Department of Biomedical Engineering
Biomedical Signal Processing Lab.
Yaniv Zigel, Ph.D.
Head, BSP Lab.
Department of Biomedical Eng.
Biomedical signal processing has become a necessary tool for extracting clinically significant information hidden in physiologic signals. This information can be the physiological state of a patient or even their psychological state in some cases.
The vision of our research is to advance the development of computer based diagnostic and monitoring systems which offer fully automated analysis. These diagnostic systems can help the physician in making well founded decisions and reduce the subjectivity of manual measurements and decision making process. Automated monitoring systems can save life in real-time situations in cases such as apneas and cardiac failures.
The research activity at the is concentrate on bioelectrical signals such as the electrocardiograms (ECG) and electroencephalogram (EEG), and physiological acoustic signals such as snore sounds, cough sounds, phonocardiogram (PCG) and voice/speech signals. In these research activities, new methods of signal processing and pattern recognition that extract useful information from the physiological signals are being developed.
The research activities in the last four years can be divided into four main research areas:
1. Sleep research: ¨ Recognition and classification of snoring sounds for obstructive sleep apnea (OSA) diagnosis.
¨ Identifying sleep apnea patients from their speech signals,
¨ Analyzing EEG signals from OSA patients.
2. Heart research: ¨ Identifying atrial electrical activity in arrhythmias using ECG signals.
medical home: ¨ Fall
detection system of elderly using floor vibrations and
¨ Development of a cough detection monitoring device,
research: ¨ Age estimation from speech signals,
¨ Robust speaker recognition to reverberant speech,
¨ Room volume estimation from speech signals.
Sleep Research – Analysis and Diagnosis of Obstructive Sleep Apnea
Obstructive sleep apnea (OSA) is a highly prevalent disease in which upper airways collapse during sleep causing a complete or partial breath cessation. One of the foremost consequences of these breathing pauses is a major fragmentation of sleep, causing a marked deterioration of sleep quality. The main complaints of OSA patients are snoring and excessive daytime sleepiness, which result in a substantial decrease in life quality. Untreated OSA is a major risk factor for both chronic and acute conditions such as cardiovascular events and even sudden death.
The gold standard for OSA assessment is the polysomnography (PSG) in which patients are connected to numerous sensors. The cost of PSG and the complexity of the study motivate further development of alternative approaches for screening subjects at risk for OSA. OSA severity is defined by the number of obstructive apnea and hypopnea events per hour of sleep (apnea hypopnea index – AHI).
Snoring, caused by the vibration of soft tissues in the upper airways (UA) is one of the common symptoms of OSA. Snoring has long been viewed as a potential indicator for monitoring OSA, and it is especially advantageous when using a non-contact microphone as the screening tool. However, as of yet, it has not been properly exploited in diagnosis. Similar to the vocal tract in speech production, the upper airways acts as a variable acoustic filter in the generation of snoring sounds. Patients with OSA commonly have narrower and more collapsible UA then non-OSA subjects. Therefore, it is expected that the acoustic characteristics of snores from OSA and non-OSA subjects will be different. In this research we hypothesize that the snore signal carries important information about the OSA condition.
Our group has suggested a system that analyses the entire nocturnal audio signal and combines several developed acoustic features with a classifier in order to classify subjects as either OSA snorers or benign (non-OSA) snorers [Ben-Israel, Tarasiuk, and Zigel, 2010]. OSA patients were recruited prospectively and consecutively at the Sleep-Wake Disorders Unit, Soroka Medical Center (Head of the unit: Prof. Ariel Tarasiuk). The results that were achieved using 60 subjects with different OSA severities were very good; the obtained sensitivity (specificity) was 96.5% (90.6%) for resubstitution and 87.5% (82.1%) for cross validation (CV), where sensitivity and specificity stands for correct detection of OSA and non-OSA subjects, respectively. A patent for this system has been submitted [Zigel, Tarasiuk, and Ben-Israel, 2010]. Recently, our research was granted by the Israeli chief scientist, Ministry of Industry, Trade & Labor [KAMIN, 2011-2013].
Since there are studies that have confirmed that OSA is associated with anatomical and functional abnormalities of the upper airway, and since speech signal properties are also influenced by the anatomical structure of the upper airway, we investigated the correlation between OSA and speech signals [Zigel, Tarasiuk, and Goldshtein, 2008]. Using speech signal processing techniques, we explored acoustic speech features of 93 subjects who were recorded using a text-dependent speech protocol and a digital audio recorder immediately prior to polysomnography study. Following analysis of the study, subjects were divided into OSA (n=67) and non-OSA (n=26) groups. A Gaussian mixture model (GMM)-based system was developed to model and distinguish between the groups; discriminative features such as vocal tract length (VTL) and linear prediction coefficients (LPC) were selected using feature selection technique. Specificity and sensitivity of 83% and 79% were achieved for the male OSA and 86% and 84% for the female OSA patients, respectively [Goldshtein, Tarasiuk, and Zigel, 2011].
Another study was performed by our group in this area of OSA. We analyzed EEG signals in order to estimate slow-wave activity (SWA), a marker of sleep homeostasis, in children with OSA before and after adenotonsillectomy (TA) compared with untreated OSA children (comparison group) [Ben-Israel, Zigel, et al., 2010]. The conclusions were:
homeostasis is considerably impaired in pre-pubescent children with OSA, and
2) adenotonsillectomy restores more physiological sleep homeostasis in children with OSA.
Heart research - Identifying atrial electrical activity in arrhythmias using ECG signals
In many arrhythmias, extraction of atrial activity or cancellation of ventricular activity can be a great help for the physician who tries to identify the type of arrhythmia from ECG signals. In many cases, the P wave is hidden in a QRS complex or in a T wave, and cannot be observed. Our group (in cooperation with Prof. Amos Katz) developed an algorithm for ECG signal separation and p-wave detection that can make the arrhythmia-type decision more convenient [Weissman, Katz, and Zigel, 2009]. A patent for this algorithm has been submitted [Katz, Weissman, and Zigel, 2010].
In another research, performed in cooperation with Dr. Yoram Etzion, we designed an algorithm to detect atrial electrical activity and refractoriness of rodents based on the surface ECG signal [Zigel, Mor, et al., 2009]; this algorithm can advance the tachycardia research and clinical applications in animal models and human.
Speech Research - Speaker Age Estimation from Speech Signals
Human age (speaker age) is a part of non-verbal information of a speech session that recently gained increasing importance in improving speech-based applications. For interactive voice response (IVR) systems, age information helps to adapt it to the user and can give more natural human-machine interaction; this can help children as well as elderly in various applications. Our research was granted by the Israeli chief scientist, Ministry of Industry, Trade & Labor [MAGNETON, 2008-2009] for cooperation with PuddingMedia LTD. We analyzed speech signals from different subjects in order to develop an automatic system that can estimate the speaker's age. Different aspects were examined and developed, acoustic features, statistical classifiers [Porat, Lange, and Zigel, 2010], optimization techniques [Dobry, Hecht, Avigal, and Zigel, 2009] [Dobry, Hecht, Avigal, and Zigel, 2010] and hybrid approaches [Hecht, Hezroni, Manna, Aloni-Lavi, Dobry, Alfandary, and Zigel, 2009].
Speech Research – Robust Speaker Recognition and Room-Volume Estimation
This research was performed in cooperation with Prof. Boaz Rafaely from the department of Electrical and Computer Eng, BGU. The influence of acoustic reverberation was researched in order to develop a robust speaker recognition system and a system for estimation of the volume of a room. The research was granted, partially, by the Israeli chief scientist, Ministry of Industry, Trade & Labor [MAGNETON, 2006-2007] for cooperation with NICE Systems LTD (Zigel was a PI in the first year of the MAGNETON from NICE systems side). Different aspects were examined and developed. We investigated the effect of reverberation on the optimal GMM order of speaker recognition systems [Shabtai, Zigel, and Rafaely, 2008a], and on feature normalization [Shabtai, Zigel, and Rafaely, 2008b] [Shabtai, Rafaely, and Zigel, 2011] [Shabtai, Rafaely, and Zigel, 2010]. A robust speaker recognition system was developed [Peer, Rafaely, and Zigel, 2008a] [Peer, Rafaely, and Zigel, 2008b] [Peer, Rafaely, and Zigel, 2008c]. A novel system for room-volume estimation was developed; a system based on room impulse response (RIR) parameters [Shabtai, Zigel, and Rafaely, 2009a] [Shabtai, Zigel, and Rafaely, 2009b] [Shabtai, Zigel, and Rafaely, 2010a], and a system based on speech features [Shabtai, Zigel, and Rafaely, 2010b].
Smart Medical Home - Fall Detection System of Elderly
Falls are a major risk for elderly people living independently. Rapid detection of fall events can reduce the rate of mortality and raise the chances to survive the event and return to independent living. In the last two decades, several technological solutions for detection of falls were published, but most of them suffer from critical limitations. This research was performed in cooperation with Prof. Israel Gannot from the department of Biomedical Eng, TAU. In this research, we presented a proof of concept to an automatic fall detection system for elderly people [Litvak, Zigel, and Gannot, 2008a] [Litvak, Zigel, and Gannot, 2008b]. The system is based on floor vibration and sound sensing, and uses signal processing and pattern recognition algorithm to distinguish between fall events and other events. The classification is based on special features like shock response spectrum and mel frequency cepstral coefficients. For the simulation of human falls we have used a human mimicking doll: “Rescue Randy.” The proposed solution is unique, reliable, and does not require the person to wear special devices. It is designed to detect fall events in critical cases in which the person is unconscious or in a stressful condition. After testing the proposed system [Zigel, Litvak, and Gannot, 2009], we concluded that it can detect human mimicking doll falls with a sensitivity of 97.5% and specificity of 98.6%. A patent for this system has been submitted [Gannot, Litvak, and Zigel, 2009].
Dr. Yaniv Zigel, Head
Ilan Levy, B.Sc., Lab. Eng.
Eliran Dafna, Ph.D. student
Or Perlman, M.Sc. student
Maya Kriboy, M.Sc. student
Tal Rosenwein, M.Sc. student
Oren Elisha, M.Sc.
Nir Ben-Israel, M.Sc.
Roy Porat, M.Sc.
Dan Lange, M.Sc.
Noam Weissman, M.Sc.
Jenny Goldshtein, M.Sc.
Ahmad Erew, B.Sc.
Marwan Abu Leil, B.Sc.
Yaniv Tocker, B.Sc.
Sara Seidel, B.Sc.
Mirit Levy, B.Sc.
Dana Weiss, B.Sc.
Elinor Tzvi, B.Sc.
Haim Cohen, B.Sc.
Hila Beharav, B.Sc.
Nadav Yaniv, B.Sc.
Vadim Zviling, B.Sc.
Ofer Gil, B.Sc.
Hagar Shushan, B.Sc.
Hen Barades, B.Sc.
Research students from other labs under Yaniv Zigel supervision:
Ph.D. (with Dr.
Peer, M.Sc. (with Dr.
Litvak, M.Sc. (with
Gil Dobri, M.Sc. (The Open-University)
Past and Present Cooperation
Yoram Etzion, Cardiac
Arrhythmia Research Laboratory,
Electrical and Computer
Dr. Jacob Urkin, Clalit Health Services.
Dr. Gallya Gannot, Opticul Diagnostics Ltd.
Dr. Moshe Ben-David, Opticul Diagnostics Ltd.