Elsevier

Digital Signal Processing

Volume 35, December 2014, Pages 95-104
Digital Signal Processing

Advanced daytime polysomnographic preprocessing: A versatile approach for stream-wise estimation

https://doi.org/10.1016/j.dsp.2014.09.007Get rights and content

Highlights

  • Original approach to take advantage of singular tools of estimation.

  • Successful separation of ECG/EMG-related artefacts from 80% of EEG and EOG epochs.

  • Denoising function performs substantial noise suppression in 100% of the epochs.

  • Statistical analyses attest the outperformance in terms of SIR, SNR and RMSE.

Abstract

The enhancement of monitoring biosignals plays a crucial role to thrive successfully computer-assisted diagnosis, ergo the deployment of outstanding approaches is an ongoing field of research demand. In the present article, a computational prototype for preprocessing short daytime polysomnographic (sdPSG) recordings based on advanced estimation techniques is introduced. The postulated model is capable of performing data segmentation, baseline correction, whitening, embedding artefacts removal and noise cancellation upon multivariate sdPSG data sets. The methodological framework includes Karhunen–Loève Transformation (KLT), Blind Source Separation with Second Order Statistics (BSS-SOS) and Wavelet Packet Transform (WPT) to attain low-order, time-to-diagnosis efficiency and modular autonomy. The data collected from 10 voluntary subjects were preprocessed by the model, in order to evaluate the withdrawal of noisy and artefactual activity from electroencephalographic (EEG) and electrooculographic (EOG) channels. The performance metrics are distinguished in qualitative (visual inspection) and quantitative manner, such as: Signal-to-Interference Ratio (SIR), Root Mean Square Error (RMSE) and Signal-to-Noise Ratio (SNR). The computational model demonstrated a complete artefact rejection in 80% of the preprocessed epochs, 4 to 8 dB for residual error and 12 to 30 dB in signal-to-noise gain after denoising trial. In comparison to previous approaches, N-way ANOVA tests were conducted to attest the prowess of the system in the improvement of electrophysiological signals to forthcoming processing and classification stages.

Introduction

In neuroscience, one of the most relevant fields of research concerns sleep as a pivotal state of consciousness. The regularisation of healthy sleeping patterns ensures physical and psychological recovery to cope with demanding tasks on a daily-based routine [1]. Although, sleep process is an intricate matter consisting of converging processes of different nature; the interpretation of their bioelectrical activity offers a valuable framework to decipher the structure and bodily relationships [2]. Thus, the polysomnogram (PSG) stands out as a technique for electrophysiological monitoring of sleep-related dynamics. Based on temporal and spatial capabilities, the PSG is a unique assisting tool for staging of sleep macrostructure and microstructure [3], as well as, diagnosis of sleep disorders [4]. Each PSG recording collects a set of biological signals from different sources; such as neuronal, ocular, muscular, cardiac, respiratory, body movement, etc. [5]. Those signals wait upon posterior application of feature extraction and classification mechanisms to deploy computer-aided systems to support clinical assessments.

Certainly, inspiring approaches have been formulated to improve the biosignals resolution, either to diminish artefactual effect or to shrink down noise disruption attending to the hectic variations. Hereafter, some meritorious models are elicited. The amalgamation of coloured noise and non-stationarity processes within the neuronal sources becomes distinguishable by signal extraction methods in [6]. However, the exclusive usage of simulated data in the evaluation of the algorithms restrains the possibility to assess the impact on actual clinical data. In correspondence, [7] suggests EEGLAB, a multi-purpose toolbox for processing electroencephalographic (EEG) recordings; including artefact rejection, filtering, epoch selection, etc. Although, the robustness of the estimation core is not debatable, the complexity of the related algorithms questions its efficiency with limited number of channels, samples or computational resources. On this direction, solutions founded on second-order-statistics (SOS) demonstrate sustainable error reduction over EEGLAB utilities. In one study [8], the artefact correction is applied to datasets with event-related potentials (ERP). Therefore, the adaptation of the methods to polysomnographic data could convey new findings with dissimilar electrical responses. An alternative approach is proposed by [9] making use of wavelet-based decomposition to remove artefacts of semi-simulated data with 7.5 seconds of time duration. These two aspects motivate the enquire about the suitability of sleep-related data to be tested through such a multi-resolution analysis. Respectively, [10], [11] describe comparable methodologies to weaken electrocardiographic (ECG) artefacts and additive noise from EEG channels. The manifested exclusion of electrooculographic (EOG) and electromyographic (EMG) activity as potential artefacts generators keeps an open discussion when multivariate PSG data is deconstructed. Recently in [12], a novel tandem arrangement to cope with preprocessing and classification tasks sets out remarkable conclusions in terms of consecutive separation-denoising function blocks. In this sense, the present work deploys a model to improve integration, time-to-diagnosis and algorithmic versatility; subject to hundreds of PSG epochs.

Here, we introduce a short daytime PSG (sdPSG) preprocessing model as part of an ongoing initiative to provide computer-assisted resources related to sleep studies. Therefore, its development bears a well-structured roadmap, starting on preprocessing middleware towards supporting systems in sleep staging and diagnosis of disorders. The system tracks and suppresses external and embedding interferences that tantalise the achievement of performant scores in self-guided recognition and diagnosis. The usage of a statistical-oriented middleware instead of event-specific allocation incorporates a software-based package with purpose-specific. Besides, pseudo-online preprocessing, here denoted as streaming operation mode, pursues the avoidance of iterative and time-extended computational effort by streaming the bodily activity, as soon as it is sensed and digitally converted. In order to remove additive Gaussian noise and artefact-embedded activity, whilst neither reduction nor expansion transformation on the original data is addressed [13]. Furthermore, the computational methods aim to deal adequately with the highly complex characteristics of EEG waveforms; since non-stationarity and non-linearity assets make a major difference in contrast to its counterparts. Hereafter, the preprocessing middleware makes use of sophisticated techniques to refine EEG features over 1) subgaussian and slow time-varying EOG signals; 2) supergaussian, spiky and periodic ECG leads; and 3) high frequency EMG distributions. Once, the denoising and artefact rejection tasks are fully accomplished, sdPSG channels are sufficiently spanned to provide valuable information about sleep composition or associated abnormalities [14]. This condition is meant to be applied in subsequent processing and classification routines, expecting gainful aftermaths in comparison to current approaches [15]. According to this, the present paper attempts to reveal the intrinsic constituents of the preprocessing model under low-order and time-to-diagnosis efficiency constraints, which convey to a plausible streaming orientation with operational outcomes. In order to attest the performance degree, a complete experimental framework was prepared, regarding a testing cohort and measurable metrics from qualitative and quantitative perspectives, such as signal ratio and residual error.

The paper is organised as follows: Section 2 makes a detailed description of the active modules within the preprocessing approach; including conditions of experiments, test subjects and employed transformation/decomposition techniques. Then, Section 3 portrays the final arrangement of modules and parameters for experimental proceedings. Section 4 discusses the product of performance metrics applied to actual clinical data. Afterwards, Section 5 realises a critical analysis about the obtained results, stressing strengths and downsides of the adopted methods. Finally, Section 6 argues additional insights and remarks about future challenges and opportunities of improvement.

Section snippets

Methods

In the present manuscript, we introduce a sdPSG preprocessing computational model conceived to fulfil operational and performing requirements over simultaneous electrophysiological recordings. The performing drivers are specifically oriented to surmount the most common tributaries of distortion upon biological signals, i.e. sdPSG self-embedded artefacts and additive noise.

According to this, the preprocessing system delegates to three independent modules the whitening of recorded channels,

sdPSG preprocessing

Fastening the aforementioned modules, the deployed preprocessing system follows the model depicted in Fig. 1. Departing from the insertion of raw sdPSG signals till full-preprocessed data channels, passing through data segmentation, whitening, artefacts removal and denoising. This model resembles the arrangement in [12] that surmounts the advantages of performing source separation, followed by wavelet-based noise withdrawal.

Furthermore, the final configuration sets up the segmentation module to

Results

The introduced model coincides with a relevant goal on polysomnographic analysis, where significant outcomes elicits a successful separation of ECG/EMG-related activity from EEG/EOG signals, as it was initially presented in [24]. The current paper introduces an extended evaluation framework and statistical analysis based on ANOVA testing, which reaffirms the related findings. Nonetheless, the oversight of stipulated assumptions, like statistical independence, might compromise the appropriate

Discussion

Overall, sdPSG preprocessing model demonstrates to be a reliant preliminary approach in the preparation of electrophysiological signals for subsequent and more specialised processing routines of complete PSG. The deployment of a stream-wise algorithm for statistical source detachment proves its efficiency from an operational perspective by removing explicit and even subtle ECG-related artefacts of 960 EEG/EOG epochs. In fact, the selection of SOBIRO algorithm stems from its reliability to

Conclusions

The introduced system posits a novel approach for preprocessing of sdPSG recordings making use of renowned techniques originated in diverse application fields. Also, their relevance to biosignal processing has been widely spread out in numerous publications. Nonetheless, the adaptation of computational methods to particular conditions of biophysical phenomena is an ongoing topic of discussion. Precisely, the proposed model is an original approach to take advantage of individual tools of

Acknowledgements

The authors would like to thank all 10 human volunteers who in 2011 undertook the biofeedback study, approved by RMIT Human Research Ethics Committee. R. Chaparro-Vargas acknowledges the financial support of COLCIENCIAS-COLFUTURO Conv#529 of the Colombian Government, as main sponsor for postgraduate studies.

R. Chaparro-Vargas received the M.Sc. degree in Communications Engineering in 2010 from the Technical University of Munich, Munich, Germany. Currently, he is working towards the Ph.D. degree in the School of Electrical and Computing Engineering at RMIT University, Melbourne, Australia. His research interest resides in signal processing and computational modelling applied to biomedical applications, particularly in neurophysiology.

References (24)

  • E. Estrada et al.

    Wavelet-based EEG denoising for automatic sleep stage classification

  • R.R. Vázquez et al.

    Blind source separation, wavelet denoising and discriminant analysis for EEG artefacts and noise cancelling

    Biomed. Signal Process. Control

    (2012)
  • Cited by (0)

    R. Chaparro-Vargas received the M.Sc. degree in Communications Engineering in 2010 from the Technical University of Munich, Munich, Germany. Currently, he is working towards the Ph.D. degree in the School of Electrical and Computing Engineering at RMIT University, Melbourne, Australia. His research interest resides in signal processing and computational modelling applied to biomedical applications, particularly in neurophysiology.

    D. Cvetkovic received the M.Eng. and Ph.D. degree respectively in 2002 and 2005 from the RMIT University, Melbourne, Australia. At the present, he is a Senior Lecturer in Biomedical and Electronics Engineering at School of Electrical and Computer Engineering (RMIT University) and a member of Health Innovations Research Institute (HIRi). His research at RMIT University spans over the last 15 years in the engineering areas of biomedicine, electronics, mechatronics and design engineering education.

    View full text