INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

PRECISE MONITORING OF VENTILATION is usually achieved by having a subject breathe through a mouthpiece or face mask attached to a pneumotachygraph or spirometer. Although these devices permit the accurate measurement of ventilation and its parameters, they also alter the pattern of breathing and minute ventilation (I) (2, 12, 13, 37). They are not useful for monitoring ventilation in any circumstance in which keeping a mouthpiece and nose clip in place is too difficult or impossible.

Magnetometry and inductance plethysmography record the movements of the rib cage and abdomen during respiration and, by the use of a suitable calibration procedure, convert the summed thoracic and abdominal motions into a volume signal (30, 34, 35). Although these devices eliminate the need for a mouthpiece and nose clip and, as a result, permit measurement of the volumes, flows, and timing of normal breathing, they are unfortunately sensitive to changes in posture, particularly xyphi-pubic distance (24, 34, 35, 38, 39). Even when the latter is controlled, the rib cage and abdomen do not always act as compartments with a single degree of freedom (7, 14, 21, 36), and this introduces a further source of error. Motion analysis by optoelectronic plethysmography works extremely well (5) but is quite expensive and constrains the subject because markers on the trunk must be visualized by fixed videocameras. There is a need for other devices that can measure ventilation cheaply and accurately without a mouthpiece and nose clip.

Phonospirometry, the estimation of ventilation parameters from measurements of tracheal breath sounds, provides a simple alternative to motion analysis systems and may prove to be more versatile. Since the invention of the stethoscope by René Laennec in 1819, auscultation has provided the clinician with a quick, but crude, assessment of pulmonary ventilation. Objective measurements of breath sounds were first made more than 25 yr ago (17) and have been used as a qualitative assessment of ventilation during sleep (8); but it is only with advances in computer technology and the wide application of digital signal analysis that recording and analysis of respiratory sounds has accelerated.

Normal breath sounds are primarily generated by turbulence within large- and medium-sized airways. Airflow velocity and airway dimensions influence the generation of turbulent flow, whereas the intensity of the detected sound is influenced by the sound transmission characteristics of the tissue between the regions of sound generation and the point where measurements are made. As a result, for a given subject and microphone position, sound amplitude is proportional to the flow rate (10, 15, 17, 32, 33), suggesting the possibility of deriving flow estimates from sound amplitude (33). In this paper, we present a new method to measure ventilation by converting tracheal breath sound intensity to ventilatory flow. By integration, tidal volume (VT), respiratory frequency (f), I, and other ventilatory parameters are obtained.

Many investigators have measured airflow by acoustical techniques (1, 6, 10, 18, 19, 27, 28). However, their interest has lain primarily in the derivation of parameters related to the frequency spectrum of the sound signal, or the mechanism of sound production, rather than in the use of the sound signal as a surrogate for flow. With the exception of Soufflet et al. (33), the studies quantifying the relationship between sound intensity and flow (10, 32) have focused exclusively on flows of >0.5 l/s.

Our purpose was to estimate flow from breath sound intensity during quiet breathing and, therefore, almost exclusively on flows of <0.5 l/s. This presented particular problems of an adverse signal-to-noise ratio (SNR) and the threshold of flow below which flow-derived sound cannot be detected. These problems have not previously been satisfactorily addressed. We have solved these problems by measuring the threshold of detection of flow-derived sounds (~0.3 l/s) and interpolating from the threshold to zero-flow points. Thus we were able to measure VT, I, f, inspiratory (TI) and expiratory (TE) durations, respiratory cycle time (Ttot), duty cycle (TI/Ttot), and mean inspiratory flow (VT/TI) during quiet breathing from the envelope of the sound signal with only a microphone placed on the skin over the thyroid cartilage in the neck, without the use of a mouthpiece or nose clip except for a calibration procedure. In this paper we describe this technology and present preliminary results.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects. A total of 11 nonsmoking normal volunteers with no history of pulmonary disease or recent respiratory tract infection, 3 asymptomatic mild asthmatic patients, and 3 patients with unstable airway obstruction attending the day hospital of the Montreal Chest Institute were recruited for the study. Of these, seven normal subjects and one asymptomatic asthmatic subject participated in the early studies in which we validated phonospirometry as a method to measure ventilation, estimated the errors by a Bland-Altman analysis, and tabulated the ventilatory pattern shown in Table 1. SNR and the power spectrum of normal breath sounds were measured in one of these subjects and in four additional normal volunteers. Effects of posture were studied in 5 of the 11 normal subjects. Subsequently five additional patients, two with mild asthma and three with unstable airways obstruction, attending the day hospital of the Montreal Chest Institute, were studied to measure their ventilatory pattern and its variations. These patients were studied primarily to get a feeling for how applicable the technique is to disease rather than to make conclusions as to the effect of disease on breathing pattern.

View this table: [in this window] [in a new window]   Table 1.   Effect of head position on error of VT estimation (± 1 SD)

Calibration procedure. Airflow at the mouth and tracheal breath sounds were measured simultaneously during two separate 30-s intervals. The setup is illustrated in Fig. 1. Subjects were seated, wore a nose clip, and breathed quietly on a mouthpiece. An electret microphone (model 1306, Armaco, Vancouver, Canada), coupled to a custom-designed preamplifier with a 30-fold amplification, was placed over the trachea at the level of the thyroid cartilage, a region that had previously been determined to provide the best sound signal. The microphone was inserted in a conical aluminium housing 5.5 mm in diameter around the microphone and 13 mm in diameter at the opening that was placed on the skin. The distance between the microphone and the opening was 3.9 mm. Frequency response of the electret microphone was flat (±3 dB) between 20 Hz and 8 kHz. Flow signal was obtained by using a pneumotachygraph (model no. 1A, Fleisch, Lausanne, Switzerland) and a piezoelectric pressure transducer. Both flow and sound were amplified and filtered before analog-to-digital conversion. For flow, the cutoff was set at 50 Hz; for the sound signal, cutoff was at 1,000 Hz. Both signals were sampled at 3,000 Hz by using a commercially available software package (ORIGIN, MICROCAL, Northampton, MA) and a 12-bit analog-to-digital converter (model DT-2801, Data Translation, Marlboro, MA).

View larger version (27K): [in this window] [in a new window]   Fig. 1.   Diagram of data acquisition system. Tracheal breath sounds are measured with an electret microphone (Mic), and flow is measured with a pneumotach and pressure transducer. Both signals are amplified and low-pass filtered (flow, 50 Hz; sound, 1 KHz) before digitization and storage.

Effect of neck rotation and body posture on volume measurements. Because head movements and changes of posture are to be expected during long-term measurements of ventilation, we examined the correspondence between Vm and Ve signals during systematic changes of head and body position. First, in five normal subjects and one asymptomatic asthmatic subject (SK), flow and sound signals were obtained during up-and-down and right-to-left movements of the head. Subjects were asked to breathe quietly for 10 s while facing forward, for 20 s while facing left and right, respectively, and then for a final 10 s facing forward for a second time. Second, maximum up-and-down movements of the head were studied in a similar fashion, i.e., a subject breathed quietly for 10 s with the head positioned centrally, for 20 s each with the neck maximally extended then flexed, then finally for 10 s facing foward. At a f of 15 breaths/min, 20 s would be sufficient to record about five breaths; in fact we found that f was closer to 20 breaths/min so that with rotation, extension, and flexion we measured between six and seven breaths.

Data analysis. Figure 2 is an example of the data obtained from one subject during a 30-s calibration period. Only a part of the record is illustrated. Shown from above downward are flow, corresponding sound signal, filtered sound signal, and sound envelope. During a single breath, two major sound bursts are observed (Fig. 2B), one during inspiration, the other during expiration. Sounds in phase with the heartbeat are seen as sharp transients. The digitized sound signal was band-pass filtered between 200 and 1,000 Hz to remove heart and muscle sounds, which are typically <200 Hz, and the high-frequency noise, which is >1,000 Hz. Specifically, the digitized data was transformed to the frequency domain by using a discrete Fourier transform (DFT). Then, all negative frequencies 1,000 Hz, all positive frequencies >1,000 Hz, and frequencies between 200 and 200 Hz were set to zero by using a three-point roll-off function. To provide a better attenuation in the rejection bands of the DFT-based band-pass filter, the three first frequency coefficients at the upper and lower boundaries of the pass bands were weighted by the following coefficients: 0.022, 0.23, 0.70 (11). After band-pass filtering (Fig. 2C), there was a marked attenuation of both the background noise and the cardiac artifact, whereas the breath sound signal was well preserved.

View larger version (36K): [in this window] [in a new window]   Fig. 2.   A segment of quiet breathing from 1 subject. A: flow measured by the pneumotachygraph. B: tracheal sound signal. C: sound signal after band-pass (200-1,000 Hz) filtering. D: sound envelope obtained after filtering and Hilbert transform.

View larger version (31K): [in this window] [in a new window]   Fig. 3.   Relationship between flow and sound amplitude during quiet breathing in 8 subjects (identified by 2-letter initials). Note that, for a given flow, there is a wide between-subject variation in sound amplitude. In addition, at low flows, the sound signal does not exceed the background noise. Transfer function for conversion of sound to flow was calculated for inspiration and expiration separately over the quasilinear portion of the relationship.

View larger version (21K): [in this window] [in a new window]   Fig. 4.   Comparison of measured (broken line) and calculated (solid line) flows from the data segment in Fig. 2. The thinner solid line is the interpolated flow. Note the good correspondence between the estimated and the actual flow signals during inspiration (positive deflections) and the early part of expiration (negative deflections).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The relationship between sound intensity and flow in seven normal subjects and an asthmatic subject (SK) is illustrated in Fig. 3. The general shape is similar for all subjects, although there were variations in sound intensity at a given flow from subject to subject as well as in the flow value above which tracheal sounds could be detected. In Fig. 4, flow derived from the sound signal (solid lines) is compared with the measured flow signal (dotted line). The thick segments represent the calculated flows, and the thin solid lines represent the interpolated values. During inspiration, calculated flow is quite close to the measured value and even reflects some of its details. Expiratory flows are less well estimated from the sound signal and, in this case, underestimate the true value. Similarly, the interpolated values are more accurate during the entire inspiration interval and the beginning of expiration, but the concave shape of the actual flow curve during the latter part of expiration is poorly estimated. This error was greatest when subjects had a long end-expiratory pause.

To determine more precisely how sound intensity changed with flow, in five of the normal subjects shown in Fig. 3 and in SK, we compared the power spectrum of the sound signal obtained during a 15-s breath hold to the spectra obtained during constant flow inspirations between 0.07 and 0.3 l/s. Data from a representative subject is illustrated in Fig. 5A. In this subject, inspiratory sounds were detectable above the background noise at 0.15 l/s and progressively increased in amplitude as flow was increased. Similar results were seen in the other five subjects, but the threshold for inspiratory sound detection differed from individual to individual. SNR was calculated by dividing, at every frequency, the sound amplitude during inspiration by the sound amplitude during the breath hold and then averaging the values for all frequencies between 200 and 1,000 Hz. These values, plotted as a function of the mean flow rate in Fig. 5B, show the intersubject variation in the threshold flow rate above which sound could be detected above background. The threshold value was generally specific to each subject. Above it was a quasi-linear relationship between SNR and flow in all but one subject.

View larger version (25K): [in this window] [in a new window]   Fig. 5.   Signal-to-noise ratio (SNR) of the sound signal. A: power spectrum of sound signals in a single subject during a breath hold (dotted line) and during a series of inspiratory flow rates. B: SNR in 6 subjects as a function of inspiratory flow. Note that flow had to exceed a threshold value before tracheal breath sounds could be detected. mean AR, mean amplitude ratio.

Measured (solid lines) and calculated (dotted lines) flows for the eight subjects in Fig. 3 are shown in Fig. 6. In subjects CL, QZ, SK, QL, EB, and SW, the two flow signals were very close during both inspiration and expiration, whereas in the remaining subjects, there were significant differences between inspiration and expiration. These differences were greatest during expiration, and, in all subjects, the two signals were very similar during inspiration.

View larger version (39K): [in this window] [in a new window]   Fig. 6.   Comparison of measured (solid lines) and estimated (dotted lines) flow signals in all subjects (identified by initials). Note that, in general, inspiratory flow is better estimated than expiratory flow.

Validation of transfer functions. Because we were more interested in volume than in flow, we used the error in volume estimation as our criterion of accuracy for the technique. The relationship between Ve and Vm is shown in Fig. 7A. The solid line represents the line of identity. All data points fall near the identity line, which indicates a good correspondence between Ve and Vm. The breath-by-breath error (Ve  Vm) in all subjects as a function of the average of Ve and Vm is illustrated in Fig. 7B. The mean difference between Ve and Vm (bias) was 0.009 ± 0.046 liters and was independent of VT. Only two subjects, JS and SW, had values that spanned the entire confidence interval; the other subjects' values were more tightly grouped, although some of them had a positive (QL) or negative (SW) bias.

View larger version (17K): [in this window] [in a new window]   Fig. 7.   Comparison of estimated and measured tidal volumes (VT) during quiet breathing in 8 subjects. Each symbol represents data from a single subject. A: solid line represents the identity relationship; note that all data points fall close to the identity line, indicating a good correspondence between measured and estimated VT measurements. B: plot of the difference in volume measurement vs. means of the 2 measurements. Mean (±SD) difference was 0.009 ± 0.046 liter.

Effect of head movements on volume estimation. In Fig. 8, the flow signal derived from tracheal breath sound amplitude (dotted line) is compared with measured flow (solid line) during head movements in a single subject. Figure 8A shows data obtained when the head was moved from side to side. Figure 8B illustrates up-and-down movements of the head. In this subject, the correspondence between the measured and estimated signal was good no matter what the position of the head was. The breaths during the transition from one position to another tended to be larger than the preceding and following breaths and could often be readily identified; nevertheless, they were reasonably well described by the estimated flow signal. Bland-Altman analysis revealed a slight positive bias (+0.015 ± 0.071 liters) with neck flexion and extension and a slight negative bias (0.019 ± 0.061 liters) with rotation. No bias was introduced by VT size. Standard deviation (SD) was somewhat larger than that measured (0.046 liters) with the subject facing forward. The percent error of the VT estimates are shown in Table 1. They show that, with one exception, the error in VT estimation was 16% with neck rotation. The error was not increased by full neck flexion. With full neck extension, the error was considerably greater.

View larger version (40K): [in this window] [in a new window]   Fig. 8.   Estimated (dotted lines) and measured (solid lines) flows in different head positions in a single subject. A: head movement from left to right. B: up-and-down head movement. Horizontal lines indicate breaths in a given position. Note the good correspondence between estimated and measured flows in this subject even during the transition between different positions.

Effect of body posture on volume estimation. Figure 9 shows the effect of posture on the estimation of volume. Volumes obtained from the sound signal were similar to those obtained from the pneumotach signal when subjects were standing, and there was no change in the SD of the difference between the two values (0.06 liters sitting; 0.07 liters standing). The bias increased from 0.005 (sitting) to 0.009 liter (standing), but this change was small. The biggest disparity between the estimated and the measured VT was seen in the supine position. In the identity plot (Fig. 9C, left) the majority of the data points fell above the line of identity, indicating an overestimation of VT by the sound signal. The bias in the Bland-Altman plot (Fig. 9C, right) was 0.123 liter compared with 0.005 liter when the subjects were seated. In addition, the SD of the difference between the two measurements increased to 0.113 liter. In Table 2, the percent error of VT estimates in the three positions is given; these were large in the supine position.

View larger version (33K): [in this window] [in a new window]   Fig. 9.   Estimated vs. measured VT in different body positions. The calibration equation was obtained from the flow-sound relationship in the seated position. A: values obtained in seated position. B: values obtained in standing position. C: values obtained in supine position. Note that the difference between estimated and measured volumes was not increased by standing but that there was a systematic overestimation of VT in the supine position. Mean differences in VT were 0.005 ± 0.061 liter (seated), 0.009 ± 0.069 liter (standing), and 0.123 ± 0.113 liter (supine).

Reproducibility. Because measurements in the forward-facing, seated position were repeated after neck movement and postural changes in the six subjects, reproducibility of the estimation of VT could be estimated. The two volume signals remained comparable despite the intervening movement on the part of the patient. The mean difference between the two measurement techniques was 0.001 liter, and the SD of the difference was 0.06 liter.

Ventilatory pattern. Table 3 gives values of VT, f, I, TI/Ttot, and VT/TI in the normal subjects and the patients. It is noteworthy that normal f averaged 19.3 breaths/min, VT averaged only 0.37 liter, and VT/TI averaged only 0.31 l/s. I was >2 SD greater than normal in the patients with unstable airways obstruction and in one of the asthmatic subjects. An increased VT/TI accounted for the increased I in these patients. One asthmatic patient had an unusually low f and a high VT.

View larger version (25K): [in this window] [in a new window]   Fig. 10.   Log-normal (ln) probability density functions (PDF) of minute ventilaton (I; A) and VT (B) in normal subjects (dashed lines) and patients (solid lines).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Derivation of flow from sound. Continuous noninvasive measurement of flow and volume has both clinical and research applications. At present, the most widely used noninvasive technique for respiratory measurements is respiratory inductive plethysmography. However, with this technique, measured VT is affected by changes in posture and/or the shape of the chest wall (24, 35, 38, 39), limiting its usefulness in long-term monitoring. For this purpose, other noninvasive methods need to be developed. In this study, we have explored the feasibility of deriving ventilatory flow from tracheal sound signals and using this signal to estimate VT.

Sources of error. When we compared VT derived by integration of measured flow with the volume obtained by integration of estimated flow, in subjects in the seated position, we found a good correlation between the two values (Fig. 7A). The mean difference between Ve and Vm (0.001 liter) was insignificant, but individual breaths could vary by up to 0.1 liter (Fig. 7B). However, the error spanned the entire confidence interval in only two subjects (EB and JS). In subject SW, Ve was consistently higher than Vm, and, in JS, Ve was lower than Vm. A bias such as this can be characterized and accounted for when making measurements in an individual subject. As a result, the percent error in estimating VT was <15% in all subjects when they were seated and facing forward (Table 1).

Effect of head movement. The estimated flow signal was not greatly affected by changing the head position apart from the breaths during the transition period (Fig. 8). Because these breaths tend to have higher than average VT/TI, they are easily distinguished. For our analyses, we eliminated one breath at the transition; nevertheless, it is interesting to note that both the actual and the estimated flow signals were similarly altered, suggesting that inclusion of these breaths in the data analysis will not increase the error of our estimation of VT.

Effect of posture. Moving from the seated to the standing position had little effect on VT estimation (Fig. 9; Table 2). This could be considered an advantage over inductance plethysmography, which changes calibration with posture and changes in xiphipubic distance. We have not systematically measured the effect of thoracoabdominal configuration on the relationship between sound and flow in the trachea but see no a priori reason why it should change. Almost certainly, there was a significant change in xiphipubic distance between sitting and standing postures in at least some of our subjects. They were not instructed on how to sit during the recording period and merely chose a position that was comfortable. We subsequently measured changes in xiphipubic distance in normal subjects between seated and standing posture while maintaining a rigid upright spinal configuration and found changes between 5 and 8 cm.

Ventilatory pattern. In the preliminary results we obtained in normal subjects, we found a ventilatory pattern that differed in important ways from those summarized by Shephard (31). In contrast to the results shown in Table 3, he gives normal values taken from several studies of f that ranged from 10.1 to 15.8 breaths/min with a mean value of 13.2 breaths/min. Our value of 19.3 breaths/min is 46% higher than this. McCool and colleagues (20, 21, 24, 25) measured the normal ventilatory pattern noninvasively from body surface measurements and found values for f that were only 10% higher than values obtained by spirometry. Our mean value for I of 6.9 l/min is identical to the mean value of the studies summarized by Shephard, although the range of his data (4.4-11.4 l/min) was considerably greater than the range we found (31). However, the VT we measured of 373 ml is considerably less than the mean value of 508 ml (range of 213-751 ml) reported by Shephard (31) and close to the value reported by Ashkanazi et al. (2) of 362 ml. They used a canopy system to make their measurements. Our value for VT/TI of 0.31 l/s is similar to the value of 0.298 l/s calculated from their data (2). They found that the primary effect of a mouthpiece and nose clip was to increase VT without change in TI so that VT/TI increased to 400 ml/s.

Ventilatory variability. We found considerable breath-to-breath variation in all ventilatory parameters. Nonrandom variations in ventilatory parameters are well described (see Ref. 4 for a review). Davis and Stagg (8a) found variations in VT and TI with constant VT/TI, whereas Ashkanazi et al. (2) found constant TI and TI/Ttot with variations in VT and VT/TI. We found variations in all these parameters. With the exception of TI/Ttot, these were better described by log normal than by normal PDF.

Noninvasive measurements of ventilation. In this study, we have shown that, with proper calibration, VT/TI rate can be derived from a tracheal sound signal and that VT can be estimated noninvasively to an accuracy within 15% of the measured volume signal. Because phonospirometry has the advantage of being easy to calibrate and is insensitive to postural change from sitting to standing, it has advantages over other noninvasive methods of measuring ventilation, such as inductance plethysmography and magnetometry (21, 30, 34). Its VT accuracy is equivalent to these techniques (20). It probably is not as accurate as optoelectronic plethysmography but is cheaper, easier to use, and less intrusive and does not require the subject's position and posture to be restricted so that the thorax is continuously visualized by videocameras.