: Sufficient sleep helps to restore the immune, nervous and cardiovascular systems, but is sometimes disturbed by sleep apnea (SA). The early diagnosis of sleep apnea is beneficial for the prevention of diseases. Polysomnography (PSG) recording provides comprehensive data for such assessment, but is not suitable for use at home due to discomfort during measurement and the difficulty of identification. This study proposes an unobtrusive measurement process by placing fiber optic sensors (FOSs) in a pillow (head-neck) or a bed mattress (thoracic-dorsal). APPROACH: We test two approaches: drop degrees from the baseline to validate the capability of catching respiratory drops, and linear regression models based on a new global measure, the percentage of the total duration of respiratory declination (PTDRD), to estimate the hand-scored apnea/hypopnea index (AHI). MAIN RESULTS: Based on data recorded from 63 adults, the drop degrees derived from respiratory signals exhibited statistical differences among central sleep apnea (CSA), obstructive sleep apnea (OSA) and normal breathing. The regression models based on the PTDRDs derived from head-neck FOS and thoracic-dorsal FOS also achieved good agreement with manually scored AHIs in Bland-Altman plots as well as oronasal airflow and thoracic wall movement. SIGNIFICANCE: The aforementioned performance demonstrates the capability of the FOS measurement and the efficacy of the PTDRD metrics for SA assessment.

Recognition of ventricular extrasystoles over the reconstructed phase space of electrocardiogram.

Distinguishing ventricular extrasystoles from normal heartbeats is crucial to cardiac arrhythmia analysis. This paper proposes novel morphological descriptors, the major portrait partition area (MPPA) and point distribution percentage (PDP), which are extracted from the reconstructed phase space of the QRS complex. These measures can be linked to QRS width and prolonged ventricular contraction, and offer several advantages over traditional characterization of the QRS structure: it does not require QRS boundary detection, is robust under R-peak misalignment, and including some information from nearby points. The first four principal components of MPPA variables and PDPs in the first and the third quadrants of the phase space diagram were used as inputs of neural networks. The performance of networks in distinguishing premature ventricular contraction events from normal heartbeats were evaluated under a series of 50 cross-validations based on the electrocardiogram data taken from the MIT/BIH arrhythmia database. The sensitivity and specificity obtained using the aforementioned MPPA principal components and PDPs as inputs were similar to those obtained using wavelet features and Hermite coefficients. However, the phase space information performed better in situations of noise contaminations and waveform deformations.

Classification of neuronal spikes over the reconstructed phase space.

Spike information is beneficial to correlate neuronal activity to various stimuli or determine target neural area for deep brain stimulation. Data clustering based on neuronal spike features provides a way to separate spikes generated from different neurons. Nevertheless, some spikes are aligned incorrectly due to spike deformation or noise interference, thereby reducing the accuracy of spike classification. In the present study, we proposed unsupervised spike classification over the reconstructed phase spaces of neuronal spikes in which the derived phase space portraits are less affected by alignment deviations. Principal component analysis was used to extract major principal components of the portrait features and k-means clustering was used to distribute neuronal spikes into various clusters. Finally, similar clusters were iteratively merged based upon inter-cluster portrait differences.

EEG coherence for a patient with Marchiafava-Bignami disease.

We evaluated a 39-year-old chronic alcoholic man with acute multiple cognitive impairment and callosal lesion seen in MRI, diagnosed as Marchiafava-Bignami disease (MBD), using analysis of electroencephalography (EEG) coherence. Three EEG sessions were recorded in the first week, the second and eighth month after disease onset in the MBD patient. Inter-hemispheric coherence (IhC), parasagittal coherence (PsC) and spatially averaged coherence (SAC) were computed. The results were compared with normative data from 30 age-matched healthy subjects. Mean values for IhC, PsC and SAC were significantly decreased during the acute stage of the disease (P < 0.01). The IhC values remained low (P < 0.01), however, PsC and SAC values rebounded in follow-up study. The IhC and SAC values were lowest in the frontal region, consistent with the main pathological involvement in the anterior two-thirds of the corpus callosum and early involvement of frontal cortex. In conclusion, MBD may manifest as a cerebral-disconnection state, which can be quantified using EEG-coherence analysis. EEG-coherence may serve as a useful tool for MBD diagnosis and evaluation.

Nonlinear characteristics of heart rate variability during unsupervised and steady physical activities.

The heart rate (HR) exhibits various behavior patterns in different postures and during physical activities, whereas a conventional long-term analysis of HR variability has the confounding effect whether the subject was physically active or immobilized. A specially designed ambulatory recorder that simultaneously measures the electrocardiogram and body accelerations was used to study the short-term (< or =11 beats, alpha1) fractal correlation property and the approximate entropy (ApEn) of RR interval data during sleep, sitting and standing (passive standing or mild walking) levels and immediately after rising in the morning in 15 healthy subjects. The alpha1 exponent that increased from sleep to sitting to standing implies an increased correlation of HR dynamics, which is concomitant with an increased ratio of low-frequency power to high-frequency power (LF/HF) that is usually linked with an increased sympathetic activity. A lower ApEn value during standing and after rising implies a reduced complexity of HR dynamics. Compared to the HR measures during the standing level, the LF/HF ratio showed a quick autonomic shift and alpha1 showed a rapid recruitment of fractal HR behavior after rising, whereas the ApEn value had a slower recovery of HR complexity. In conclusion, both linear and nonlinear HR behaviors during different unsupervised physical activities can be better interpreted with the aid of the recorded movement data.

Heart rate variability characterization in daily physical activities using wavelet analysis and multilayer fuzzy activity clustering.

A portable data recorder was developed to parallel measure the electrocardiogram and body accelerations. A multilayer fuzzy clustering algorithm was proposed to classify the physical activity based on body accelerations. Discrete wavelet transform was incorporated to retrieve time-varying characteristics of heart rate variability under different physical activities. Nine healthy subjects were included to investigate activity-related heart rate variability during 24 h. The results showed that the heartbeat fluctuations in high frequencies were the greatest during lying and the smallest during standing. Moreover, very-low-frequency heartbeat fluctuations during low activity level (lying) were greater than during high activity level (nonlying).