Procedia Technology 11 ( 2013 ) 1122 – 1126

2212-0173 © 2013 The Authors. Published by Elsevier Ltd.Selection and peer-review under responsibility of the Faculty of Information Science & Technology, Universiti Kebangsaan Malaysia.doi: 10.1016/j.protcy.2013.12.303

The 4th International Conference on Electrical Engineering and Informatics (ICEEI 2013)

Study of Lung and Heart Diagnosis Application Based on Frequency Separation of Breath Sound

Muhammad Sukrisno Mardiyanto*, Ria Lestari Moedomo, Achmad Munir

School of Electrical Engineering and Informatics, Institut Teknologi BandungJalan Ganesha 10 Bandung, 40132, Indonesia

Abstract

In this paper, the study of lung and heart diagnosis application is investigated based on frequency separation of breath soundfrequency. The aim of study is to investigate how to separate and process the breath sound frequency into lung sound frequency and heart sound frequency. Since the lung and the heart are the vital organ of human, therefore the needs of immediate preventive and curative actions are primarily important. Here, the frequency separation of breath sound is carried out by using the modulation filtering method. From the investigation, although the results still need some more improvement, however it shows that the method is implementable for the separation of lung sound and heart sound from the breath sounds frequency. It is then useful for the development of application for lung and heart diagnosis diseases.

© 2013 The Authors. Published by Elsevier B.V.Selection and peer-review under responsibility of the Faculty of Information Science & Technology, Universiti Kebangsaan Malaysia.

Keywords: breath sound; diagnosis application; frequency separation; lung and heart sounds.

1. Introduction

Based on the report of United Nations agency for international public health, World Health Organization (WHO),Indonesia is the 3rd country after India and China which has the most number of people infected by tuberculosis(TB). Unfortunately, the treatment for the disease, i.e. tuberculosis, sometimes needs a bit longer time due to the

* Corresponding author. Tel.: +62-22-2508135; fax: +62-22-2500940.E-mail address: sukrisno@informatika.org

Available online at www.sciencedirect.com

© 2013 The Authors. Published by Elsevier Ltd.Selection and peer-review under responsibility of the Faculty of Information Science & Technology, Universiti Kebangsaan Malaysia.

ScienceDirect

1123 Muhammad Sukrisno Mardiyanto et al. / Procedia Technology 11 ( 2013 ) 1122 – 1126

growth of TB mycobacterium organism is slow in the laboratory environment which takes around 2-3 days for sputum culture or blood, and also the transportation cost to bring the sputum or blood from the location of patient to the laboratory [1-3]. In case of influenza disease that infects respiratory tract [4-6], a number of Avian Influenza (H5N1) cases have occurred in rural areas of Indonesia. Since the case can lead to severe illness or death, the detection of H5N1 case especially in rural areas requires a fast and accurate diagnosis action. However due to the difficultness in transportation and long distance of the location from some laboratory; the diagnosis action needs additional time to be carried out. These problems also happened for other diseases such as the lung diseases and the respiratory tract diseases. Usually the diagnosis tool with high accuracy requires a large investment, takes a long time for processing and is almost located in urban areas. Whilst, for the fast action, another diagnosis tool which can conduct rapid diagnosis test is the most commonly used tool, however it gives less accurate results due to its lowsensitivity and specificity [7].

In this paper, the study of lung and heart diagnosis application to be implemented in the diagnosis tool is investigated based on frequency separation of breath sound. The purpose of study is to investigate how to separate and process the breath sound frequency into lung sound frequency and heart sound frequency. In the investigation, the modulation filtering method is proposed to be applied for the frequency separation of breath sound. Some result and discussion related to the method will be presented.

2. Brief overview of modulation filtering method

Numerous researches related to the method for separation and cancellation the heart and lung sound frequencies have been investigated in the last decade [8-12]. The recursive least squares adaptive noise cancellation filtering has been implemented for heart sound frequency reduction in lung sounds recordings [8]. Whilst in [9-10], the method based on the independent component analysis has been successfully to separate the heart sound from the lung sound. Difference with the previous methods, the heart sound frequency cancellation has been proposed based on multiscale products and linear prediction [11]. In [12] which is adopted in this paper, the method to separate the breath sound frequency into the lung and heart frequencies, is approached by applying the modulation filtering.

As shown in Fig. 1, the block diagram of modulation filtering approach is proposed in [12] to blind HS separation from LS recordings. Prior to the processing, the breath sound recording is windowed and transformed using Fourier transform, for example, to generate the time-frequency response. Here, a term of acoustic frequency is applied to distinguish the generated time-frequency response from modulation frequency representation of the signal. Hence, the second transformation, i.e. modulation transform, is applied to the temporal trajectory of magnitude component of each acoustic frequency signal. The processing step to obtain the spectro–temporal signal representation of the breath sound recording is illustrated in Fig. 2. After the processing, the results of modulation spectrum will consist of information regarding the rate of signal spectral components.

Fig. 1. Block diagram of modulation filtering approach to blind heart sound (HS) separation from lung sound (LS) recordings.

windowingbreath sound

DF

T

modulation processing

bin 1

modulation processing

bin N

IDF

T-1

IDF

T -

2 windowing

windowing overlap and add

)(ms

Tm ,...,1

),( 1 mfs

),( 1 mfs

),( 1 mfs

),(~1 mfs

),( 1 mfs

),( mfs N

),( mfs N

),( mfs N

),( mfs N

)(ˆ ms

)(~ ms

HS

LS

overlap and add

),(~ mfs N

1124 Muhammad Sukrisno Mardiyanto et al. / Procedia Technology 11 ( 2013 ) 1122 – 1126

0 10 20 300

0.2

0.4

0.6

0.8

1

Modulation frequency (Hz)

Mag

nitu

de

bandpass characteristic bandstop characteristic

Fig. 2. Processing step to obtain spectro–temporal signal Fig. 3. Frequency characteristic of modulation filter

To figure out the process flow, after the breath sound is windowed and transformed, the spectral component is considered as s(f, m) where f = 1, ..., N and m = 1, ..., T, denote the short-term spectral component at the fth

frequency bin and mth time step of the short-term analysis. It should be noted that N and T denote total number of frequency bands and time steps, respectively. In case of the input is fixed frequency band f = F, so s(F, m) with m =1, ..., T, represents the Fth band temporal trajectory. Since the modulated spectral component of heart sounds typically fall between approximately 2-20 Hz, therefore 2 finite impulse response modulation filters are employed. Figure 3 plots the frequency characteristic of modulation filters used in the processing. It shows that the first response, i.e. the bandpass filter, has the low and high cut-off frequencies at 1Hz and 20Hz, respectively. Hence, the second response is the bandstop filter which has the same cut-off frequencies. In addition, the frequencies above 20Hz are kept in such away to improve the naturalness of separated signals.

3. Investigation result and discussion

Fig. 4. Sinusoidal signal and power spectrum diagram Fig. 5. Result of modulation frequency at 30Hz-2kHz acoustic frequency

In the investigation, the breath sound is recorded using VR+® of Blackberry cellular phone in the format of *.amr file. In order to be able processed using Matlab®, the *.amr recorder file has to be converted into *.wav file. Figure 4 shows the output of recorded breath sound (in arbitrary scale) for the sinusoidal signal (top) and the power spectrum diagram (bottom) after being processed using Matlab® Signal Processing Toolbox (SPT). Next, the Discrete Fourier Transform (DFT) is applied to the breath sound signals to obtain the magnitude and phase of signal. To obtain the modulation spectral analysis, it is required to conduct the modulation transform from acoustic frequency into

base transform

breath soundn

mod

. tra

nsfo

rm

f(ac

oust

ic fr

eq.)

f(ac

oust

ic fr

eq.)

m (time) fm (modulation freq.)

temporal trajectory of spectral component

1125 Muhammad Sukrisno Mardiyanto et al. / Procedia Technology 11 ( 2013 ) 1122 – 1126

modulation frequency (fm). The process of modulation transforming is focused on the breath sound frequency in the range of 30Hz–2kHz acoustic frequency. The result of modulation transform (in arbitrary scale) is depicted in Fig. 5 for modulated breath sound periodic signal (top) and modulated breath sound frequency (bottom).

The output of modulation transform is then processed using Inverse Discrete Fourier Transform (IDFT) to obtain the magnitude and phase of heart and lung sound (LS) and heart sound (HS). Next, the output of IDFT module is overlapped and added to reconstruct the LS and HS signals. The process is conducted using 2 finite impulse response modulation filters which have bandpass and bandstop characteristics. After reconstruction, the output LS signal and HS signal (in arbitrary scale) are depicted in Figs. 6 and 7, respectively. Fig. 6, it is shown 3 signals:the produced LS signal (bottom) is not yet similar with the recorded LS signal (top), however the intermediary LS signal (middle) looks more similar with the recorded LS signal (top). The similar thing also appears in Fig. 7, where the produced HS signal (bottom) is not yet similar with the recorded HS signal (top), where the intermediary LS signal (middle) looks more similar with the recorded HS signal (top). These problems are probably evoked by the last process, the overlap-and-add process, which requires redefinition and more improvement.

Fig. 6. Recorded, intermediary, and produced lung sound (LS) signals Fig. 7. Recorded, intermediary, and produced heart sound (HS) signals

4. Conclusion

The study of lung and heart diagnosis application using modulation filtering method has been investigated based on frequency separation of breath sound frequency. The investigation which has implemented using Matlab® Signal Processing Toolbox (SPT) has been applied to process the recorded breath sound from VR+® of Blackberry cellular phone. From the investigation result, although the produced lung sound (LS) and heart sound (HS) signals have not been yet similar to the recording ones which require redefinition and more improvement, however it has been shown that the method was applicable for the separation of lung and heart frequencies from the breath sounds frequency.

1126 Muhammad Sukrisno Mardiyanto et al. / Procedia Technology 11 ( 2013 ) 1122 – 1126

Therefore, the result will give more advantages in developing application for lung and heart diagnosis diseases in the future.

Acknowledgements

The authors thank Prof. Dr. Soewarno, Dean of School of Electrical Engineering and Informatics, Institut Teknologi Bandung (STEI ITB), who has given support for this research. Sincere thanks to Dr. Achmad Imam Kistijantoro, Dr. Munawar Ahmad, Dr. Bachti Alisjahbana (Med), Tjahjono Djatmiko MBA, who has given enormous contribution during the initial stage of this research series.

Special thanks to Joko Sarwono, PhD., from Engineering Physics Department, Institut Teknologi Bandung, forvaluable inputs and discussions for the current work, and also to Dr. Ary Setijadi and Dr. Rifki Wijaya from STEI ITB. Thanks and respect to Dr. Alexander Iskandar, Dr. Acep Purqon and Dr. Fourier D.E. Latief from Physics Department ITB, who have given some valuable inputs and some precious time for the study of previous research paper.

This research is partially supported by the Research Grant from Institut Teknologi Bandung (ITB) under the scheme program of Research and Innovation 2013.

References

[1] Ellner JJ. Tuberculosis. In: Goldman L, Schafer AI, eds. Goldman's Cecil Medicine. 24th ed. Philadelphia, PA: Elsevier Saunders, chap 332. 2011.

[2] Fitzgerald DW, Sterling TR, Haas DW. Mycobacterium tuberculosis. In: Mandell GL, Bennett JE, Dolan R, eds. Mandell, Douglas, and Bennett’s Principles and Practice of Infectious Diseases. 7th ed. Philadelphia, PA: Elsevier Churchill-Livingstone, chap 250; 2009.

[3] Becker KW. Research into Adventitious Lung Sound Signals Originating From Pulmonary Tuberculosis Using Electronic Auscultation, Department of Mechanical and Mechatronic Engineering, Faculty of Engineering, Stellenbosch University; 2009.

[4] WHO. Avian influenza – situation in Indonesia – update, January 19th, http://www.who.int/csr/don/2012_01_19/en/index.html; 2012[5] Grimes A. Influenza: Seasonal Flu and H1N1. Symptoms, Treatment, and Prevention. Continuing Education Courses for Healthcare

Professionals, Wild Iris Medical Education, Inc; 2010.[6] Uyeki TM. Human Infection with Highly Pathogenic Avian Influenza A (H5N1) Virus. Review of Clinical Issues, Emerging Infections, CID

49; 2009.[7] Journal Watch. H1N1 Rapid Tests: Poor Sensitivity, 2009;12:11.[8] Gnitecki J. Moussavi Z. Pasterkamp H. Recursive least squares adaptive noise cancellation filtering for heart sound reduction in lung sounds recordings. In: Proc. of IEEE International Conference on Engineering in Medicine and Biology Society 2003. p. 2416-2419[9] Pourazad M. Moussavi Z. Farahmand F. Ward R. Heart sounds separation from lung sounds using independent component analysis. In: Proc. of IEEE International Conference on Engineering in Medicine and Biology Society 2005. p. 2736–2739.[10]Chien JC. Huang MC. Lin YD. Chong FC. A study of heart sound and lung sound separation by independent component analysis technique.

In: Proc. of IEEE International Conference on Engineering in Medicine and Biology Society 2006. p. 5708-5711.[11]Tapia DF. Moussavi Z. Thomas G. Heart sound cancellation based on multiscale products and linear prediction,” IEEE Trans. on Biomedical

Eng. 2007; 54:2. p. 234-243.[12]Falk TH. Chan WY. Modulation filtering for heart and lung sound separation from breath sound recordings. In: Proc. of IEEE International

Conference on Engineering in Medicine and Biology Society 2008. p. 1859–1862.[13]Ria Lestari, Munawar Ahmad, Bachti Alisjahbana, Tjahjono Djatmiko. The Lung Diseases Diagnosis Software: Influenza Case Study,

International Conference on Rural Information and Communication Technology (rICT) 2011, School of Informatics and Electrical Engineering, Institut Teknologi Bandung, November 2011.

[14]Ria Lestari, Munawar Ahmad, Bachti Alisjahbana, Tjahjono Djatmiko The Lung Diseases Diagnosis Software: Influenza and Tuberculosis Cases Study, International Conference on Biomedical Science (ICBS) 2012, School of Life Science and Technology, Institut Teknologi Bandung, February 2012.

[15]Ria Lestari, Munawar Ahmad, Bachti Alisjahbana, Tjahjono Djatmiko. The Lung Diseases Diagnosis Software: Influenza and Tuberculosis Case Studies in Cloud Computing Environment, International Conference on Cloud Computing and Social Networking (ICCCSN) 2012, School of Informatics and Electrical Engineering, Institut Teknologi Bandung, April 2012, 26-27 April 2012.

[16]Ria Lestari Moedomo, M. Sukrisno Mardiyanto, Munawar Ahmad, Bachti Alisjahbana, Tjahjono Djatmiko. The Breath Sound Analysis For Diseases Diagnosis and Stress Measurement, Internasional Conference on System Engineering and Technology (ICSET) 2012, School of Informatics and Electrical Engineering, Institut Teknologi Bandung,10-12 September 2012.

[17]Ria L. Moedomo, M. Sukrisno Mardiyanto, Achmad Munir. Preliminary Study : The Breath-Sound Frequency Division For The Lung Diseases Diagnosis Application, International Conference on Mathematics and Natural Science (ICMNS) , Physics Department ITB, 2012.