- Email: [email protected]
Fourier analysis of infantile stridor: preliminary data
InternationalJournalofPediatricOtorhinolao'ngologv,
10 (1985)191-199
191
Elsevier POR 00336
Fourier analysis of infantile stridor: preliminary data Lincoln Gray 1, James C. Denneny, III 2, Hugo Carvajal 3 and Robert Jahrsdoerfer 1 t Department of Otolaryngology- Head and Neck Surgery, University of Texas Medical School, Houston, TX 77030; 2 University of lndiana, Riley's Children Hospital, Department of Otolaryngology, Indianapolis, IN 46223; and J Department of Pediatrics, P.O. Box 20708, Houston, TX 77725 (U.S.A.)
(Received November 8th, 1984) (Revised version receivedFebruary 2nd, 1985) (Accepted July 28th. 1985)
Key words: stridor-Fourier analysis-spectral analysis-computers-neonates-airway
Summary Preliminary data from 3 patients suggest that computerized (Fourier) analysis of infantile stridor can be instructive. Several recordings are quickly collected with a microphone and digital oscilloscope at the patient's bedside. The data are later sent to a computer for spectral analysis. Averaging of several spectra from each patient depicts only those sounds that are consistent from sample to sample. Subtraction of background noise from the averaged stridor removes all but those sounds that are produced by the patient. The results show that the spectra are relatively consistent from sample to sample within the same patient, and that different patients with different pathologies have distinct patterns in their spectra. It thus appears that further acoustical studies of infantile stridor will be productive. We are optimistic that data from a larger series of patients will indicate those spectral patterns which are characteristic of a specific laryngotracheal pathology, and perhaps facilitate rapid diagnosis without invasive procedures. An additional potential of this analysis is that pre- and post-operative spectra can be subtracted to show those sounds that improved or worsened. Such serial comparisons during the management of airway problems could be useful in evaluating treatment.
Correspondence: L. Gray, Department of Otolaryngology-Headand Neck Surgery. University of Texas
Medical School, Houston, TX 77030, U.S.A. 0165-5876/85/$03.30 © 1985 ElsevierScience Publishers B.V. (Biomedical Division)
192 Introduction
It is important for the practicing Otolaryngologist to correctly manage pediatric airway problems. Frequently, these patients present with noticeable stridor. The term stridor comes from Latin meaning a harsh and creaking sound [13]. It is noisy respiration produced by turbulent airflow through a partial obstruction [4]. Clinical evaluation of these abnormal breathing sounds gives valuable clues about the underlying pathology and hence to the proper course of treatment. High-pitched stridor, for example, may suggest the presence of a glottic or subglottic lesion [5]. Stridor with a zone of intensity around 2 kHz may be indicative of bronchial pathology [9]. Cotton and Reilly have stated that 'in cases of stridor, the most important part of the examination is auscultation' [4]. The clinical art of listening to these abnormal sounds has been enhanced by recent technological advances. Sonographic techniques have permitted some acoustical analysis of infants' cries and breathing sounds [9,11]. This acoustical analysis revealed 4 basic categories of stridor: pharyngeal, laryngeal, subglottic and bronchial. Sonograms provide a visual representation of the stridorous sounds, but the information is not stored on a computer and thus difficult to use in statistical analyses. A completely digital acoustical analysis, such as the Fast Fourier transform (FFT), might provide as much useful information as the analog acoustical analysis (sonograms) yet be in a form where further statistical evaluation of the stridors will be easy. Fourier analysis has been used recently in studies of cries [6,11] and sounds generated within infected lungs [10]. Cotton and Reilly hope 'that with the help of computers, data could be stored and compared to assist physicians in accurate clinical impressions and as an adjunct to endoscopic evaluation. When following chronic respiratory conditions, such data could be used for monitoring possible improvement or regression' [4]. Thus, the purpose of this study was to provide some preliminary evaluation of data collected from a computerized acoustical analysis of infantile stridor. Hopefully, these procedures can then be used in a more extensive series of patients to objectively evaluate any correlation among the acoustical characteristics of stridor with clinical diagnosis, prognosis and recovery.
Methods
Data were collected at the patient's bedside with a sound level meter (Bruel and Kjaer model 2215) and a digital oscilloscope (Nicolet model 3091). The sound level meter was used as a high-quality calibrated microphone and amplifier system. The microphone was held 2-10 cm from the patient's mouth. The analog output from the sound level meter was displayed and then stored by the digital oscilloscope positioned close to the patient's bed. The sensitivities on both the sound level meter and the oscilloscope were adjusted to produce a signal that usually filled the oscilloscope screen but did not exceed its limits. The oscilloscope is capable of storing up to ten 2000-point samples on a bubble-memory cassette. The data were digitized at a rate
193 of one 12-bit point every 200 /~s (5 kHz). Each sample thus represents 0.4 s of recorded stridor. This allows a Fourier analysis of frequencies between approximately 2 and 2500 cycles per second. Usually, 8 different samples of the stridor were collected, and then two samples of the background noise were collected for use in the subsequent computerized analysis. Collection of these data required about 10 min at the patient's bedside, but more than half of this time was required to set up the equipment. After collection of these data in the hospital the oscilloscope was carried to a laboratory where the 10 stored samples were retrieved from the bubble memory and transmitted to a computer (LSI 11/03). Twenty-four zeros were added to the front and back of the data, to produce a power-of-two number of samples, and then the data were multiplied by a Hanning window to remove the unwanted effects of spectral spread or 'leakage' produced by discrete Fourier analysis [7]. The data were then subjected to a fast Fourier analysis [2,3,14] and the resulting power spectrum was stored in the computer. To produce the graphs shown in Fig. 1, the power spectra from individual samples of the patient's stridor were averaged. Averaging power spectra from different samples of the same sound depicts the acoustical characteristics that are consistent from sample to sample [1]. From this average of sampled stridors, the average of the power spectra from the background noise was subtracted. Subtracting the power spectra of background noises eliminates the contribution of extraneous sounds, and the powers that remain show only those portions of the recorded sounds that were produced by the infant. The spectra displayed are smoothed slightly, by computing a running average of each 10 adjacent spectral lines. This smoothing removes some of the variation in the curves that tends to obscure visual evaluation of general patterns. In summary, it is believed that the power spectra presented in Fig. 1 depict the consistent acoustical structure of the sounds produced by each patient. That is, only those sounds that are consistent from sample to sample are shown, not frequencies produced by sources other than the patient. The graphs show power (in decibels along an arbitrary scale) as a function of frequency (from 2 to approximately 2500 cycles per second). As many readers may know, a pure tone would be depicted as a single vertical line in such a graph because a sinusoidal sound has all of its power at one frequency. White noise and clicks would appear as horizontal lines because these sounds contain all frequencies at equal power. Cries, as discussed below, have prominent peaks at about 400 Hz and higher multiples of that fundamental frequency [11].
Results
The results from 5 tests on 3 patients are shown in Fig. 1. These averaged power spectra from multiple samples of stridor minus any background noise clearly demonstrate that each patient had a consistent acoustical pattern to the abnormal sounds they produced when breathing.
194 125
125
o
-I0
-ID
0
I0
25
125
i
i
0
L
i
,
,
,
,
,
tO
,
25
125 2b
-10
i
i
i
0
i
I0
i
i
i
i
i
i
25
i
-I00
'
'
i
' 1~3
n
i
i
•
25
25 LEGEND
I0
0
i 0
i
i
i IO
kHz
Fig. 1. Power spectra of infantile stridors. Each row is a different patient, and each column is a different test of that patient. Patient 1 had a subglottic hemangioma; spectra l a and lb were collected 8 days apart as the condition worsened. Patient 2 had subglonic stenosis; spectra 2a and 2b were collected before and after surgery, respectively. Patient 3 had laryngomalacia.
Notice that in each graph there is a variable but low level of power at many frequencies. Imagine a broad and nearly horizontal line across the bottom of these graphs. This is the 'noise floor' and represents a combination of the background noise that was not subtracted, and errors from the acoustical, electrical and mathematical manipulations. There are, however, loud frequencies in most spectra whose powers rise above this noise floor by 12 dB or more. This means that there are frequencies produced by the infant that, on the average from sample to sample, are roughly 4 times more intense than would be expected from random background noise. Notice also that the patients appear to have distinctly different patterns to their stridors. These different patterns, as we see them, are sketched in Fig. 2. The acoustical analysis will be discussed after a brief presentation of the patients.
195
Two Sharp Peaks (Subglot tic Hernangiomo)
One Sharp Peak (Subglot tic Stenosis)
8
One Broad Peak (Laryngomolocia)
C
No Peak (Resolved Strider)
Fig. 2. Schematic representation of possible patterns in infantile strider.
Case presentations Patient 1 is a 6-month-old white female who was born prematurely to a 40-year-old mother. She spent the first few months of her life in the neonatal intensive care unit. The patient then did well at home with her feeding and breathing. At 6 months of age she was readmitted to the intensive care unit for problems with inspiratory strider with dyspnea, and the first samples of her strider were collected. Her strider was sampled again after 8 days as her condition worsened. Bronchoscopy and direct laryngoscopy revealed a subglottic hemangioma that had been suspected from a subglottic narrowing on radiographs of her neck and a hemangioma on her ankle. A tracheostomy was performed to provide an airway, and the subglottic hemangioma was resected with a laser. Patient 2 is a one-year-old black male who was admitted to the pediatric intensive care unit with a provisional diagnosis of epiglottitis versus croup. The first samples of an inspiratory strider were collected soon after admission. Examination of the lateral neck films revealed the epiglottis to be normal as did the examination of the oropharynx and hypopharynx. The patient was presumed to have croup or laryngotracheobronchitis, and was treated conservatively using humidified oxygen. When this treatment failed to resolve the prominent strider, it was decided to perform endoscopy. Direct laryngoscopy and bronchoscopy revealed subglottic stenosis with edema. It was believed that he probably had a congenitally narrowed subglottic area that was worsened by the edema from viral inflammation. A cricoid split operation was done and the patient remained intubated for an additional week. Following
196 removal of the endotracheal tube the patient did very well with minimal stridor. This was confirmed by a second set of acoustical samples. Patient 3 is an 8-month-old black male who presented severe inspiratory stridor and respiratory distress aggravated by upper respiratory infection. After cinefluoroscopy of the airway and barium swallow, the patient underwent direct laryngoscopy and bronchoscopy conforming a diagnosis of laryngomalacia. He was treated with humidified air and vaponephrine until his inflammatory process resolved. The acoustical samples of his stridor were collected during a routine office visit. The laryngomalacia was improving. The patient still had evidence of significant stridor but was gaining weight and continuing to ventilate well.
Discussion
The power spectra from the stridors of patient 1 (parts la and lb of Fig. 1 and part A of Fig. 2) have two sharp peaks. A major peak in the area of 900 Hz remained consistent from tl~.e first to the second test. The second peak changed dramatically over the 8 days between the two tests, originally 4 dB and 500 Hz lower than the main peak, but later equally intense and only 250 Hz lower than the main peak. An increase in the size of the subglottic hemangioma probably explains this easily visualized difference in stridor because the endoscopic examination revealed a lumen of less than 1 mm. The power spectrum from the stridor of patient 2 (part 2a of Fig. 1 and part B of Fig. 2) has, in contrast, a single sharp peak at about 500 Hz. At the time of the second test (part 2b of Fig. 1 and part D in Fig. 2), the stridor had resolved, and the gradually sloping noisy wave form is the spectrum of a barely audible rattle. Patient 3 was only tested once and the spectrum of his stridor shows a single peak at approximately the same frequency as the second patient: 500 Hz. In contrast to the previous patient, the powers from frequency to frequency at the region of this spectral peak in the stridor are much more variable from sample to sample. This shows up as a jagged line in the graph, depicting approximately 5 dB variability in the intensities of one frequency to its nearby neighbor. This is thought to be due to variability in each frequency that comprises the stridor. One advantage of power spectra stored on computers is that pre- and post-operative acoustical analyses can be compared. An example of such analysis is shown in Fig. 3. The post-operative spectrum of patient 2 (part 2b of Fig. 1) was subtracted from the pre-operative spectrum (part 2a of Fig. 1) after slightly more smoothing, and the difference displayed. Powers above the zero level of difference are filled in and likely depict sounds that were corrected by surgery. Such subtraction of power spectra is somewhat analogous to digital subtraction angiography, and in a larger series of patients could be a valuable source of information on changes in stridor with time and treatment. There is a continuing parallel interest in similar acoustical analyses of normal and pathological cries of neonates [6,9,11]. Stridors and cries have distinctively different spectra. Stridors are usually soft and unvoiced, like 'blowing wind' [13]. Their
197
0
1.0 kHz
2.5
Fig. 3. Pre- minus post-operativespectra from Patient 2.
spectra are generally noisy with broad and low peaks. Cries are usually loud and voiced. They have a regular harmonic structure, and their power spectrum resembles a tall picket fence whose height decreases only slightly from a fundamental frequency of about 500 Hz to over 6 kHz [9,11]. Similar data can be collected to determine which of the many sounds that neonates make are most suggestive of certain pathologies. Analyses of stridors may be more valuable in the diagnosis of some disorders, such as stenoses of the upper airway, while analyses of cries may be more valuable for another set of disorders, such as neuropathies. The absolute intensity of these sounds is not described in this analysis, only the relative powers of the various frequencies in the total sound. Because the levels of background noise were different in each patient, and there was not a constant distance from the mouth to the microphone, comparisons of the absolute levels of these sounds from averaged trace to averaged trace are probably not valid in this preliminary study. Additional controls and records are needed in future studies to enable direct comparison of the amplitudes in different spectra. Records should include the distance from the patient's mouth to the microphone, and sensitivity adjustments on all equipment. From these measurements, a computer can calculate how much the decibel scales on each spectrum need to be adjusted up or down to coincide. Another useful record might include the measurement of a calibrated noise source (such as the Bruel and Kjaer Sound Level Calibrator Type 4230) turned on in the room at the same distance from the microphone as the patient. Direct readings from the sound level meter, such as A-weighted and octave-band levels, would also be helpful in directly relating the amplitudes of different samples. A potential problem in this procedure, termed aliasing, needs to be examined in future studies. Because the data were collected as discrete digitized samples, frequencies in the signal greater than half the sampling rate (5 kHz) can appear as lower frequencies [12]. Thus, a peak at say 500 Hz in Fig. 1 could, in theory, be due to sounds at 4500 or 5500 Hz [8]. Such aliasing is not likely to be a major problem in
198 these data because previous studies [9] and our impressions indicate that these stridors have predominately low-frequency components. In any event, the important conclusion from this study is not diminished by any aliasing; different spectral patterns are found in different pathologies. More exact characterization of the signal, through analog recording followed by digital sampling at different rates, is desirable in future studies. This will determine the amplitude of high-frequency components in various stridors and thus identify an optimal sampling rate for digital analysis. Future studies of this kind might benefit from the collection of data on a portable tape recorder, followed by computer analysis at a later time. In this way, both analog and digital records of the patients' stridors would be available for further analyses. The equipment used in this study was perhaps too complicated for general use. The versatility of the digital oscilloscope and laboratory computer could be helpful in pilot studies to determine the best of many methods to collect and analyze data. Once the basic research is completed, however, clinically oriented testing could be done with dedicated equipment. Digital acoustical analysis can be done by pressing a few buttons on existing portable units. Conclusion
Preliminary data suggest that computerized (Fourier) analysis of infantile stridors can be instructive. The acoustical analysis is relatively consistent from sample to sample within the same patient. Distinct power spectra are evident from different pathologies. There are many advantages to a digital analysis of infantile stridors. The data are easy to collect. The results from each patient are stored in a form that will facilitate future statistical evaluation of patients, procedures and pathologies. Averaging of spectra can remove most of the intra-subject variability, and thus depict only those sounds that are consistent from sample to sample. Subtraction of background noises from the recorded stridors can show only those sounds that are produced by the patient; thus, meaningful data can be collected in relatively noisy places such as neonatal intensive care units. Pre- and post-operative tests can be easily compared; subtraction of spectra shows those sounds that improved or worsened. The acknowledged problems with these preliminary data can be corrected by using existing equipment and changing the computer programs. It thus appears that a larger study of the acoustical characteristics of infantile stridor is indicated. Comparisons of spectra at various times during the management of airway problems can be used to evaluate the effectiveness of various types of medical treatment including surgery and steroids. Most importantly, data from a larger series of patients may indicate what spectral patterns are characteristic of what pathologies. If consistent acoustical patterns exist, they will facilitate rapid diagnosis without invasive procedures. At the least, these tests will be valuable in alerting the surgeon to non-standard cases. It is possible that the collection and evaluation of these acoustical data could be almost completely automated with existing technology. A dedicated micro-computer
199 c o u l d g u i d e an a s s i s t a n t t h r o u g h the p r o c e d u r e s of d a t a collection, i n s u r i n g that a p p r o p r i a t e m e a s u r e m e n t s were r e c o r d e d . In a few m i n u t e s the c o m p u t e r c o u l d then p e r f o r m the F o u r i e r analysis, g r a p h the results, p r i n t the p r o b a b i l i t i e s of various d i a g n o s e s b a s e d on these data, a n d p r i n t w a r n i n g s a b o u t u n u s u a l cases.
Acknowledgements W e t h a n k the staff of the P e d i a t r i c I n t e n s i v e C a r e U n i t at H e r m a n n H o s p i t a l for their help in collecting these data. R. W o t r i n g p e r f o r m e d m a n y of the a n a l y s e s a n d p r e p a r e d the figures. D. K u b o s u m i p r e p a r e d the m a n u s c r i p t . Mr. P. G a m b l e of N i c o l e t Inc. p r o v i d e d the b u b b l e - m e m o r y unit. S u p p o r t was p r o v i d e d b y N I H ( N S 17320 a n d 20474) a n d b y the U n i v e r s i t y of T e x a s M e d i c a l School.
References 1 Bendat, J.S. and Piersol, A.G., Random Data: Analysis and Measurement Procedures, Wiley-lnterscience, New York, 1971. 2 Bergland, G.D., A guided tour of the fast Fourier transform, IEEE Spectrum, 6 (1969) 41-52. 3 Blackman, R.B. and Tukey, J.W., The Measurement of Power Spectra, Dover, New York, 1959. 4 Cotton, R. and Reilly, J.S., Stridor and airway obstruction. In C.D. Bluestone and S.E. Stool (Eds.), Pediatric Otolaryngology, Saunders, Philadelphia, PA, 1983, pp. 1190-1204. 5 Friedberg, J., Hoarseness. In C.D. Bluestone and S.E. Stool (Eds.), Pediatric Otolaryngology, Saunders, Philadelphia, PA, 1983, pp. 1181-1189. 6 Gerber, S.E., Acoustic analysis of neonates' vocalizations. Paper presented to the Society for Ear, Nose and Throat Advances in Children, 1984. 7 Harris, F.J., On the use of windows for harmonic analysis with the discrete Fourier transform, Proc. IEEE, 66 (1978) 51-83. 8 Hewlett Packard Inc., The Fundamentals of Signal Analysis, Application note 243, 1981. 9 Hirschberg, J., Acoustic analysis of pathological cries, stridors and coughing sounds in infancy, Int. J. Pediat. Otorhinolaryngol., 2 (1980) 287-300. 10 Moil, M., Kinoshita, K., Morinari, H., Shiraishi, T., Koike, S. and Murao, S., Waveform and spectral analysis of crackles, Thorax, 35 (1980) 843-850. 11 Murry, T. and Murry, J., Infant Communication: Cry and Early Speech, College-Hill Press, Houston, TX, 1980. 12 Ramirez, R.W., The FFT, Fundamentals and Concepts, Prentice-Hall, Englewood Cliffs, N J, 1985. 13 Stedman's Medical Dictionary, Williams and Wilkins, Baltimore, MD, 1982. 14 Thrane, N., The discrete Fourier transform and FFT analysers, Bruel & Kjaer Tech. Rev., 1 (1979) 3-25.
10 (1985)191-199
191
Elsevier POR 00336
Fourier analysis of infantile stridor: preliminary data Lincoln Gray 1, James C. Denneny, III 2, Hugo Carvajal 3 and Robert Jahrsdoerfer 1 t Department of Otolaryngology- Head and Neck Surgery, University of Texas Medical School, Houston, TX 77030; 2 University of lndiana, Riley's Children Hospital, Department of Otolaryngology, Indianapolis, IN 46223; and J Department of Pediatrics, P.O. Box 20708, Houston, TX 77725 (U.S.A.)
(Received November 8th, 1984) (Revised version receivedFebruary 2nd, 1985) (Accepted July 28th. 1985)
Key words: stridor-Fourier analysis-spectral analysis-computers-neonates-airway
Summary Preliminary data from 3 patients suggest that computerized (Fourier) analysis of infantile stridor can be instructive. Several recordings are quickly collected with a microphone and digital oscilloscope at the patient's bedside. The data are later sent to a computer for spectral analysis. Averaging of several spectra from each patient depicts only those sounds that are consistent from sample to sample. Subtraction of background noise from the averaged stridor removes all but those sounds that are produced by the patient. The results show that the spectra are relatively consistent from sample to sample within the same patient, and that different patients with different pathologies have distinct patterns in their spectra. It thus appears that further acoustical studies of infantile stridor will be productive. We are optimistic that data from a larger series of patients will indicate those spectral patterns which are characteristic of a specific laryngotracheal pathology, and perhaps facilitate rapid diagnosis without invasive procedures. An additional potential of this analysis is that pre- and post-operative spectra can be subtracted to show those sounds that improved or worsened. Such serial comparisons during the management of airway problems could be useful in evaluating treatment.
Correspondence: L. Gray, Department of Otolaryngology-Headand Neck Surgery. University of Texas
Medical School, Houston, TX 77030, U.S.A. 0165-5876/85/$03.30 © 1985 ElsevierScience Publishers B.V. (Biomedical Division)
192 Introduction
It is important for the practicing Otolaryngologist to correctly manage pediatric airway problems. Frequently, these patients present with noticeable stridor. The term stridor comes from Latin meaning a harsh and creaking sound [13]. It is noisy respiration produced by turbulent airflow through a partial obstruction [4]. Clinical evaluation of these abnormal breathing sounds gives valuable clues about the underlying pathology and hence to the proper course of treatment. High-pitched stridor, for example, may suggest the presence of a glottic or subglottic lesion [5]. Stridor with a zone of intensity around 2 kHz may be indicative of bronchial pathology [9]. Cotton and Reilly have stated that 'in cases of stridor, the most important part of the examination is auscultation' [4]. The clinical art of listening to these abnormal sounds has been enhanced by recent technological advances. Sonographic techniques have permitted some acoustical analysis of infants' cries and breathing sounds [9,11]. This acoustical analysis revealed 4 basic categories of stridor: pharyngeal, laryngeal, subglottic and bronchial. Sonograms provide a visual representation of the stridorous sounds, but the information is not stored on a computer and thus difficult to use in statistical analyses. A completely digital acoustical analysis, such as the Fast Fourier transform (FFT), might provide as much useful information as the analog acoustical analysis (sonograms) yet be in a form where further statistical evaluation of the stridors will be easy. Fourier analysis has been used recently in studies of cries [6,11] and sounds generated within infected lungs [10]. Cotton and Reilly hope 'that with the help of computers, data could be stored and compared to assist physicians in accurate clinical impressions and as an adjunct to endoscopic evaluation. When following chronic respiratory conditions, such data could be used for monitoring possible improvement or regression' [4]. Thus, the purpose of this study was to provide some preliminary evaluation of data collected from a computerized acoustical analysis of infantile stridor. Hopefully, these procedures can then be used in a more extensive series of patients to objectively evaluate any correlation among the acoustical characteristics of stridor with clinical diagnosis, prognosis and recovery.
Methods
Data were collected at the patient's bedside with a sound level meter (Bruel and Kjaer model 2215) and a digital oscilloscope (Nicolet model 3091). The sound level meter was used as a high-quality calibrated microphone and amplifier system. The microphone was held 2-10 cm from the patient's mouth. The analog output from the sound level meter was displayed and then stored by the digital oscilloscope positioned close to the patient's bed. The sensitivities on both the sound level meter and the oscilloscope were adjusted to produce a signal that usually filled the oscilloscope screen but did not exceed its limits. The oscilloscope is capable of storing up to ten 2000-point samples on a bubble-memory cassette. The data were digitized at a rate
193 of one 12-bit point every 200 /~s (5 kHz). Each sample thus represents 0.4 s of recorded stridor. This allows a Fourier analysis of frequencies between approximately 2 and 2500 cycles per second. Usually, 8 different samples of the stridor were collected, and then two samples of the background noise were collected for use in the subsequent computerized analysis. Collection of these data required about 10 min at the patient's bedside, but more than half of this time was required to set up the equipment. After collection of these data in the hospital the oscilloscope was carried to a laboratory where the 10 stored samples were retrieved from the bubble memory and transmitted to a computer (LSI 11/03). Twenty-four zeros were added to the front and back of the data, to produce a power-of-two number of samples, and then the data were multiplied by a Hanning window to remove the unwanted effects of spectral spread or 'leakage' produced by discrete Fourier analysis [7]. The data were then subjected to a fast Fourier analysis [2,3,14] and the resulting power spectrum was stored in the computer. To produce the graphs shown in Fig. 1, the power spectra from individual samples of the patient's stridor were averaged. Averaging power spectra from different samples of the same sound depicts the acoustical characteristics that are consistent from sample to sample [1]. From this average of sampled stridors, the average of the power spectra from the background noise was subtracted. Subtracting the power spectra of background noises eliminates the contribution of extraneous sounds, and the powers that remain show only those portions of the recorded sounds that were produced by the infant. The spectra displayed are smoothed slightly, by computing a running average of each 10 adjacent spectral lines. This smoothing removes some of the variation in the curves that tends to obscure visual evaluation of general patterns. In summary, it is believed that the power spectra presented in Fig. 1 depict the consistent acoustical structure of the sounds produced by each patient. That is, only those sounds that are consistent from sample to sample are shown, not frequencies produced by sources other than the patient. The graphs show power (in decibels along an arbitrary scale) as a function of frequency (from 2 to approximately 2500 cycles per second). As many readers may know, a pure tone would be depicted as a single vertical line in such a graph because a sinusoidal sound has all of its power at one frequency. White noise and clicks would appear as horizontal lines because these sounds contain all frequencies at equal power. Cries, as discussed below, have prominent peaks at about 400 Hz and higher multiples of that fundamental frequency [11].
Results
The results from 5 tests on 3 patients are shown in Fig. 1. These averaged power spectra from multiple samples of stridor minus any background noise clearly demonstrate that each patient had a consistent acoustical pattern to the abnormal sounds they produced when breathing.
194 125
125
o
-I0
-ID
0
I0
25
125
i
i
0
L
i
,
,
,
,
,
tO
,
25
125 2b
-10
i
i
i
0
i
I0
i
i
i
i
i
i
25
i
-I00
'
'
i
' 1~3
n
i
i
•
25
25 LEGEND
I0
0
i 0
i
i
i IO
kHz
Fig. 1. Power spectra of infantile stridors. Each row is a different patient, and each column is a different test of that patient. Patient 1 had a subglottic hemangioma; spectra l a and lb were collected 8 days apart as the condition worsened. Patient 2 had subglonic stenosis; spectra 2a and 2b were collected before and after surgery, respectively. Patient 3 had laryngomalacia.
Notice that in each graph there is a variable but low level of power at many frequencies. Imagine a broad and nearly horizontal line across the bottom of these graphs. This is the 'noise floor' and represents a combination of the background noise that was not subtracted, and errors from the acoustical, electrical and mathematical manipulations. There are, however, loud frequencies in most spectra whose powers rise above this noise floor by 12 dB or more. This means that there are frequencies produced by the infant that, on the average from sample to sample, are roughly 4 times more intense than would be expected from random background noise. Notice also that the patients appear to have distinctly different patterns to their stridors. These different patterns, as we see them, are sketched in Fig. 2. The acoustical analysis will be discussed after a brief presentation of the patients.
195
Two Sharp Peaks (Subglot tic Hernangiomo)
One Sharp Peak (Subglot tic Stenosis)
8
One Broad Peak (Laryngomolocia)
C
No Peak (Resolved Strider)
Fig. 2. Schematic representation of possible patterns in infantile strider.
Case presentations Patient 1 is a 6-month-old white female who was born prematurely to a 40-year-old mother. She spent the first few months of her life in the neonatal intensive care unit. The patient then did well at home with her feeding and breathing. At 6 months of age she was readmitted to the intensive care unit for problems with inspiratory strider with dyspnea, and the first samples of her strider were collected. Her strider was sampled again after 8 days as her condition worsened. Bronchoscopy and direct laryngoscopy revealed a subglottic hemangioma that had been suspected from a subglottic narrowing on radiographs of her neck and a hemangioma on her ankle. A tracheostomy was performed to provide an airway, and the subglottic hemangioma was resected with a laser. Patient 2 is a one-year-old black male who was admitted to the pediatric intensive care unit with a provisional diagnosis of epiglottitis versus croup. The first samples of an inspiratory strider were collected soon after admission. Examination of the lateral neck films revealed the epiglottis to be normal as did the examination of the oropharynx and hypopharynx. The patient was presumed to have croup or laryngotracheobronchitis, and was treated conservatively using humidified oxygen. When this treatment failed to resolve the prominent strider, it was decided to perform endoscopy. Direct laryngoscopy and bronchoscopy revealed subglottic stenosis with edema. It was believed that he probably had a congenitally narrowed subglottic area that was worsened by the edema from viral inflammation. A cricoid split operation was done and the patient remained intubated for an additional week. Following
196 removal of the endotracheal tube the patient did very well with minimal stridor. This was confirmed by a second set of acoustical samples. Patient 3 is an 8-month-old black male who presented severe inspiratory stridor and respiratory distress aggravated by upper respiratory infection. After cinefluoroscopy of the airway and barium swallow, the patient underwent direct laryngoscopy and bronchoscopy conforming a diagnosis of laryngomalacia. He was treated with humidified air and vaponephrine until his inflammatory process resolved. The acoustical samples of his stridor were collected during a routine office visit. The laryngomalacia was improving. The patient still had evidence of significant stridor but was gaining weight and continuing to ventilate well.
Discussion
The power spectra from the stridors of patient 1 (parts la and lb of Fig. 1 and part A of Fig. 2) have two sharp peaks. A major peak in the area of 900 Hz remained consistent from tl~.e first to the second test. The second peak changed dramatically over the 8 days between the two tests, originally 4 dB and 500 Hz lower than the main peak, but later equally intense and only 250 Hz lower than the main peak. An increase in the size of the subglottic hemangioma probably explains this easily visualized difference in stridor because the endoscopic examination revealed a lumen of less than 1 mm. The power spectrum from the stridor of patient 2 (part 2a of Fig. 1 and part B of Fig. 2) has, in contrast, a single sharp peak at about 500 Hz. At the time of the second test (part 2b of Fig. 1 and part D in Fig. 2), the stridor had resolved, and the gradually sloping noisy wave form is the spectrum of a barely audible rattle. Patient 3 was only tested once and the spectrum of his stridor shows a single peak at approximately the same frequency as the second patient: 500 Hz. In contrast to the previous patient, the powers from frequency to frequency at the region of this spectral peak in the stridor are much more variable from sample to sample. This shows up as a jagged line in the graph, depicting approximately 5 dB variability in the intensities of one frequency to its nearby neighbor. This is thought to be due to variability in each frequency that comprises the stridor. One advantage of power spectra stored on computers is that pre- and post-operative acoustical analyses can be compared. An example of such analysis is shown in Fig. 3. The post-operative spectrum of patient 2 (part 2b of Fig. 1) was subtracted from the pre-operative spectrum (part 2a of Fig. 1) after slightly more smoothing, and the difference displayed. Powers above the zero level of difference are filled in and likely depict sounds that were corrected by surgery. Such subtraction of power spectra is somewhat analogous to digital subtraction angiography, and in a larger series of patients could be a valuable source of information on changes in stridor with time and treatment. There is a continuing parallel interest in similar acoustical analyses of normal and pathological cries of neonates [6,9,11]. Stridors and cries have distinctively different spectra. Stridors are usually soft and unvoiced, like 'blowing wind' [13]. Their
197
0
1.0 kHz
2.5
Fig. 3. Pre- minus post-operativespectra from Patient 2.
spectra are generally noisy with broad and low peaks. Cries are usually loud and voiced. They have a regular harmonic structure, and their power spectrum resembles a tall picket fence whose height decreases only slightly from a fundamental frequency of about 500 Hz to over 6 kHz [9,11]. Similar data can be collected to determine which of the many sounds that neonates make are most suggestive of certain pathologies. Analyses of stridors may be more valuable in the diagnosis of some disorders, such as stenoses of the upper airway, while analyses of cries may be more valuable for another set of disorders, such as neuropathies. The absolute intensity of these sounds is not described in this analysis, only the relative powers of the various frequencies in the total sound. Because the levels of background noise were different in each patient, and there was not a constant distance from the mouth to the microphone, comparisons of the absolute levels of these sounds from averaged trace to averaged trace are probably not valid in this preliminary study. Additional controls and records are needed in future studies to enable direct comparison of the amplitudes in different spectra. Records should include the distance from the patient's mouth to the microphone, and sensitivity adjustments on all equipment. From these measurements, a computer can calculate how much the decibel scales on each spectrum need to be adjusted up or down to coincide. Another useful record might include the measurement of a calibrated noise source (such as the Bruel and Kjaer Sound Level Calibrator Type 4230) turned on in the room at the same distance from the microphone as the patient. Direct readings from the sound level meter, such as A-weighted and octave-band levels, would also be helpful in directly relating the amplitudes of different samples. A potential problem in this procedure, termed aliasing, needs to be examined in future studies. Because the data were collected as discrete digitized samples, frequencies in the signal greater than half the sampling rate (5 kHz) can appear as lower frequencies [12]. Thus, a peak at say 500 Hz in Fig. 1 could, in theory, be due to sounds at 4500 or 5500 Hz [8]. Such aliasing is not likely to be a major problem in
198 these data because previous studies [9] and our impressions indicate that these stridors have predominately low-frequency components. In any event, the important conclusion from this study is not diminished by any aliasing; different spectral patterns are found in different pathologies. More exact characterization of the signal, through analog recording followed by digital sampling at different rates, is desirable in future studies. This will determine the amplitude of high-frequency components in various stridors and thus identify an optimal sampling rate for digital analysis. Future studies of this kind might benefit from the collection of data on a portable tape recorder, followed by computer analysis at a later time. In this way, both analog and digital records of the patients' stridors would be available for further analyses. The equipment used in this study was perhaps too complicated for general use. The versatility of the digital oscilloscope and laboratory computer could be helpful in pilot studies to determine the best of many methods to collect and analyze data. Once the basic research is completed, however, clinically oriented testing could be done with dedicated equipment. Digital acoustical analysis can be done by pressing a few buttons on existing portable units. Conclusion
Preliminary data suggest that computerized (Fourier) analysis of infantile stridors can be instructive. The acoustical analysis is relatively consistent from sample to sample within the same patient. Distinct power spectra are evident from different pathologies. There are many advantages to a digital analysis of infantile stridors. The data are easy to collect. The results from each patient are stored in a form that will facilitate future statistical evaluation of patients, procedures and pathologies. Averaging of spectra can remove most of the intra-subject variability, and thus depict only those sounds that are consistent from sample to sample. Subtraction of background noises from the recorded stridors can show only those sounds that are produced by the patient; thus, meaningful data can be collected in relatively noisy places such as neonatal intensive care units. Pre- and post-operative tests can be easily compared; subtraction of spectra shows those sounds that improved or worsened. The acknowledged problems with these preliminary data can be corrected by using existing equipment and changing the computer programs. It thus appears that a larger study of the acoustical characteristics of infantile stridor is indicated. Comparisons of spectra at various times during the management of airway problems can be used to evaluate the effectiveness of various types of medical treatment including surgery and steroids. Most importantly, data from a larger series of patients may indicate what spectral patterns are characteristic of what pathologies. If consistent acoustical patterns exist, they will facilitate rapid diagnosis without invasive procedures. At the least, these tests will be valuable in alerting the surgeon to non-standard cases. It is possible that the collection and evaluation of these acoustical data could be almost completely automated with existing technology. A dedicated micro-computer
199 c o u l d g u i d e an a s s i s t a n t t h r o u g h the p r o c e d u r e s of d a t a collection, i n s u r i n g that a p p r o p r i a t e m e a s u r e m e n t s were r e c o r d e d . In a few m i n u t e s the c o m p u t e r c o u l d then p e r f o r m the F o u r i e r analysis, g r a p h the results, p r i n t the p r o b a b i l i t i e s of various d i a g n o s e s b a s e d on these data, a n d p r i n t w a r n i n g s a b o u t u n u s u a l cases.
Acknowledgements W e t h a n k the staff of the P e d i a t r i c I n t e n s i v e C a r e U n i t at H e r m a n n H o s p i t a l for their help in collecting these data. R. W o t r i n g p e r f o r m e d m a n y of the a n a l y s e s a n d p r e p a r e d the figures. D. K u b o s u m i p r e p a r e d the m a n u s c r i p t . Mr. P. G a m b l e of N i c o l e t Inc. p r o v i d e d the b u b b l e - m e m o r y unit. S u p p o r t was p r o v i d e d b y N I H ( N S 17320 a n d 20474) a n d b y the U n i v e r s i t y of T e x a s M e d i c a l School.
References 1 Bendat, J.S. and Piersol, A.G., Random Data: Analysis and Measurement Procedures, Wiley-lnterscience, New York, 1971. 2 Bergland, G.D., A guided tour of the fast Fourier transform, IEEE Spectrum, 6 (1969) 41-52. 3 Blackman, R.B. and Tukey, J.W., The Measurement of Power Spectra, Dover, New York, 1959. 4 Cotton, R. and Reilly, J.S., Stridor and airway obstruction. In C.D. Bluestone and S.E. Stool (Eds.), Pediatric Otolaryngology, Saunders, Philadelphia, PA, 1983, pp. 1190-1204. 5 Friedberg, J., Hoarseness. In C.D. Bluestone and S.E. Stool (Eds.), Pediatric Otolaryngology, Saunders, Philadelphia, PA, 1983, pp. 1181-1189. 6 Gerber, S.E., Acoustic analysis of neonates' vocalizations. Paper presented to the Society for Ear, Nose and Throat Advances in Children, 1984. 7 Harris, F.J., On the use of windows for harmonic analysis with the discrete Fourier transform, Proc. IEEE, 66 (1978) 51-83. 8 Hewlett Packard Inc., The Fundamentals of Signal Analysis, Application note 243, 1981. 9 Hirschberg, J., Acoustic analysis of pathological cries, stridors and coughing sounds in infancy, Int. J. Pediat. Otorhinolaryngol., 2 (1980) 287-300. 10 Moil, M., Kinoshita, K., Morinari, H., Shiraishi, T., Koike, S. and Murao, S., Waveform and spectral analysis of crackles, Thorax, 35 (1980) 843-850. 11 Murry, T. and Murry, J., Infant Communication: Cry and Early Speech, College-Hill Press, Houston, TX, 1980. 12 Ramirez, R.W., The FFT, Fundamentals and Concepts, Prentice-Hall, Englewood Cliffs, N J, 1985. 13 Stedman's Medical Dictionary, Williams and Wilkins, Baltimore, MD, 1982. 14 Thrane, N., The discrete Fourier transform and FFT analysers, Bruel & Kjaer Tech. Rev., 1 (1979) 3-25.