Metin Akay

Wavelet applications in medicine

The author describes how, because of their compatibility with nonstationary random processes, wavelet transforms are particularly powerful when it comes to analyzing biomedical signals. In this paper, he presents a brief survey of the field which makes this fact clear. Applications discussed include detecting coronary artery disease, and heart sounds related to turbulent blood flow analysis.

Heart rate variability during sleep and the early development of posttraumatic stress disorder

BACKGROUND Noradrenergic function has been linked to posttraumatic stress disorder (PTSD) and might have a role in mediating sleep disturbances of the disorder. Our objective was to relate a peripheral manifestation of noradrenergic function, sympathetic nervous system activity as indexed by heart rate variability during sleep, to the development of PTSD in subjects with recent traumatic injuries. METHODS Subjects who had recall of life-threatening experiences were recruited from one of two regional trauma centers. Select subjects received a polysomnographic recording within 1 month of the trauma. Digitized electrocardiogram recordings were extracted from early and late rapid-eye-movement (REM) and preceding non-REM sleep periods. Autoregression was applied to R-R interval time series to calculate the ratios of low-frequency to high-frequency spectral densities (LF/HF ratios), which index sympathetic activation. Posttraumatic stress disorder status was determined at 2 months. RESULTS There was a significant state x group interaction: LF/HF ratios were higher during the REM sleep of the nine subjects who were positive for PTSD symptoms, compared with the 10 subjects who were PTSD negative. CONCLUSIONS Our findings are consistent with the possibility that increased noradrenergic activity during REM sleep contributes to the development of PTSD.

Fractal dynamics of body motion in patients with Parkinson's disease

In this paper, we assess the complexity (fractal measure) of body motion during walking in patients with Parkinsons disease. The body motion of 11 patients with Parkinsons disease and 10 healthy elderly subjects was recorded using a triaxial accelerometry technique. A triaxial accelerometer was attached to the lumbar region. An assessment of the complexity of body motion was made using a maximum-likelihood-estimator-based fractal analysis method. Our data suggest that the fractal measures of the body motion of patients with Parkinsons disease are higher than those of healthy elderly subjects. These results were statistically different in the X (anteroposterior), Y (lateral) and Z (vertical) directions of body motion between patients with Parkinsons disease and the healthy elderly subjects (p < 0.01 in X and Z directions and p < 0.05 in Y direction). The complexity (fractal measure) of body motion can be useful to assess and monitor the output from the motor system during walking in clinical practice.

Unconstrained monitoring of body motion during walking

Discusses using the matching pursuit algorithm to characterize time-frequency patterns of body motion in poststroke hemiplegic patients. We have been working on the quantification of body motions in healthy young and elderly subjects, patients with Parkinsons disease (PD), and poststroke hemiplegic (PSH) patients using an accelerometry technique and advanced signal processing methods. In this article, we use the matching pursuit (MP) algorithm to characterize the time-frequency patterns of the acceleration signal recorded from both healthy subjects and poststroke hemilpegic patients. The MP algorithm was chosen since it provides better time and frequency resolutions than other time-frequency analysis methods and is an algorithm that decomposes any signal into several already-known time-frequency patterns, which are called atoms. It also provides detailed information about each time-frequency pattern including its energy, time and frequency localization, and phase and scale (time duration), which can be used for the comparison and the statistical analysis.

Fractal dynamics of body motion in post-stroke hemiplegic patients during walking

In this paper, we quantify the complexity of body motion during walking in post-stroke hemiplegic patients. The body motion of patients and healthy elderly subjects was measured by using the accelerometry technique. The complexity of body motion was quantified using the maximum likelihood estimator (MLE-) based fractal analysis methods. Our results suggest that the fractal dimensions of the body motion in post-stroke hemiplegic patients at several Brunnstrom stages were significantly higher than those of healthy elderly subjects (p < 0.05). However, in the hemiplegic patients, the fractal dimensions were more related to Brunnstrom stages.

Wavelets for biomedical signal processing

In this paper, we will discuss the general idea of the wavelet representation, in its continuous and discrete versions, as well as in terms of a multiresolution approximation. In addition, the general expression for the affine class, and the relationship between the affine and Cohens classes are presented. Also, the shift-scale invariant class is defined. This class basically combines the properties of both classes. Finally a recent development, namely, the use of unitary transformations in both Cohens and the affine classes, with the consequent generation of even more specific tools for signal analysis will be discussed.

Investigating the Contamination of Electroencephalograms by Facial Muscle Electromyographic Activity Using Matching Pursuit

It has been widely recognized and previously reported that electrical fields from facial muscle electromyographic (EMG) activity can contaminate the electroencephalogram (EEG), even when closely spaced, bipolar electrode configurations are used (personal observations). We suspected that EEG signals evoked in response to pressure changes in the upper airway may include EMG contamination subsequent to muscle reflexes triggered by the stimuli. We evaluated the potential contamination of the background EEG by voluntary activation of a facial muscle by obtaining simultaneous recordings in human subjects of the EEG (from Cz-C4) and masseter muscle EMG (from a bipolar surface electrode pair) before (quiet) and after voluntary tensing (VTen). Matching pursuit analysis permitted identification of different time-frequency patterns for each signal during the quiet period because the EMG signal has mostly atoms above 30 Hz compared to the EEG signal. However, the EEG showed periods of low-frequency activity unmatched in the EMG TF pattern below 30 Hz. During the tensing, most of the atoms of both the EEG and EMG shifted to the higher frequency regions above 100 Hz, making the separation difficult. These results further suggest that the matching pursuit method may not separate the background EEG from phasic EMG signals, both of which are nonstationary in nature.

Investigating the complexity of respiratory patterns during recovery from severe hypoxia

Progressive hypoxemia in anesthetized, peripherally chemodenervated piglets results in initial depression of the phrenic neurogram (PN) culminating in phrenic silence and, eventually, gasping. These changes reverse after the 30 min reoxygenation (recovery) period. To determine if changes in the PN patterns correspond to changes in temporal patterning, we have used the approximate entropy (ApEn) method to examine the effects of maturation on the complexity of breathing patterns in chemodenervated, vagotomized and decerebrated piglets during severe hypoxia and reoxygenation. The phrenic neurogram in piglets was recorded during eupnea (normal breathing), severe hypoxia (gasping) and recovery from severe hypoxia in 31 piglets (2-35 days). Nonlinear dynamical analysis of the phrenic neurogram was performed using the ApEn method. The mean ApEn values for a recording of five consecutive breaths during eupnea, a few phrenic neurogram signals during gasping, the beginning of the recovery period, and five consecutive breaths at every 5 min interval for the 30 min recovery period were calculated. Our data suggest that gasping resulted in reduced duration of the phrenic neurogram, and the gasp-like patterns exist at the beginning of the recovery. But, the durations of phrenic neurograms during recovery were increased after 10 min postreoxygenation, but were restored 30 min post recovery. The ApEn (complexity) values of the phrenic neurogram during eupnea were higher than those of gasping and the early (the onset of) recovery from severe hypoxia (p < 0.01), but were not statistically different than 5 min post recovery regardless of the maturation stages. These results suggest that hypoxia results in a reversible reconfiguration of the central respiratory pattern generator.

Effects of hypoxia on the complexity of respiratory patterns during maturation.

During hypoxic gasping, the hypoxic neurogram has a steeper rate of rise, an augmented amplitude, and a shorter duration than is seen during eupnea. Because hypoxia reduces neural activity, we hypothesized that gasping would be characterized by low complexity (irregularity) values compared with eupnea in piglets. In this study, we define and quantify changes in the complexity of the phrenic neurogram, the output of the respiratory neural network in piglets using the approximate entropy (ApEn) method which provides a model independent measure of the complexity of the phrenic neurogram. The phrenic neurogram in vagotomized, peripherally chemodenervated, decerebrated piglets was recorded from the C5 phrenic nerve during eupnea and gasping at four postnatal ages; 3-6 days of age (n=8), 7-13 days of age (n=3), 15-21 days of age (n=4), 29-35 days of age (n=10). Nonlinear dynamical analysis of the phrenic neurogram was performed using the approximate entropy method. The mean approximate entropy values for a recording of 5 consecutive breaths during eupnea and 6-29 consecutive breaths during gasping for each piglet in each group during eupnea was calculated. Our results suggested that the mean approximate entropy values for the 3-6 days age group were 1.46+/-0.003 during eupnea and 0.85+/-0.001 during hypoxic gasping. For the 7-13 days age group, the mean approximate entropy values were 1.35+/-0.009 during eupnea and 1.00+/-0.001 during hypoxic gasping. For the 15-21 days age group, they were 1.33+/-0.005 during eupnea and 0.94+/-0.001 during hypoxic gasping. Finally, for the 29-35 days age group, they were 1.38+/-0.002 during eupnea and 0.93+/-0.001 during hypoxic gasping. The shift from eupnea to gasping caused a drastic drop in the mean values of the approximate entropy values at each of these four age groups. These differences in the complexity values of the phrenic neurogram between eupnea and gasping are statistically different at each age group (p<0.001). These findings suggest that during hypoxic gasping, regardless of degree of development, the output of the central pattern generator becomes less complex probably because hypoxia reduces the neural activity.

Statistical mapping of scalp-recorded ictal EEG records using wavelet analysis.

Summary:  Purpose: The wavelet transform (WT) is well suited for the analysis of signals whose characteristics vary rapidly over time. We devised a computerized method for objective scoring of scalp‐recorded seizures that takes advantage of the WT.