Venous air embolism (VAE) is hazardous when it occurs in large quantity during surgery [1,2]. Precordial Doppler ultrasound is one of the most sensitive and noninvasive monitors of VAE during surgery [3-6]. However, this method requires the constant attention of anesthesiologists listening to the subtle changes in the Doppler heart sound (DHS), and it provides poor quantitative information on the volume of air entering the heart chamber-information that may permit anesthesiologists to take action before VAE causes sequelae such as hypotension and dysrhythmias, which are directly related to the rate of air entrainment and volume of embolic air .
In our previous study, we examined the spectral characteristics of the embolic heart sound (EHS) during small volume VAE (0.01-0.1 mL) in anesthetized dogs . We reported that the EHS contained high-frequency components lying outside the range of normal DHS, which could be revealed by a properly chosen cutoff frequency. A simple high-pass filtering of EHS at different volumes of VAE showed a monotonous increase in the high-frequency power corresponding to the volume of air injected.
However, our previous study examined only small volumes of air. To further develop our method into a clinically useful tool for estimating air emboli, it was necessary to evaluate VAE in larger volumes, e.g. >or=to10 mL. Hence, the aim of the present study was twofold. First, we investigated whether wavelet analysis would be useful in analyzing the signals of the EHS in a real-time manner. Second, we determined whether the embolic power of the EHS for larger volumes of air is proportionate to the volume of VAE that has been reported for small volumes of VAE. Wavelet analysis [9-12] was chosen because of its speed in examining spectral details, allowing potential application for real-time monitoring during VAE. Analysis of similar time-varying signals such as evoked potentials  and the electrocardiogram  has been successful.
The study was approved by the Animal Research Committee of the Veterans General Hospital-Taipei. Seven mongrel dogs weighing 13.4 +/- 1.7 kg were anesthetized with pentobarbital (40 mg/kg IV) and placed in the supine position. The trachea was intubated with a cuffed tracheal tube, and the lungs were ventilated mechanically by using a Harvard ventilator (Model 34RH; Harvard Apparatus, Natick, MA) delivering 100% oxygen. Tidal volume (15-20 mL/kg) was adjusted to maintain PETCO2 within normocapnic range. A 16-gauge Teflon catheter (Arrow International, Reading, PA) was inserted via the right external jugular vein for air infusion. Supplemental pentobarbital (20 mg/kg IV) was given as needed for anesthetic maintenance. The rectal temperature was controlled at 37 +/- 1[degree sign]C using a heating lamp, and 0.9% normal saline in glucose solution was infused (3 mL [center dot] kg-1 [center dot] h-1 IV) for intravascular volume maintenance. Cardiac rhythm was monitored by lead II electrocardiography (Model 90601A; SpaceLab, Redmond, WA).
A coupling gel (Aquasonic 100; Parker Laboratories, Orange, NJ) was applied generously to the surface of the 2.4-MHz Doppler ultrasonic flow transducer (Versatone Doppler; MedaSonics, Fremont, CA). The transducer was positioned between the third and fourth and fourth and fifth intercostal spaces to the right of the sternum and was secured to the chest. Correct placement of the transducer over the heart was confirmed by the clearly audible sound of turbulence induced by rapid injection of 10 mL of saline into the external jugular vein. The circuit of the Doppler device is constructed to amplify only those components representing forward flow with respect to the probe. During the entire experiment, the line output of the Doppler transducer was recorded on a tape recorder (Sony Professional, Tokyo, Japan). The DHS signal was then digitized at 4 kHz with 16-bit resolution on a multimedia personal computer. The overall frequency response of the recording system was within 3 dB from 20 to 18,000 Hz.
After obtaining a 5-min control baseline recording, we sequentially injected small volumes of air (0.01, 0.02, 0.05, 0.07, 0.1, 0.15, 0.2, 0.3, 0.4, and 0.8 mL) into the external jugular vein through the catheter. Each injection was made by using a microsyringe (Hamilton, Reno, NV) at intervals of 2-5 min. After a 10-min stabilization period, a series of larger volumes of air (0.8, 1.6, 2.4, 4.8, and 9.6 mL) was infused at various controlled rates of 0.011-0.128 mL [center dot] kg-1 [center dot] min-1 over 5 min by using a precalibrated syringe pump (Model PSK-01; Nikkiso Co. Ltd., Tokyo, Japan). The precision of the syringe pump was considered to be +/- 1% at an infusion rate ranging from 0.1 to 300 mL/h. We also allowed a 5-min interval between successive infusions. No premature ventricular contraction or other electrocardiogram changes were detected.
The basic principle of wavelet analysis is to approximate a signal at fine-to-coarse steps (called resolutions or scales). The process is similar to band-pass filtering of the signal at successively narrower band-widths. The signal is divided into two parts, i.e., the low-pass and high-pass signals, by wavelet analysis at scales equal to 1. The former is defined as the approximation signal (i.e., the coarser approximation of the signal), whereas the latter is difference signal (i.e., the difference between the two signals). The same procedure is then applied to the approximation signal during wavelet analysis at successive scales. One advantage of wavelet analysis is the existence of computationally efficient algorithm for its implementation . We can compute the approximation signal A2s f(x) and difference signal D2s f(x) of the DHS signal at various scales using the following algorithms: Equation 1 and Equation 2 where we define, for normalization purpose, the original DHS signal as the approximation signal at s = 0, i.e. A20 f(x). The hk and gk are the impulse responses (i.e., the inverse Fourier transform of the frequency response) of the low-pass and high-pass filters, respectively. The choice of wavelet functions only affects the impulse responses of the filters and the processing time. In the present study, we used the Daubechies 20-coefficient wavelet  for the DHS analysis. This choice was a compromise between performance and processing time. Further details of the wavelet analysis can be found elsewhere [9-12].
A real-time monitoring software package for VAE was developed in our laboratory using wavelet analysis as the core algorithm. It computes the signal power of the wavelet-processed DHS in a real-time manner. For statistical comparisons, the signal power was normalized with respect to the mean signal power during the control period. The embolic DHS was extracted by a method of thresholding on the signal power. The total power associated with the embolic DHS induced by a particular air dose was calculated and called cumulative embolic power (CEP). Linear regression analyses were performed to identify a significant relationship between the CEP and injected volumes of air. Student's t-test was used to compare the slope and intercept between a bolus injection and a continuous infusion. A value of P < 0.05 was considered statistically significant. The EHS results predicted by the wavelet method were compared with those predicted by the conventional method, in which an anesthesiologist (WH) blinded to the study protocol reviewed the original DHS audio signals. The sensitivity, specificity, positive predictive value, and negative predictive value of the wavelet analysis were calculated by using standard formulae. In particular, these variables were evaluated at both s = 1 and s = 2 in combination with different threshold levels (2-4 times the mean control level). Briefly, sensitivity is the number of cases predicted by using wavelet analysis to have true EHS, which was divided by the total number of EHS detected by listening. The specificity is the number of cases predicted by using wavelet analysis not to have true EHS, which was divided by the total number of EHS detected by listening. Positive predictive value is the number of cases predicted by using the wavelet method to have true EHS, which was divided by the total number of EHS predicted by using wavelet analysis. Negative predictive value is the number cases predicted by using wavelet analysis not to have true EHS, which was divided by the total number of cases predicted not to have EHS by using wavelet analysis.
A representative tracing of the original and wavelet-processed DHS signals is shown in Figure 1. Although the approximation signals (A1 and A2) grossly resembled the original DHS, the difference signals (D1 and D2) showed striking features distinguishing EHS from background. The quality of this differentiation for D1 easily exceeded that of D2 during the real-time processing.
(Figure 2) shows the results of the difference signal power at different scales. At s = 1 or s = 2, EHS could be easily extracted by using a threshold method (e.g., threshold level = 2) as the contrast of embolic to normal signal power became greatly enhanced. The periodic rhythm in the baseline DHS signal that reflects chest wall movement produced by the ventilator did not affect the performance of wavelet analysis.
The CEP of the difference signals were computed at s = 1 and s = 2 for each volume of air injection due to different threshold levels (2-4 times control). Linear regression, correlation coefficient (r), and level of significance of the regression slope were also obtained from the above data on a log-log scale (Table 1). It is obvious that the difference signals at s = 1 performed better than those at s = 2. In particular, a twofold threshold yielded best results. At this threshold, each subject did not differ statistically in each slope and intercept of the regression line (P < 0.05) derived from either bolus injection or continuous infusion. For the seven dogs, the small relative error (i.e., standard deviation divided by mean) of the slope and the intercept suggested the interanimal consistency of the present method. Linear relationship (y = 1.08x + 7.89; r = 0.75, P < 0.001), as well as the 95% confidence intervals for slope (0.89-1.27) and intercept (7.65-8.13), were found at s = 1 (using 2 times threshold) between the CEP and air volumes injected by bolus and continuous infusion on the natural log-log scale (Figure 3).
The sensitivity, specificity, positive predictive value, and negative predictive value of wavelet analysis at different scales and threshold levels are shown in Table 2. The sensitivity of the wavelet analysis was inversely proportionate to the threshold levels. At the smallest volume of air (0.01 mL), it decreased to 71.4% when the threshold level was set at 4. The positive predictive value was 100% regardless of the threshold level. Because denominator is 0, the specificity and negative predictive value were not applicable. The difference signals at s = 1 and at 2 times threshold were found effective and reliable for detecting EHS in a real-time manner. In clinical application, these characteristics will be useful in developing an automatic quantitative monitor for VAE.
The peaks of the signal power during continuous infusion (Figure 4, lower panel) was actually a multiple of the signal power during bolus infusion (Figure 4, upper panel). Each peak against the background baseline indicates the EHS.
Using wavelet analysis, we could scan the DHS signal at different scales in a computationally efficient way to accurately quantify the VAE. In particular, we found that the difference signal at a scale equal to 1 greatly facilitated the extraction of the abnormal EHS from the normal DHS signals, especially with a threshold of 2 or 3. This feature allows the design of a real-time system for automatic VAE monitoring.
The results of the present study show remarkable similarity in the embolic signal power between the bolus and continuous infusions. The audio signal of the EHS during continuous infusion was also similar to that during a series of bolus injections. As such, we used the embolic signal power of bolus injection as the basis to compute the CEP for each infusion to estimate the volume of air emboli. The pattern of the signal power during continuous infusion actually bears great similarity to a superposition of what happened during the bolus infusion at different times (Figure 4). Although the rate of the continuous infusion was kept constant for each volume, the EHS occurred intermittently. This phenomenon probably reflects the pulsatile nature of the syringe pump during infusion. In addition, our results also suggest the nonlinearity of the air emboli traveling from the external jugular vein to the right atrium. In fact, in some cases, a bolus injection of 0.8 mL resulted in an EHS similar to that observed with the continuous infusion. Hence, it is conceivable that larger volumes of air entering the heart would be broken down into progeny microbubbles of smaller diameters. It is interesting to note that the CEP obtained after a bolus injection of 0.8 mL was often greater than that after a continuous infusion of the same amount. This finding is in agreement with those of other studies suggesting that the risk associated with VAE may be related more to the rate than to the volume of air entering the venous system . Therefore, our results suggest that the magnitude of the CEP may be useful in predicting the severity of VAE. However, the accuracy of the embolic power is highly dependent on the nature of the reflected Doppler signal, which in turn depends on the biophysical aspects of gas bubbles in blood . Encapsulated bubbles, multiple bubbles with scattered signals, bubbles blocking the sound waves of others, and gross movement artifacts are among the most common factors that affect accuracy of the embolic power.
The Doppler ultrasound instrument presents the audible heart sound with amplitude and frequency proportional to the power of reflected ultrasonic signals and velocity of reflective structures. Because the ultrasonic beams cannot pass through air, strong reflection is expected to cause large shifts in the amplitude of the EHS. This is the physical rationale of the present study to quantify the volume of air emboli . Because the EHS contains high-frequency components, whereas normal DHS present mainly within the low-frequency range, it is conceivable that air emboli move faster than the surrounding red blood cells when entering the heart chamber. In fact, this conforms to the classical Newtonian law of motion (f = m x a), revealing that under the same contractile force of the ventricle, air emboli of lighter mass (m) have greater acceleration (a) and, hence, higher speed with respect to the red blood cells. This speculation assumes that air emboli behave like particles traveling in the blood. However, in a fluid such as blood, significant viscous drag is imposed on the bubbles, which tends to slow their velocity. Additionally, because air bubbles lack mass relative to the blood that surrounds them, they have low inertia, which further diminishes their velocity. Also, the buoyancy of the bubbles affects their motion in the blood stream . Therefore, the etiology of the high-frequency components in the EHS signal is still undetermined and awaits further investigation.
Our conclusions were drawn with the assumption that air injected into the external jugular vein passes through the Doppler detection field once without being trapped or coalescing in the atrial appendages. Only under these conditions is the reflected Doppler signal proportional to the injected volume of air . With repeated air infusions and under prolonged anesthesia, variabilities in the DHS were inevitable during the experiment. Anatomically, the atrial appendages in the supine position are located superior to the atrioventricular valve. Thus, it is likely that a portion of the air entering the chamber accumulates at the appendages. The cumulative embolic power measurement thus reflects the presence of this accumulated air, resulting in a significant overestimation of volume. The above notion was further supported by the present findings that the threshold for detecting the air emboli required readjustment after successive air infusions, as the baseline power level might have drifted because of cumulated VAE. In dogs, the ultrasonic probe is usually positioned over the third to fourth intercostal space to the right of the sternum. A more superior location provides a larger portion of the signal from the vena cava, whereas a more inferior position provides a greater signal from the atrium. In the present study, the transducer was correctly placed over the third to fourth or fourth to fifth intercostal space whenever we clearly heard turbulence induced by a bolus injection of 10 mL of saline into the central vein. Given the variable anatomy of the chest wall and the difference in placing the ultrasonic probe, the choice of wavelet analysis allows fast adjustment to obtain optimal settings for the filter. The relative error in the slopes and the intercepts of the regression lines demonstrated the repeatability of wavelet analysis among the dogs. Hence, our present method could permit the use of a computerized solution for a real-time VAE monitoring system.
In some dogs, an increase in heart rate would occur during the infusion of air at high volumes. This may be due to a sympatholytic reflex mediated through receptors in the pulmonary vasculature . Reflex tachycardia was occasionally encountered, possibly secondary to hypotension induced by supplements of pentobarbital for anesthetic maintenance. It is interesting to find that as red blood cells traveled faster at accelerated heart rates, the high-frequency components of the DHS signals increased, causing a transient enhancement in the embolic signal power.
We argue that the injection of 10 mL of air, far from the lethal volume of air embolus in dogs, is too small to simulate a clinically significant VAE, because as much as 250 mL of air was aspirated from the right atrium during a VAE episode in humans . In addition, the dog can tolerate as much as 150 mL of air, which is equivalent to 1.5 mL [center dot] kg-1 [center dot] min-1 for at least five minutes. Higher volumes may be clinically more important, but they were not tested because they cause tachycardia or cardiac arrhythmia. Furthermore, this study sought to provide a method of early detection so as to allow treatment to begin at approximately 10% of the lethal VAE dose. In this regard, a 20% lethal volume is no different from twofold the 10% doses occurring successively in time. In consideration of the extremely high reflective property of air, the reflected Doppler signal will be extremely strong at these high volumes and may result in the saturation of the audio amplifier. Therefore, the present data are still clinically relevant.
Periodic rhythm (0.25 Hz), as a result of the chest wall movement related to the ventilator, had minimal effects on the wavelet analysis (see Figure 2). The sensitivity of the wavelet method was excellent within the range of 0.01-10 mL of VAE, which is sufficient to include both subclinical and clinically relevant embolic air volumes.
In conclusion, these results confirm our hypothesis that wavelet analysis is as effective as a real-time monitor during continuous VAE and that it is possible to determine larger volumes of air emboli based on previous injections of small volumes of air.
1. Buckland RW, Manners IM. Venous air embolism during neurosurgery: a comparison of various methods of detection in man. Anaesthesia 1976;31:633-43.
2. Young ML, Smith DS, Murtagh F, et al. Comparison of surgical and anesthetic complications in neurosurgical patients experiencing venous air embolism in the sitting position. Neurosurgery 1986;18:157-61.
3. Edmonds-Seal J, Maroon JC. Air embolism diagnosed with ultrasound: a new monitoring technique. Anaesthesia 1969;24:438-40.
4. Michenfelder JD, Miller RH, Gronert GA. Evaluation of an ultrasonic device (Doppler) for the diagnosis of venous air embolism. Anesthesiology 1972;36:164-7.
5. English JB, Westenskow D, Hodges MR, Stanley TH. Comparison of venous air embolism monitoring methods in supine dogs. Anesthesiology 1978;48:425-9.
6. Malinow AM, Naulty JS, Hunt CO, et al. Precordial ultrasonic monitoring during cesarean delivery. Anesthesiology 1987;66:816-9.
7. Shupak RC. Air embolism and its influence on anesthetic management. Cur Rev Clin Anesth 1991;11:113-24.
8. Lui PW, Lin YM, Chan FHY, et al. Spectral characteristics of embolic heart sounds detected by precordial Doppler ultrasound during venous air embolism in dogs. Br J Anaesth 1993;71:689-95.
9. Mallet SG. A theory of multiresolution signal decomposition: the wavelet representation. IEEE Trans Patt Anal Machine Intell 1989;11:674-93.
10. Rioul O, Vetterli M. Wavelets and signal processing. IEEE Signal Proc 1991;8:15-38.
11. Meyer Y. Wavelets and operators. Cambridge: Cambridge University Press, 1992.
12. Daubechies I. Ten lectures on wavelets. Philadelphia: Society for Industrial and Applied Mathematics, 1992.
13. Thakor NV, Guo XR, Sun YC, Hanley DF. Multiresolution wavelet analysis of evoked potentials. IEEE Trans Biomed Eng 1993;40:1085-93.
14. Li C, Zheng C, Tai C. Detection of ECG characteristic points using wavelet transforms. IEEE Trans Biomed Eng 1995;42:21-8.
15. Bulter BD. Biophysical aspects of gas bubbles in blood. Med Instrument 1985;19:59-62.
16. Chan BCB, Lui PW, Poon PWF, et al. Quantification of venous air embolism using Doppler heart sound. Proc Int Conf Biomed Eng 1996;287-90.
17. Young HD, Freedman RA. University physics. 9th ed. Reading, MA: Addison-Wesley, 1996.
18. Bunegin L, Wahl D, Albin MS. Detection and volume estimation of embolic air in the middle cerebral artery using transcranial Doppler sonography. Stroke 1994;25:593-600.
19. Adornato DC, Gildenbery PL, Ferrario CM, et al. Pathophysiology of intravenous air embolism in dogs. Anesthesiology 1978;49:120-7.
20. Albin MS, Carroll RG, Maroon JC. Clinical considerations concerning detection of venous air embolism. Neurosurgery 1978;3:380-4.