Printer Friendly

A survey on automated wheeze detection systems for asthmatic patients.


Nowadays, asthma is becoming a common disease that may occur at any age and have become a public health challenge to the world today [1]. It is a chronic inflammatory diseases of the respiratory airway and can be hyper- responsiveness to a variety of stimuli [2]. The asthmatic patient suffers attacks such as coughing, dyspnea, and the main manifestation is wheezing [3]. Sounds generated during breathing can be a good source of information on lung's health [4]. Any characteristic changes of the normal lung sounds can imply a diseased condition that probably is invading the lung. Each type of disease is different from each other and the variation can be ascertained from sound characteristic, pitch, amplitude, frequency, duration, etc. [5]. With regard to asthma, symptoms originating from the wall oscillations of narrowed airways at critical flow rates causes wheeze to occur [3]. Wheeze is one of the adventitious sounds present in lung that is clinically defined as abnormal. The presence of wheeze, its location, duration and its relation to the respiratory cycle can be very useful to assist the physician as it has become a crucial practice in diagnosing and managing a number of pulmonary pathologies such as chronic obstructive pulmonary disease (COPD), bronchiolitis and commonly asthma [6]. Wheezes are continuous adventitious sounds that are superimposed on the normal breath sounds. According to the American Thoracic Society (ATS), the word "continuous" can be defined as the duration of the wheeze that is longer than 250 ms. The ATS also defines wheezes as high-pitched continuous sounds with a dominant frequency of 400 Hz or more. Wheezes can be detected and classified based on the frequency characteristics of its sinusoidal waves that justifies the musical character of the wheeze [7]. Conventionally, stethoscope is used to diagnose and monitor wheezes in asthmatic patients. Although it is well known that auscultation with stethoscope is reliable, fast and non-invasive, continuous monitoring of the respiration condition is impossible [8, 9]. Due to increasing number of asthmatic patients at present, there is a growing demand for automatic monitoring of the wheeze to assist the physicians in diagnosing and monitoring the patient. For asthmatic patients, continuous and automatic monitoring is essential as the daily symptoms can provide crucial information to the medical diagnosis [8]. Therefore, the electronic stethoscope, which is capable of recording and storing lung sounds, is available for many years now. This stethoscope can not only store the data obtained, but such data can be retrieved in future to aid in the interpretation of disease by medical personnel [10]. However, the problem seems to amplify as different physicians interpret the lung sounds differently. To overcome these problems, computerized approach has been developed over the past three decades for automated wheeze detection [11]. It is a bit time-consuming, but low in cost and reliable. Many researchers were involved in developing and improving these automated systems and many have succeeded in their research. A survey of literature shows that the main methodologies can roughly be classified into two categories: Fourier peaks detection and spectrogram image analysis. This review paper is organized as follows. In methodology section, we present how articles have been searched and selected for the review in this paper. In the results section, information from all the chosen papers is summarized in a tabular form and an explanation on the content of these papers is given. In the following section, further discussion on the automated wheeze system based on the table developed is presented. And finally, in conclusion, we show how the system can be very beneficial for asthmatic patients. Some avenues for future work are also suggested.


Samples and Methodology

A systematic search of articles published as early as 1985 to 2012 was conducted in Universiti Malaysia Perlis (UniMAP). Keywords that were used throughout the search are lung auscultation, lung sound, wheezes, crackle, adventitious sounds and others that are synonymous to the search criteria. Only papers with important information for automated wheeze detection and papers published in English only were reviewed. Some papers may have presented other adventitious sound together with wheezes but only wheeze detection method will be taken into consideration. Moreover, abstracts that have been published as workshop, conference or symposium proceedings were not be considered due to insufficient information presented in them.



A total of 94 articles were found initially. After, elimination of duplicates and obviously irrelevant titles (n = 67), only 27 articles chosen for this systematic review (see Figure 1). From the total of 94 papers obtained, most of the discarded papers were not related to the discussed automated wheeze detection system. They included several other respiration sounds that are not related to asthma or their study design did not even include any wheezes present even though other lung sounds were discussed. None of the papers included in this systemic review were books or other reports not subject to peer review. From 27 papers chosen, only 21 articles which satisfied the selection criteria are discussed briefly in Table 1. As source of the data obtained is important in developing an automated system, we have listed all the hardware involved in data collection method. It is assured that data collected are from patients suffering from airway obstruction, which led to wheeze formation and the most are from asthmatic patients. No hardware is presented in case researchers used database as their data source. Listed hardware can be very useful for new researchers in developing new automated system for their research. In collecting the data, the placement of the auscultation hardware has been presented as well. This is due to that each point of data collection can result in different quality of the obtained signal. Wheezes have some characteristics that may produce a better signal for certain auscultation placement. Therefore, the placement of auscultation hardware presented may be very useful for collecting the best data for wheezes. The processing method together with the extracted feature used has been listed in Table 1. This seems to be the most crucial part in the successful of an automated system. A poor selection of the processing method and a feature to be extracted from data may cause a system to fail. Therefore, from the table, researchers will be able to choose the best processing method and a best feature to be extracted for a better automated system development. Result from every article may also act as a good reference for the system to be developed.


The authors strongly believe that this article is first of its kind since no previous work on summarizing the previous research on automatic wheeze detection systems beneficial seems to be available. In this section, the data collection procedures and the data processing methods by previous researchers are discussed. Few recommendations based on the analysis from the previous work are also suggested.

Data Collection

Nowadays, there are several methods used clinically in detecting wheeze. Methods such as spirometer and other pulmonary function tests are widely used. For an example, Jane et al. [14] in their studies have proved that respiratory sound analysis complementary to the classical spirometry has showed that the automated system can also be reliable as a pulmonary test for detecting wheeze. These methods are time-consuming and also extra effort is needed on part of the patient for such tests. The simplest method to detect wheeze is using microphone mounted on a stethoscope. This method is non-invasive and not time-consuming. This has drawn much attention from researchers to develop automated wheeze detection system. Since automated auscultation technique is not yet implemented in current medical application, the use of current pulmonary function test along with acoustic analysis can improve the reliability of such automated system. While, Hossain [15] have introduced the lung sound-flow rate model that can be used for the detection of airway narrowing occurring in asthmatic patients, Cortes and team [16] in their work proved the unnecessary of using the additional details together with lung sounds. They have successfully developed a system to assist the patient on whom spirometric test cannot be carried out by solely using lung sound signal. From Table 1, it can be observed that most of the previous researchers have collected data by placing the microphone over the trachea region [2, 4, 6, 10, 14, 17, 22, 23]. Riella et al. [12] stressed the reliability of the trachea to be a better location for analyzing wheezes than the lung while Oud et al. [13] claimed that trachea is reliable because all air-propagated lung sound from the two lungs integrates in trachea. Moreover, more frequency information is preserved at the trachea, as the chest wall filters out higher frequencies [12]. This shows that the trachea is the most favorable and reliable position for wheeze data collection.

Data Processing

There are several methods being used by the previous researchers to process the lung sounds for detecting wheeze. The lung sounds are either processed in time domain or frequency domain. The lung sounds are non-stationary and nonlinear. Therefore, analyzing the lung sounds in the time-frequency domain will be more helpful for getting valuable information from the non-stationary signals [26]. Regardless the usefulness of time-frequency domain for analyzing lung sounds, very few researchers have used the methods according to Table 1. The work of Taplidou et al. [25] is the only work being carried out using time-frequency domain. For recognizing the wheeze, the previous researchers have opted to use statistical analysis or machine learning methods. Method such as Fisher Discriminant Analysis (FDA), kurtosis and linear regression analysis are the most commonly used for statistical analysis while neural network (NN), radial basis function (RBP), learning vector quantization (LVQ), Gaussian mixture models (GMM), fast Fourier transform are the most commonly used for machine learning methods. Methods such as genetic algorithm, support vector machine, and practical swarm optimization are yet to be explored in developing the wheeze detection system. The researchers also have not yet touched upon the area of hybrid machine learning algorithm in the detection of wheezes. Hybrid algorithms such as neuro-fuzzy, genetic fuzzy systems, evolutionary neural networks etc, have proved to be effective with other applications [27]. It is recommended to use such algorithms for wheeze detection in the future. As far as the extracted feature are concerned, various type of features have been used by the previous researches such as time-frequency spectrum, entropy, Mel frequency ceptral coefficients (MFCC), power spectral density (PSD), Mean, standard deviation (SD), Peak Frequency (FP), kurtosis, mean crossing, welch spectra, and subband based cepstral (SBC) [3, 4, 8, 9, 11, 17, 19, 21]. It is impossible to choose which features is the best to be extracted, this is because by using the same feature, a system performance can be differ with each other due to different signal processing and classification technique applied. For an example, Corbera et al. [19] successfully obtained 100% specificity while Hsueh et al. [22] only managed to obtained 89% specificity for their system eventhough both of the system uses PSD [19, 22]. Moreover, the higher specificity obtained by Corbera et al. also due to the use of PSD together with SD features resulting a better result for their system. From Table 1, can observed that the most feature used is PSD [6, 8, 14, 16, 18, 22].


This review provides an insight on automatic wheeze detection systems developed by previous researchers and its potential in both clinical and research setting. The literature review was carried out on the 27 articles that satisfied the selection criteria. A brief summary on the 21 out of 27 articles was tabulated. The table clearly discussed the hardware used, sensor placement, data processing methods, feature extraction methods and important outcome of the 21 articles. A brief discussion was done based on the previous works on automated wheeze detection system. Few recommendations on the placement of the sensor and methods that can be applied in future research were also discussed. The research on the developed system has been carried out for the past three decades. Even though wheeze detection systems have been successfully implemented, they are only at research level and still commercially available real-time wheeze detection system is yet to be developed. Therefore, the future researchers are encouraged to develop commercial system that will improve the diagnosis of wheeze in clinical environment.


We confirm that this manuscript has not been published elsewhere and is not under consideration by another journal. All authors have approved the manuscript and agree with submission to the Journal of the Association of Basic Medical Sciences of Federation of Bosnia and Herzegovina. The authors have no conflicts of interest to declare.


[1] Niksic D., Saracevic E., Cemerlic-Kulic A., Kurspahic-Mujcic A., Bajraktarevic S., Niksic H. Evaluation of efficiency practical issues in the management of childhood asthma. Bosn J Basic Med Sci. 2005;5(4):40-5.

[2] Mehmedagic A., Vranic E., Hadzovic S., Pokrajac M., Milic J., Mehic B., Zutic H., Konjicija A. The new trends in theophylline therapy. Bosn J Basic Med Sci. 2002;2(1-2):62-65.

[3] Forkheim, Kevin E., Scuse D., Pasterkamp H. A comparison of neural network models for wheeze detection. WESCANEX 95 Communications, Power, and Computing Conference Proceedings, IEEE 1995; 1:214-219.

[4] Oud M. Lung function interpolation by means of neural-network-supported analysis of respiration sounds. Med Eng Phys 2003; 25(4): 309-316.

[5] Jain A, Vepa J, editors. Lung sound analysis for wheeze episode detection. 30th Annual International Conference of the IEEE on Engineering in Medicine and Biology Society; 2008: 2582-2585.

[6] Fenton T.R., Pasterkamp H., Tal A., Chernick V. Automated spectral characterization of wheezing in asthmatic children. IEEE Transactions on Biomedical Engineering 1985; 32 (1):50-55.

[7] Parkhi A., Pawar M., editors. Analysis of Deformities in Lung Using Short Time Fourier Transform Spectrogram Analysis on Lung Sound. International Conference on Computational Intelligence and Communication Networks (CICN) IEEE; 2011:177-181.

[8] Jane R., Cortes S., Fiz J., Morera J., editors. Analysis of wheezes in asthmatic patients during spontaneous respiration. 26th Annual International Conference of the IEEE on Engineering in Medicine and Biology Society, IEMBS'04; 2004: 3836-3839.

[9] Zhang J., Ser W., Yu J, Zhang T., editors. A novel wheeze detection method for wearable monitoring systems. International Symposium on Intelligent Ubiquitous Computing and Education IEEE, 2009; 331-334.

[10] Marshall A., Boussakta S., Pearson S., editors. Applications of signal recognition algorithms to diagnosis and monitoring in chest medicine. The 3rd IEEE International Seminar on Medical Applications of Signal Processing, 2005: 121-124.

[11] Bahoura M., Pelletier C., editors. Respiratory sounds classification using Gaussian mixture models. Canadian Conference on Electrical and Computer Engineering IEEE, 2004; 3:1309-1312.

[12] Oud M., Maarsingh E. Spirometry and forced oscillometry assisted optimal frequency band determination for the computerized analysis of tracheal lung sounds in asthma. Physiol Meas 2004;25(3):595-606.

[13] Riella R., Nohama P., Borges R., Stelle A., editors. Automatic wheezing recognition in recorded lung sounds. Proceedings of the 25th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2003; 3:2535-2538.

[14] Jane R., Salvatella D., Fiz J., Morera J., editors. Spectral analysis of respiratory sounds to assess bronchodilator effect in asthmatic patients. Proceedings of the 20th Annual International Conference on Engineering in Medicine and Biology Society; 1998; 6: 3203-3206.

[15] Hossain I., Moussavi Z., editors. Finding the lung sound-flow relationship in normal and asthmatic subjects. 26th Annual International Conference of the IEEE on Engineering in Medicine and Biology Society, IEMBS'04, 2004; 2: 3852-3855.

[16] Cortes S., Jane R., Fiz J., Morera J., editors. Monitoring of wheeze duration during spontaneous respiration in asthmatic patients. 27th Annual International Conference of the IEEE on Engineering in Medicine and Biology Society, IEEE-EMBS 2005; 6141-6144 .

[17] Mayorga P., Druzgalski C., Morelos R., Gonzalez O., Vidales J., editors. Acoustics based assessment of respiratory diseases using GMM classification. Annual International Conference of the IEEE on Engineering in Medicine and Biology Society (EMBC), 2010: 6312-6316.

[18] Homs-Corbera A., Jane R., Fiz J.A., Morera J., editors. Algorithm for time-frequency detection and analysis of wheezes. Proceedings of the 22nd Annual International Conference of the IEEE on Engineering in Medicine and Biology Society, 2000;4: 2977-2980.

[19] Jin F., Sattar F., Goh D.Y.T., editors. Automatic wheeze detection using histograms of sample entropy. 30th Annual International Conference on Engineering in Medicine and Biology Society, 2008 EMBS 2008: 1890-1893.

[20] Aydore S., Sen I., Kahya Y.P., Mihcak M.K., editors. Classification of respiratory signals by linear analysis. Annual International Conference on Engineering in Medicine and Biology Society IEEE, EMBC; 2009: 2617-2620.

[21] Wisniewski M., Zielinski T., editors. Digital analysis methods of wheezes in asthma. International Conference on Signals and Electronic Systems IEEE (ICSES), 2010: 69-72.

[22] Hsueh M.L., Chien J.C., Chang F.C., Wu H.D., Chong F.C., editors. Respiratory wheeze detection system. 27th Annual International Conference of the Engineering in Medicine and Biology Society, 2005 IEEE-EMBS ; 2006: 7553-7559.

[23] Homs-Corbera A., Fiz J.A., Morera J., Jane R. Time-frequency detection and analysis of wheezes during forced exhalation. IEEE Transactions on Biomedical Engineering, 2004;51(1):182-186.

[24] Riella R., Nohama P., Maia J. Method for automatic detection of wheezing in lung sounds. Braz J Med Biol Res 2009;42(7):674-684.

[25] Taplidou S.A., Hadjileontiadis L.J. Wheeze detection based on time-frequency analysis of breath sounds. Comput Biol Med 2007; 37(8):1073-1083

[26] Kandaswamy A., Kumar C.S., Ramanathan R.P., Jayaraman S., Malmurugan N. Neural classification of lung sounds using wavelet coefficients. Comput Biol Med 2004; 34(6)523-537.

[27] Yang J.Y., Yang M.Q., Luo Z., Ma Y., Li J., Deng Y., et al. A hybrid machine learning-based method for classifying the Cushing's Syndrome with comorbid adrenocortical lesions. BMC Genomics. 2008; 9 (Suppl 1):S23.

Submitted: 4 September 2012 / Accepted: 16 October 2012

Syamimi Mardiah Shaharum (1) *, Kenneth Sundaraj (1), Rajkumar Palaniappan (1)

(1) Al-Rehab Research Group, Universiti Malaysia Perlis (UniMAP), Kampus Pauh Putra, 02600 Perlis, Malaysia

* Corresponding author: Syamimi Mardiah Shaharum, AI-Rehab Research Group, Universiti Malaysia Perlis (UniMAP), Kampus Pauh Putra, 02600 Perlis, Malaysia Phone: +6049767399; Fax : +6049885167 e-mail:
TABLE 1. An overview of developed wheeze automated systems

Author              Year   Hardware and          Placement

Forkheim et         1995   Eight microphones     Chest and back
  al. (3)
Zhang et al. (9)    2009   Microphones           NA
                             WM-64 PN)
Mayorgaetal. (17)   2010   NA (database)         NA
Coberaetal. (18)    2000   Piezoelectric PPG     NA
Jane et al. (8)     2004   Microphones and       trachea
Fenton et           1985   Two contact           Lung and trachea
  al. (6)                    accelerometer
                             EMT-25B, Elema
Feng Jin et         2008   Single Electrets      Trachea
  al. (19)                   Condenser
                             (ECM-77B, Sony)
Riella et           2003   NA (database)         NA
  al. (13)
Aydore et           2009   14 Electret           Chest wall
  al. (20)                   Microphones
                             (Sony ECM
                             -44 BPT),
Wisniewski et       2010   14 microphones        Chest wall
  al. (21)                   (Sony ECM
M. Oud (4)          2003   Air Coupled           Trachea
Jain et al. (5)     2008   NA (Database)         NA
Bahoura et          2004   NA (Database)         NA
  al. (11)
Cortes et           2005   Piezoelectric PPG     tracheal
  al. (16)                   sensor (Technion,
Hsueh et al. (22)   2005   Microphone Sensor     thorax
Jane et al. (14)    1998   PFT Horizon           Near the cricoids
                             Spirometer            cartilage
Corbera et          2004   Spirometry (PFT       NA
  al. (23)                   Spirometer,
                             Horizon, USA),
                             Germany), PPG
Hossain et          2004   Piezoelectric         right upper lung
  al. (15)                   accelerometer         lobe
                             (Siemens EMT
Oud et al. (12)     2004   Handmade Air          Trachea
                             Spirometer (Vmx
                             R.O. S.,
                             The Netherlands)
Riella et           2009   NA (Database)         Trachea
  al. (24)
Taplidou et         2007   Five Electrets        Trachea
  al. (25)                   Condenser
                             (ECM77B, Sony,
                             (974010, Siemens,

Author              Processing method     Features

Forkheim et         Back propagation      FT Spectrum
  al. (3)             (BP), Neural
                      Network (NN),
                      Fast Fourier
                      Transform (FFT),
                      Radial Basis
                      Function (RBP),
                      Self Organizing
                      Map (SOM),
                      Learning Vector
Zhang et al. (9)    Entropy-Based         Entropy
                      Wheeze Detection
                      (EBWD), STFT
Mayorgaetal. (17)   Finite Impus          Mel Frequency
                      Response              Cepstral
                      (FIR) filter,         Coefficients
                      Hamming               (MFCC)
                      window, Fast
                      Models (GMM)
Coberaetal. (18)    Local Adaptive        Power Spectral
                      Wheezes Detection     Density (PSD),
                      Algorithm             Mean, Standard
                      (LAWDA), Hanning      Deviation (SD)
Jane et al. (8)     Autoregressive        PSD, Peak
                      (AR) model            Frequency (FP)

Fenton et           EFT                   Power Spectra
  al. (6)
Feng Jin et         Short-Time Fourier    Mean distortion
  al. (19)            Transform (STFT),     of histogram
                      Sample Entropy Fast
                      Gabor Spectrogram
Riella et           FIR filter, FFT,      Spectral Mean
  al. (13)            Hanning window,
                      Perceptron, BP
Aydore et           Fisher                Kurtosis, Renyi
  al. (20)            Discriminant          entropy, f50/
                      Analysis              f90 ratio, mean
                      (FDA), Neyman         crossing
Wisniewski et       Teager Energy         Kurtosis, Renyi
  al. (21)            Operator (TEO),       entropy, f50/
                      Otsu Threshold,       f90 ratio
                      STFT, Sample
M. Oud (4)          Feed Fo ward NN,      Welch spectra
                      Hanning Window,
                      Feed Forward NN
Jain et al. (5)     FFT, Hanning          Spectral analysis
Bahoura et          Gaussian Mixture      MFCC. Subband
  al. (11)            Models (GMM).         Based Cepstral
                      MLP.VQ                (SBC)
Cortes et           AR                    PSD,FP
  al. (16)
Hsueh et al. (22)   Shabtai-Musih         PSD
                      Algorithm, Hanning
                      Window, LAWDA
Jane et al. (14)    AR                    PSD

Corbera et          LAWDA--Shabtai        Average Mean
  al. (23)            Musih                 Frequency
Hossain et          Hilbert               Mean amplitude,
  al. (15)            Transformation,       mean flow,
                      FFT, Linear           average power
Oud et al. (12)     Hanning window,       Welch Spectra
Riella et           STFT, FFT, ANN        Spectral analysis
  al. (24)
Taplidou et         STFT, Hanning         TF analysis
  al. (25)            Window

Author              Result

Forkheim et         Comparison between
  al. (3)             types of
                      neural network
                      obtained, 94.5%
                      accuracy for RBF,
                      91% accuracy for
                      SOM, 94.5%
                      accuracy for LVQ
Zhang et al. (9)    Able to identify
                      85% wheezes
                      samples and
                      systems have
                      been implemented
                      into wearable
Mayorgaetal. (17)   Accuracy of 98.7%
Coberaetal. (18)    Algorithm has
                      good sensitivity
                      (100% to 70%),
Jane et al. (8)     Higher value of
                      peak frequency
                      were found in
                      patients with
Fenton et           Trachea is a better
  al. (6)             source for wheeze
Feng Jin et         97.9% accuracy for
  al. (19)            wheezes during
                      expiration while
                      85.3% accuracy
                      for wheezes in
Riella et           Proposed technique
  al. (13)            can be quite
                      useful mainly
                      for a continuous
                      analysis using
                      many respiratory
                      cycle from
Aydore et           95.1% accuracy for
  al. (20)            training and
                      93.5% accuracy
                      for testing
Wisniewski et       Review proved that
  al. (21)            measurement of
                      wheezes in
                      e-health systems
                      can be a very
                      powerful tool.
M. Oud (4)          Variance
                      can be utilized
                      to reduce the
                      dimension of
                      the spectral
                      data by means
                      of Principal
                      Analysis (PCA).
Jain et al. (5)     Yielded system
                      with 84%
                      and 86%
Bahoura et          GMM-MFCC
  al. (11)            successfully
                      classify sounds
                      into two
                      category (wheeze
                      and normal
Cortes et           System can be used
  al. (16)            for patients with
                      high obstruction
                      on whom
                      test cannot be
                      carried out.
Hsueh et al. (22)   System have been
                      tested and obtained
                      sensitivity and
Jane et al. (14)    Respiratory sound
                      analysis showed
                      to be
                      to classical
Corbera et          An algorithm to
  al. (23)            detect wheeze in
                      plane, with
                      was developed
                      and validated by
                      medical doctor.
Hossain et          The LS-Flow
  al. (15)            model may be
                      used as means
                      of detecting
                      the airway
                      narrowing in
Oud et al. (12)     Forced Expiratory
                      Volume in 1 s
                      (FEV1) versus
                      (Rrs (4))
                      parameter found
                      that Rrs is
                      to FEV.
Riella et           The successful use
  al. (24)            of the technique
                      can be helpful
                      in clinical
                      diagnosis mainly
                      when the
                      analysis needed
                      to be performed
                      continually using
                      many respiratory
                      cycles from the
                      same patient.
Taplidou et         The system developed
  al. (25)            can be successfully
                      used due to its noise
                      robustness, fast
                      implementation and
                      increased accuracy.
COPYRIGHT 2012 Association of Basic Medical Sciences
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2012 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Shaharum, Syamimi Mardiah; Sundaraj, Kenneth; Palaniappan, Rajkumar
Publication:Bosnian Journal of Basic Medical Sciences
Article Type:Report
Geographic Code:4EXBO
Date:Nov 1, 2012
Previous Article:Assessment of relatedness between neurocan gene as bipolar disorder susceptibility locus and schizophrenia.
Next Article:Trefoil factor 3 isolated from human breast milk downregulates cytokines (IL8 and IL6) and promotes human beta defensin (hBD2 and hBD4) expression in...

Terms of use | Privacy policy | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters