Long-term signal detection, segmentation and summarization using wavelets and fractal dimension: A bioacoustics application in gastrointestinal-motility monitoring

https://doi.org/10.1016/j.compbiomed.2006.08.013Get rights and content

Abstract

The current paper describes a wavelet-based method for long-term processing and analysis of gastrointestinal sounds (GIS). Windowing techniques are used to select sequential blocks of the prolonged multi-channel recordings and proceed to various wavelet-domain processing stages. De-noising, significant-activity detection, automated segmentation and extraction of summary curves are applied in an integrated mode, allowing for enhanced content manipulation and analysis. The proposed analysis scheme combines flexible long-term graphical representation tools, while maintaining the ability of quick browsing via visualization and auralization of the detected short-term events. This work is part of a project aiming to implement non-invasive diagnosis over gastrointestinal-motility (GIM) physiology. However, the proposed techniques might be applied to any study of long-term bioacoustics time series.

Introduction

Nowadays, wavelets become a very powerful signal processing and analysis tool, utilized in many scientific fields. Representative examples and applications include audio signal processing, de-noising and compression algorithms, modeling and prediction of digital sequences, feature extraction and spectral analysis. Utilization of wavelet transform in biomedical signals has also been reported for psycho-physiology monitoring and medical diagnosis purposes. For example, we may refer processing of EMG, EEG and EKG signals [1], [2], [3], [4], [5], as well as of various bioacoustics time series, such as lung sounds, cardiac sounds, voice signals and others [6], [7], [8], [9], [10], [11], [12]. Wavelet analysis and processing of gastrointestinal acoustic phenomena fall into the last sub-category, aiming to provide alternative diagnostic approaches for GIM-related abnormalities [11], [12], [13], [14], [15].

Wavelet transform has been introduced as an alternative time–frequency representation method suitable for non-stationary signal analysis. In contrast to its predecessor Short Time Fourier Transform (STFT) [17], wavelets feature improved time–frequency resolution, adapted to signal attributes as well as to human perception [18], [19]. In particular, effective representation of audio signals requires fine time resolution at high frequencies and fine frequency resolution at low frequencies [18]. Human hearing exhibits similar characteristics, also adapted to this logarithmic nature of frequency resolution [20]. Thus, wavelet transform is ideal for demands of that kind, featuring “constant Q analysis” properties [18], [20]. Fast implementation algorithms, reduced complexity and computational saving are some additional factors that contributed to the rapid acceleration and wide expansion of wavelets. As a result, extensions to the initial theoretic framework, establishment of new formulas concerning the involved maths, development of adaptive solutions and more delicate wavelet algorithms have become an evolving research field [21], [22], [23], [24].

The current paper focuses on signal processing and content manipulation strategies for long-term multi-channel recordings. The core of the implemented bioacoustics framework is wavelet processing, where de-noising is carried out in combination with signal detection, segmentation and summarization. The proposed Long-Term Wavelet-based Detection, Segmentation and Summarization (LT-WDSS) method is intended for gastrointestinal-motility (GIM) monitoring, via capturing and evaluation of digestive sounds (DS). However, it can be applied to any long-term signal analysis problem, for similar, sound-related applications or biomedical monitoring techniques.

Section snippets

Problem definition

Bowel sound (BS) auscultation techniques aim at the evaluation of gastrointestinal mechanical activity in order to relate it with physiological patterns and functional disorders. Gastrointestinal phonography (GIP) gathers many advantages: it is non-invasive, flexible and easy to operate, can be applied for long periods of time causing minimal discomfort to subjects [11], [12], [13], [14], [15], [16], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37]. These

Method

The proposed method aims to exploit advantages of all the previously described approaches, providing enhanced manipulation and representation of the underlying medical information. In this manner, both summarization signals and isolated events are employed to provide more thorough analysis schemes. Additionally, multi-channel recordings are utilized to overcome difficulties related with the weak nature of the propagated acoustic waves, as well as to provide sound-field topographic

Experimental setup

Various long-term, multi-channel recordings were employed during the development and the evaluation phases of the implemented methods. Experiments of GIS monitoring took place at the Gastroenterology Department of the Papageorgiou General District Hospital of Thessaloniki,4 with a protocol approved by the Hospital Ethics Committee. Healthy volunteers, meaning subjects without medical record of gastrointestinal disorders, were involved in this procedure. All the combinations of the experimental

Summary

The current paper focuses on the design and implementation of a wavelet-based set of procedures, for long-term signal de-noising, detection, segmentation and summarization. This set is suggested for multi-channel GIS monitoring, aiming to provide alternative medical solutions for the studying of GIM physiology. Nevertheless, its utilization can be easily extended to many other research fields. Comparisons with the existing methods have been executed, bringing forward many advantages.

Charalampos A. Dimoulas was born in Munich, Germany on August 14, 1974. He received the Diploma in Electrical Engineering from the Department of Electrical Engineering, Faculty of Technology, Aristotle University of Thessaloniki, Greece, in 1997. He is currently a Ph.D. degree candidate in the same Department, finishing his Ph.D. dissertation. His doctoral thesis deals with long-term, multi-channel audio-visual processing techniques for the study of human bioacoustic phenomena. His current

References (67)

  • J.Q. Zhang

    An eigenvalue residuum-based criterion for detection of the number of sinusoids in white Gaussian noise

    Digital Signal Process.

    (2003)
  • M.J. Katz

    Fractals and the analysis of waveforms

    Comput. Biol. Med.

    (1988)
  • R.S.S. Badwal

    The application of fractal dimension to temporomandibular joint sounds

    Comput. Biol. Med.

    (1993)
  • H. Shono et al.

    A new method to determine a fractal dimension of non-stationary biological time-serial data

    Comput. Biol. Med.

    (2000)
  • M.K. Koymik et al.

    Comparison of STFT and wavelet transform methods in determining epileptic seizure activity in EEG signals for real-time application

    Comput. Biol. Med.

    (2005)
  • F. Dirgenali, S. Kara, S. Okkesim, Estimation of wavelet and short-time Fourier transform sonograms of normal and...
  • R.C. Guido et al.

    Trying different wavelets on the search for voice disorders sorting

  • E. Fonseca et al.

    Discrete wavelet transform and support vector machine applied to pathological voice signals identification

  • R. Coifman et al.

    Adapted waveform ‘de-noising’ for medical signals and images

    IEEE Eng. Med. Biol.

    (1995)
  • L.J. Hadjileontiadis et al.

    A wavelet-based reduction of heart sound noise from lung sounds

    Int. J. Med. Inf.

    (1998)
  • L.J. Hadjileontiadis et al.

    Separation of discontinuous adventitious sounds from vesicular sounds using a wavelet-based filter

    IEEE Trans. Biomed. Eng.

    (1997)
  • L.J. Hadjileontiadis

    Wavelet-based enhancement of lung and bowel sounds using fractal dimension thresholding—part I: methodology

    IEEE Trans. Biomed. Eng.

    (2005)
  • L.J. Hadjileontiadis

    Wavelet-based enhancement of lung and bowel sounds using fractal dimension thresholding—part II: application results

    IEEE Trans. Biomed. Eng.

    (2005)
  • L.J. Hadjileontiadis et al.

    Enhancement of bowel sounds by wavelet-based filtering

    IEEE Trans. Biomed. Eng.

    (2000)
  • C. Dimoulas, G. Kalliris, G. Papanikolaou, A. Kalampakas, Novel wavelet domain Wiener filtering de-noising techniques:...
  • C. Dimoulas, G. Kalliris, G. Papanikolaou, A. Kalampakas, Abdominal sounds pattern classification using advanced signal...
  • R. Ranta, C. Heinrich, V. Louis-Dorr, D. Wolf, F. Guillemin, Wavelet-based bowel sounds denoising, segmentation and...
  • L. Cohen

    Time–frequency distributions—a review

    Proc. IEEE

    (1989)
  • Ol. Rioul et al.

    Wavelets and signal processing

    IEEE Signal Process. Mag.

    (1991)
  • M. Unser et al.

    Wavelet theory demistified

    IEEE Trans. Signal Process.

    (2003)
  • T. Painter et al.

    Perceptual coding of digital audio

    Proc. IEEE

    (2000)
  • Ol. Rioul et al.

    Fast algorithms for discrete and continuous wavelets transforms

    IEEE Trans. Inf. Theory

    (1992)
  • D.L. Donoho

    De-noising by soft-thresholding

    IEEE Trans. Inf. Theory

    (1995)
  • Cited by (63)

    • Study on Brain Dynamics by Non Linear Analysis of Music Induced EEG Signals

      2016, Physica A: Statistical Mechanics and its Applications
    View all citing articles on Scopus

    Charalampos A. Dimoulas was born in Munich, Germany on August 14, 1974. He received the Diploma in Electrical Engineering from the Department of Electrical Engineering, Faculty of Technology, Aristotle University of Thessaloniki, Greece, in 1997. He is currently a Ph.D. degree candidate in the same Department, finishing his Ph.D. dissertation. His doctoral thesis deals with long-term, multi-channel audio-visual processing techniques for the study of human bioacoustic phenomena. His current scientific interests include data acquisition, audio-signal processing, wavelets, artificial neural networks, pattern recognition, noise measurements, psycho-physiological monitoring, digital video systems, multimedia content management, description and retrieval within the MPEG-7 framework.

    George Kalliris was born in Nicosia, Cyprus, in 1964. In 1989 he received his M.Sc. Diploma in Electrical Engineering from the Aristotle University of Thessaloniki. His diploma thesis was carried out at the Laboratory of Electroacoustics and TV systems of the Telecommunications Division in the field of computer-aided sound signal analysis. In 1995 he received a Ph.D. from the same University. His doctoral research was on the subject of digital signal restoration of noise corrupted speech and music recordings. During and after completing his doctoral studies he worked in research and development projects and as part time teacher in the Aristotle University of Thessaloniki. His current position is Assistant Professor of Electronic Media in the Department of Journalism and Mass Communication Media at the same University. His current research interests include electronic media, radio and TV studio design, virtual studios and characters, digital audio-video processing–production–broadcasting techniques, multimedia content, restoration, management and retrieval.

    George Papanikolaou was born in Gorlitz, Poland, in 1951. In 1974 he received a B.Sc. in Telecommunications and Electronics from the Technical University of Gdansk. In 1975 he received his M.Sc. in Electroacoustics from same University. In 1978 he received a Ph.D. in Electroacoustics. After completing his studies he worked as lecturer in Aristotle University of Thessaloniki, Greece and from 1992 he is Associate Professor in the Department of Electrical Engineering and Computer Engineering, as well as in the Department of Musical Science and the Department of Journalism and Mass Media at the same University. His current field of interests include electroacoustics measurements, room acoustics, noise and vibration analysis, multi-channel 3D audio recording/reproduction systems, digital audio and video signal processing.

    Athanasios Kalampakas was born in 1954 in Litochoro, Greece. He graduated in Medicine (Ptychion Iatrikes of Aristotelian University of Thessalonica, (AUTH)) in 1979. During 1980–92 he fullfiled his army duty (compulsory) as medical officer and the rural area duty as general practitioner. Following (1992–98) he trained as gastroenterologist, initially at Larissa Gen Hospital (internal medicine) for two years and then at Ippokrateion Gen Hospital, Thessaloniki, (Gastroenterology). In 1992, he was submitted his MD Thesis in AUTH and was graduated with Honours Degree. The subject of his thesis was oesophageal motility. Through the research for his thesis, he specialized on motility disorders of the alimentary track. In 1988, he got a scholarship and joined the Royal London Hospital (RLH) Medical College in London as Research Fellow for his Ph.D. thesis. He worked there for 4 years on the subject of endoscopic hemostasis and on the subject of motility disorders working in the affiliated with the RLH, G.I Sci Res.Unit, currently Wingate Research Unit. His work on the application of microwaves for endoscopic hemostasis received a lot of worldwide attention (Gastroenterology 1993). After return to Greece, in the beginning of 1993, he joined the D’Medical Clinic of AUTH working mainly on motility disorders. On 1999, he was appointed his last job at the Papageorgiou District General Hospital as consultant of Gastroenterology and director of the Gastroenterology Unit. The subjects of his research interests comprise analysis of abdominal sounds, oesophageal motility disorders, small and large bowel motility disorders, new methods of endoscopic haemostasis, microwave tumour destruction, multimedia applications in medicine, new methods of telemedicine and distance learning, decision analysis and diagnostic algorithms in Gastroenterology and artificial intelligence in Gastroenterology.

    View full text