Chanakya Reddy Patti

Automated Sleep Spindle detection using novel EEG features and mixture models.

Research in automated Sleep Spindle detection has been highly explored in the past few years. Although a number of automated techniques were developed, many of them were based on using fixed parameters or thresholds which do not consider subject specific differences. In this research study, we introduce a novel method of sleep spindle detection using Gaussian Mixture Models with no fixed parameters or thresholds. The algorithm was tested on an online public spindles database consisting of six 30 minute sleep excerpts extracted from whole night recordings of 6 subjects. The results obtained were better when compared with other methods. We obtained an overall sensitivity of 74.9% at a 28% False Positive proportion.

Application of random forest classifier for automatic sleep spindle detection

Sleep spindle detection using supervised learning methods such as Artificial Neural Networks and Support Vector Machines had been researched in the past. Supervised learning methods such as the above are prone to overfitting problems. In this research paper, we explore the detection of sleep spindles using the Random Forest classifier which is known to over fit data to a much lower extent when compared to other supervised classifiers. The classifier was developed using data from 3 subjects and it was tested on data from 12 subjects from the MASS database. A sensitivity of 71.2% and a specificity of 96.73% was achieved using the random forest classifier.

Automatic sleep-wake detection using electrooculogram signals

In this study, we developed an automatic algorithm for sleep-wake detection based on Electrooculography (EOG) in healthy and non-healthy patients. Several features were extracted in time and frequency domains from the EOG signal. The artificial neural network (ANN) was used as a classifier. This pilot study consisted of three aims; the first aim was to utilise only the EOG signal for automatic sleep-wake stage detection. The second objective was to investigate which features were the most effective in detecting the sleep-wake phases in healthy and non-healthy individuals. The third important aim is to investigate which suitable and effective channel can be utilized for detecting the sleep-wake stages. The database was built up using 7 healthy subjects and 9 patients with mixed sleep apnoea, sleep apnoea hypopnea syndrome (SAHS), dyssomnia and periodic limb movements of sleep (PLMS). The inter-rater reliability was 91.3%. The sensitivity and specificity were 84.5% and 91.5%, respectively. Cohens kappa between visual and automatic algorithm in detection of the sleep-wake stages was 0.74.

Automated sleep spindle detection using IIR filters and a Gaussian Mixture Model

Sleep spindle detection using modern signal processing techniques such as the Short-Time Fourier Transform and Wavelet Analysis are common research methods. These methods are computationally intensive, especially when analysing data from overnight sleep recordings. The authors of this paper propose an alternative using pre-designed IIR filters and a multivariate Gaussian Mixture Model. Features extracted with IIR filters are clustered using a Gaussian Mixture Model without the use of any subject independent thresholds. The Algorithm was tested on a database consisting of overnight sleep PSG of 5 subjects and an online public spindles database consisting of six 30 minute sleep excerpts. An overall sensitivity of 57% and a specificity of 98.24% was achieved in the overnight database group and a sensitivity of 65.19% at a 16.9% False Positive proportion for the 6 sleep excerpts.

Automatic detection of sleep arousal events from polysomnographic biosignals

Manual scoring of arousals is generally conducted by sleep experts in spite of being time-consuming and subjective. Our objective of this study was to develop an algorithm for automatic detection of sleep arousals without distinguishing between the types of arousal and sleep disorder groups. The processed and analysed data multiple overnight Polysomnography (PSG) recordings, consisting of 9 human subjects (6 male, 3 female), with age range of 34-69 and different conditions (4 patients with obstructive sleep apnoeas, 4 healthy and 1 patient with periodic limb movement disorder). PSG biosignals were processed to extract necessary features. K-nearest neighbours (KNN) was used as the classifier and performance of algorithm were evaluated by Leave-One-Out Cross-Validation. The average sensitivity, specificity and accuracy of algorithm was 79%, 95.5% and 93%, respectively. These results demonstrate that our algorithm can automatically detect arousals with high accuracy. Furthermore, the algorithm is capable to be upgraded for classification of various types of arousals based upon their origin and characteristics.

Sleep onset detection with multiple EEG alpha-band features: Comparison between healthy, insomniac and schizophrenic patients

In the past several studies have evaluated the human sleep onset (wake to sleep transition) using the electroencephalographic (EEG) measurements. This paper has evaluated the detection accuracy of sleep stages for multiple features based on the EEG alpha activity, during SO in healthy, insomniac and schizophrenic patients. The features include topographic features such as Directed Transfer Function, Full frequency DTF, Welch Coherence, Minimum Variance Distortionless Response Coherence and Partial Directed Coherence. Sleep stages Wake, NREM (Non-rapid Eye Movement) stages 1 and 2 were classified using Artificial Neural Networks (ANN) classifier and evaluated using classification accuracy. The results suggest that using topographic set of features yield an agreement of 81.3 % with the whole database classification of human expert.

Linear and non-linear interdependence of EEG and HRV frequency bands in human sleep

The characterisation of functional interdependencies of the autonomic nervous system (ANS) stands an evergrowing interest to unveil electroencephalographic (EEG) and Heart Rate Variability (HRV) interactions. This paper presents a biosignal processing approach as a supportive computational resource in the estimation of sleep dynamics. The application of linear, non-linear methods and statistical tests upon 10 overnight polysomnographic (PSG) recordings, allowed the computation of wavelet coherence and phase locking values, in order to identify discerning features amongst the clinical healthy subjects. Our findings showed that neuronal oscillations θ, α and σ interact with cardiac power bands at mid-to-high rank of coherence and phase locking, particularly during NREM sleep stages.

Sleep spindle detection using multivariate Gaussian mixture models

In this research study we have developed a clustering‐based automatic sleep spindle detection method that was evaluated on two different databases. The databases consisted of 20 all‐night polysomnograph recordings. Past detection methods have been based on subject‐independent and some subject‐dependent parameters, such as fixed or variable thresholds to identify spindles. Using a multivariate Gaussian mixture model clustering technique, our algorithm was developed to use only subject‐specific parameters to detect spindles. We have obtained an overall sensitivity range (65.1–74.1%) at a (59.55–119.7%) false positive proportion.

EEG biomarker of Sleep Apnoea using discrete wavelet transform and approximate entropy

Sleep Apnoea Syndromes (SAS) is a sleep disorder which caused breathing pauses during sleep at night. There is various method of analyzing sleep EEG signals can be found in the literature. In this paper both linear; Discrete Wavelet Transform (DWT) and non-linear; Approximate Entropy (ApEn) extraction methods were performed to differentiate features of each sleep stages between apnoea and healthy person. The efficiency of both extraction methods was compared by using ANOVA. In our study, we observed the non-linear feature of ApEn improves the ability to discriminate healthy and sleep apnoea at different sleep stages.

Support Vector Machine Classification of EEG Nonlinear Features for Primary Insomnia

Primary insomnia is a term used to describe a subtype of insomnia that constitutes the disorder itself and is not a consequent to any other psychiatric or sleep disorder. Hitherto, there is no clear objective markers from Polysomnography (PSG) signal to characterize insomnia. Although linear methods like spectral analysis of EEG frequency bands have been used to detect physiological arousal in patients with insomnia, these methods may not be sufficient enough to extract valuable information and detect abnormalities in the signals. The EEG signal itself originate from a complex neuronal activity in the brain, therefore the use of nonlinear measures may show some hidden information that could better explain the activation of this hyperarousal. The aim of the present study is to classify the primary insomnia patient from the healthy based on the supervised learning machine technique of SVM and the usage of nonlinear features of EEG signal. The classification result by using SVM achieved an overall of 83% of accuracy, 85 and 80% of sensitivity and specificity respectively.