IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 47, NO. 2, JANUARY 2000
239
Nonlinear Adaptive Noise Compensation in Electrogastrograms Recorded from Healthy Dogs Martin P. Mintchev, Arnaud Girard, and Kenneth L. Bowes
Abstract—Adaptive noise compensation is a popular method for improving signal-to-noise ratio in a variety of biomedical applications with its major disadvantage being the requirement for a reference channel containing noise strongly correlated to the noise in the primary channel. In many biomedical applications the utilization of a channel containing such noise without any representation of the information signal is difficult if not impossible. In this study we investigated the possibility of applying adaptive compensation in nonideal noise environments containing substantial presence of information signal in the reference channel. The signal in the reference channel was subjected to nonlinear manipulations for reducing the signal-to-noise ratio, thus diminishing the representation of information signal. The methodology was tested on canine electrogastrographic (EGG) signals of four unconscious dogs which underwent laparotomy and implantation of six pairs internal stainless steel electrodes in addition to the eight-channel abdominal EGG. Fourteen-channel (six internal and eight cutaneous) were obtained from each dog for 1 2 h. The signals were digitized and processed by computer. All internal signals showed regular and coupled gastric electrical activity with frequency of repetition in the normogastric range [3–9 cycles-per-minute (cpm)]. A single pair of primary and reference channels was selected from each cutaneous recording and exponential manipulators in the reference channels were introduced. The manipulators were tuned to maximize the percent distribution of spectral components in the canine normogastric range of each frequency spectrum calculated from the signal at the output of the adaptive compensator. Significant increment in the percent distributions in the normogastric range ( 0 01) was noted after tuning the exponential manipulator, and in many frequency spectra the recovery of the genuine dominant frequency peak of gastric electrical activity as determined by the internal recordings was noted. This study indicated that low percent distributions registered by some EGG channels are related to external nonlinear factors, the impact of which can be partially compensated. Index Terms—Adaptive compensation, electrogastrography (EGG).
GEA LMS
Gastric electrical activity. Least mean squares. I. INTRODUCTION
A. Principles of Adaptive Compensation
A
DAPTIVE noise compensation of canine and human EGG has been suggested for elimination of artifacts in these recordings [1]–[3]. The principles of adaptive filtering based on LMS compensation algorithm were described by Widrow et al. [4]. The cornerstone of this methodology is the assumption that along with the primary channel containing information signal and noise , a reference channel can be made available comwhich is strongly correlated with the prising solely of noise (Fig. 1). If this important connoise in the primary channel dition is fulfilled the compensatory procedure is quite straightis obtained through a forward. The compensation signal and combination of the reference signal, the output signal . Under ideal conditions the adaptathe weighting factors to the noise tion would equalize the compensation signal and by subtracting the compensation signal from the remains. It has primary signal only the information signal been shown mathematically [1]–[5] that the adjustment of for an optimal can be achieved by minimizing the energy is of the output signal . Assuming that the output signal signal and the the difference between the primary (which in fact is a manipulated image compensation signal ), one could write of the noise (1) After squaring (1)
NOMENCLATURE cpm EGG FHT
(2)
Cycles per minute. Electrogastrograms, electrogastrographic. Fast Hartley transform.
one can subsequently calculate the expected value of the result [2]
Manuscript received April 1, 1999; revised August 18, 1999. This work was supported in part by Natural Sciences and Engineering Research Council of Canada, Canada Foundation for Innovation and the Whitaker Foundation. Asterisk indicates corresponding author. *M. P. Mintchev is with the Department of Electrical and Computer Engineering, University of Calgary, 2500 University Drive, N.W., Calgary, Alta., Canada T2N 1N4. He is also with the Departments of Surgery, University of Alberta, Edmonton, Canada. (e-mail:
[email protected]). A. Girard is with the Department of Engineering, Universite d'Orleans, Orleans, ESPEO, France. K. Bowes is with the Department of Surgery, University of Alberta, Edmonton, Canada T6G 2B7. Publisher Item Identifier S 0018-9294(00)00885-5.
(3) and correspond to Keeping in mind that the crosscorrelation between noncorrelated signals (4) Thus, (3) now becomes
0018–9294/00$10 © 2000 IEEE
(5)
240
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 47, NO. 2, JANUARY 2000
The goal of the adaptive compensation process is to minimize the second term of (5) (6) where the noise is made by the adaptive process as close as . possible to the noise Equation (6) represents the essence of Widrow–Hoff LMS algorithm [4].
Fig. 1. Traditional adaptive compensation requires primary channel containing information signal s1 and noise n1, and a reference channel containing noise n2 strongly correlated with the noise n1 in the primary channel.
B. Nonideal Noise Environments. The assumption of the reference channel not containing any information signal, however, could be very restrictive, particularly in EGG applications. Kentie et al. [1], [2] noted that obtaining a reference channel containing noise strongly correlated with the noise in the primary channel but not containing any EGG signal is difficult if not impossible endeavor. The presence of information signal in the reference channel can be very damaging to the adaptive compensation process. Given that signal now contains and (the manipulated versions of the noise mixed with the representation of in the reference channel in the reference channel present due the information signal to the nonideal noise environment), it can be written that (7) Fig. 2. Schematic representation of the experimental setup used for the canine model.
Equation (5) can be rewritten (8) Adaptive compensation seeks to minimize (8)
(9) Equation (9) differs substantially from (6) because the component containing the information signal is now also a subject of minimization. Practically this means that the output of the adaptive compensator would be distorted and its reliability would be jeopardized depending on the magnitude of . Naturally, an appropriate solution for an adaptive compensator in nonideal environment would be the minimization of , which is equivalent to the minimization of the information signal component [or to the reduction of the signal-to-noise ratio (SNR)] in the reference channel. C. Applications of Adaptive Compensation in EGG Subsequent to the pioneering work of Kentie et al. (1981) [1], [2], Chen et al. [3] applied successfully the original method proposed by Widrow et al. [4] for compensation of respiration artifacts in human EGG. It should be noted, however, that respiration signals with their frequency range greater than 10 cpm in humans or greater than 12 cpm in dogs, do not pose a filtering problem whatsoever neither for the human nor for the canine EGG (maximal frequency range 1–8 cpm for humans, 1–10 cpm for dogs, normal frequency range 2.5–4 cpm for humans, 3–8 cpm
for dogs) with the signal having relatively narrow frequency spectrum due to the multiple signal transformations related to the nature of this noninvasive recording technique [6]. Naturally, respiration signals can be obtained easily from both humans or dogs using strain-gauge or temperature-based respiration sensors. Thus, the presence of EGG can be virtually eliminated and if such signal is utilized as a reference channel, the adaptive compensation could indeed be very adequate. Question arises, however, about the need for such fancy two-channel technique solely for the removal of respiration artifacts, which could be easily eliminated using a simple frequency-sampling digital filter [7], or even utilizing analog filtering techniques [8]. The real problem of EGG is not the respiration noise but the motion artifacts and the nonlinearities caused by external factors which dynamically distort the EGG recordings and in our opinion are the major reason for misinterpreting the EGG signals in the recent years [9]. While the origin and the impact of motion artifacts is more or less elucidated [3], [10], the issues related to the impact of external factors on EGG and particularly the nonlinear transformations of gastric electrical signals when they are recorded noninvasively from the abdomen have been discussed only recently [9], [11]. If an adaptive compensator is to be designed to combat these factors, finding a “pure noise source” for a reference channel would be a futile exercise since the external factors should definitely be recorded with an electrical sensor, and in this case the reference
MINTCHEV et al.: NONLINEAR ADAPTIVE NOISE COMPENSATION IN EGG’S RECORDED FROM HEALTHY DOGS
241
(a)
(b)
(c) Fig. 3. (a) GEA and (b) and (c)EGG recordings in which the poor quality of the EGG signal (channel 12 of the setup from Fig. 2) is evident while the corresponding internal recordings (channels 3 and 6 of the setup from Fig. 2) are perfectly regular and coupled. (b) Time- and (c) frequency-domain representations of selected primary channels revealed unrecognizable regularity and very low percent distribution of the spectral components in the normogastric range.
channel would inevitably contain some representation of the information EGG signal if the correlation between the
noises in the primary and in the reference channel is to be kept significant.
242
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 47, NO. 2, JANUARY 2000
(a)
(b) Fig. 4. Reference channels with a strong presence of information EGG signal were selected to test the methodology. In this example channel 9 of the setup from Fig. 2 was chosen.
D. Aim of the Present Study The aim of this study is to offer a methodology for adaptive noise compensation of EGG’s taking into account [1] the nonideality of the noise environment, and [2] the impact of nonlinear influences and distortions. II. METHODS A. Experimental Animals Six pairs of bipolar electrodes (stainless steel wires) were inserted into the gastric wall (three anterior; three posterior) of four anesthetized dogs at laparotomy. The animals were anesthetized using Penthotal (Abbott, Montreal, P.Q., Canada). The initial dosage of anaesthetic was 30 mg/kg and was supplemented with 3 mg/kg as needed based on monitoring the restoration of the blinking effect. Eight-channel abdominal
bipolar EGG was recorded using a technique described before [11]–[13], (Fig. 2). Fourteen-channel (six internal and eight cutaneous) -h recording were made from each dog in the basal state. Both types of signals (internal GEA and EGG) were digitized initially with 10 Hz. Prior to the digitization, analog active filters with two different pass bands were used because of the difference in the frequency bands of GEA and EGG. The high cutoff frequency of these two-pole Butterworth analog filters was 0.8 Hz for the internal GEA signals, and 0.3 Hz for the EGG. The low cutoff frequencies for both types of signals were set at 0.02 Hz. The signals were digitally filtered in the same frequency range using higher-slope frequency-sampled filters [8]. The conversion was done every 51.2 s (512 points with sampling frequency of 10 Hz). Internal GEA was assessed quantitatively by counting the waves in the successive 51.2-s intervals and assessing their
MINTCHEV et al.: NONLINEAR ADAPTIVE NOISE COMPENSATION IN EGG’S RECORDED FROM HEALTHY DOGS
243
(a)
(b) Fig. 5. (a) The idea of nonlinear manipulation of the reference signal can be presented as swings of the unity gain in the reference channel. (b) A nonlinear manipulator is added before the adaptive amplifier in the reference channel.
TABLE I CHANGES IN THE PERCENT DISTRIBUTION OF SPECTRAL COMPONENTS IN THE NORMOGASTRIC RANGE WITH THE TUNING OF THE PARAMETER . THE INFINITELY HIGH VALUE OF IS EQUIVALENT TO NOT USING NONLINEAR MANIPULATIONS IN THE REFERENCE CHANNEL ALTOGETHER
regularity and coupling using a technique described before [11]. EGG’s were converted in frequency domain using the FHT [13] and percent distributions of the spectral components in the normogastric range (3–9 cpm), higher-frequency range (9–30 cpm) and lower-frequency range (0.5–3 cpm)
were calculated. In our previous research [6], [9], [11]–[13], we have explained the reasons for the dramatic differences in the waveform and the spectral characteristics between the internal GEA and cutaneous EGG, and have derived a methodology to quantify normal EGG signals based on comparative studies of internal and cutaneous recordings [11]–[13]. Moreover, we have also shown that in multichannel EGG recordings some channels do not represent properly the dynamics of the internal GEA in certain intervals [11], [12]. Although at least three out of the eight EGG channels in each dog showed high (above 75%) average percent distribution of spectral components in the normogastric range [11], [12], some of the remaining EGG channels were of poor quality and could easily be labeled as “abnormal” according to criteria employed by other authors [14]–[16], although internally recorded GEA remained always perfectly normal and coupled. A single pair of primary and reference sources was selected from these “abnormal” EGG channels in each dog (Channels 12 and 9, see Fig. 2). The location of the electrodes for the primary and the reference channels in all dogs was kept the same. The main selection criterion was the minimal average percentage of spectral components in the normogastric range in the EGG signal, combined with perfectly regular internal GEA (Fig. 3). This combination was considered indicative that cutaneous EGG was dynamically distorted by the influence of nonlinear external
244
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 47, NO. 2, JANUARY 2000
factors. The selection of the reference channel was based on a strong presence of information EGG signal (Fig. 4). The experiments were approved by the University of Alberta Animal Welfare Committee. B. Nonlinear Manipulations in the Reference Channel Simple real-time nonlinear manipulations of the signals in the reference channel can be easily implemented using nonlinear digital gain of exponential type [Fig. 5(a)] [17]. The method could be thought of as making the line of the unity gain at the input of the reference channel swing in different but controllable directions (10) Obviously, the impact of these nonlinear changes of the refin a erence signal should be a reduction of the SNR controllable fashion. It is of primary importance, therefore, to choose appropriate values for the parameter , and to iteratively control its adjustment for optimal performance. The iterative to . Thus, the changes of can have wide scope from adaptive multiplier in the reference channel would manipulate , but the signals not directly the input signals at the output of the nonlinear manipulator [Fig. 5(b)]. C. Tuning the Nonlinear Compensation In order to monitor the impact of the nonlinear manipulation in the output signal, we used as a basis the fact that the internal recordings in all four dogs were normal and coupled according to previously published criteria [11], [13] [see Fig. 3(a)]. Consequently, we sought maximization of the percent distribution of the EGG signals in the normogastric range (3–9 cpm) obtained from the same recordings and the iterative adjustments of were performed in the direction of such maximization. Spectral representations of the output signal obtained after the adaptation were individually evaluated in order to determine the maximal percent distribution in the normogastric range for a given value of the nonlinear parameter . The iterations started with a negative and continued with a stepwise increment by 50 points until the maximal percent distribution in the normogastric range for the individual spectrum was obtained (Fig. 6). Similar iterations were performed for each subsequent 51.2-s segwas obtained for each time-doment. Distinct value of main interval corresponding to a single spectral representation. Thus, every 51.2 s the dynamic nature of the nonlinear factors influencing the recording was taken into account, obtaining a set of nonlinear manipulators for each dog. One set was obof optimal values for the nonlinear manipulator tained for each channel pair involved in the adaptive compensation process for each dog. These sets were examined for statistical difference using a standard two-tailed Student -test . Changes in the percent distribution values for the normogastric range for each 51.2-s interval before any manipulation, after ordinary adaptive compensation, and after nonlinear adaptive
Fig. 6. Block-diagram of the tuning procedure aiming at maximizing the percent distribution in the normogastric range.
compensation were also assessed statistically using a standard . two-tailed Student -test III. RESULTS Despite the nonideality of the noise environment, the adaptive compensation without using nonlinear manipulator in the reference channel improved dramatically the percent distribution in (Fig. 7). The application of the normogastric range nonlinear multipliers in the reference channel, however, lead to in the sets of norfurther significant improvement mogastric percent distributions in all four dogs after tuning the nonlinear parameter for each 51.2-s interval after the adaptation time (Fig. 8). In many cases before the application of the nonlinear manipulation, the dominant peak of the 51.2-s spectral representations at the output of the adaptive compensator was outside the normogastric range, although the internal recordings explicitly showed completely regular gastric electrical activity (see Figs. 7
MINTCHEV et al.: NONLINEAR ADAPTIVE NOISE COMPENSATION IN EGG’S RECORDED FROM HEALTHY DOGS
245
Fig. 7. (a) Compensated EGG signal and (b) its spectrum without using any nonlinear manipulator in the reference channel. Dramatic improvement in the percent distribution in the normogastric range was noted compared to the spectrum from Fig. 3(c).
and 3). After introducing the nonlinear manipulator and tuning toward maximizing the percent distribution in the normogastric range, the dominance of the correct EGG frequency component corresponding to the internal GEA frequency became evident (see Fig. 8). The tuning procedure applied to separate 51.2-s intervals after the adaptation time delivered a single maximum of the percent distribution in the predefined normogastric range of 3–9 cpm, with negative -parameters ranging from −5000 to −1250. An example of several percent distributions of EGG spectral com-
ponents in the normogastric range for different parameters obtained in the tuning process is shown on Table I. For each dog 35 values of the nonlinear parameter were obtained (one for each 51.2-s interval). Statistical examination of the sets of values obtained from different dogs failed to show . statistically significant difference between the sets IV. DISCUSSION In the present study we suggest that the greatest percentage of “abnormal” EGG’s do not objectively represent abnormal gas-
246
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 47, NO. 2, JANUARY 2000
Fig. 8. (a) Compensated EGG channel and (b) its spectrum using the optimal nonlinear parameter of the exponential manipulator in the reference channel. Over 15% improvement in the percent distribution in the normogastric range compared to the spectrum from Fig. 7 was noted.
tric electrical activity but are of poor quality due to the influence of a variety of unquantifiable nonlinear factors. Contrary to Chen et al. [3], we regarded these nonlinear influences related to external factors to be of paramount importance in EGG, not the respiration noise, which is of significantly higher frequency and can be eliminated using ordinary filtering techniques. Moreover, we suggest that the environment in which noise elimination from the EGG’s is to be achieved is not ideal from the point of view of adaptive compensation and some information signal
is always present in the reference channel. With the suggested technique of real-time nonlinear manipulations in the reference channel of the EGG adaptive compensator, the percent distributions in the normogastric range of some “abnormal” EGG channels in all four experimental dogs improved so dramatically, that these percent distributions became similar to the ones seen in normal EGG channels evaluated using the criteria published before [11], [13]. This improvement is probably related to two factors: 1) worsening of the signal-to-noise ratio in the reference
MINTCHEV et al.: NONLINEAR ADAPTIVE NOISE COMPENSATION IN EGG’S RECORDED FROM HEALTHY DOGS
channel of the adaptive compensator, thus, minimizing the presence of information signal in it and 2) minimization of some of the nonlinear influences affecting the signal quality. The observed improvements of the percent distributions in the normogastric range due to the adaptive process itself, however, underline another important aspect of the contemporary misinterpretation of EGG—there are procedures that can increase significantly these percent distributions, and considering the latter an important diagnostic parameter is dangerous and unreliable assumption. This fact is emphasized more in the present study of healthy dogs by the observation that in some time intervals recorded from abdominaly recorded EGG channels the normogastric percent distributions can be as low as 20%–30%, although simultaneous internal GEA recordings show that the internal GEA is perfectly regular and coupled. The present practice (see e.g., 14–16) has been to eliminate these intervals from the signal processing process. We disagree with this approach and suggest that dynamic nonlinear influences caused by various external factors are the major source for these disturbances. As evidence to support this hypothesis we present the described nonlinear adaptive compensation methodology, with which some of these nonlinear influences can be partially compensated and the normal EGG range can be restored to correspond to the normal GEA seen internally. Inevitably, these findings rise once again the question of the inappropriate terminology labeling the lower frequency components recorded from EGG as “bradygastrias,” and the higher EGG components as “tachygastrias” [14]–[16], which in their better percentage are probably not objectively existing phenomena [11], [12] but are a product of the dynamic nonlinear impact of various external factors. Since we based our standards on internal recordings, the iterative tuning of the suggested adaptive compensation method was directed at maximizing the percent distribution of EGG spectral components in the normogastric range, which in ideal recording conditions and lack of nonlinear and random noise influences should have been 100%. However, this study shows from another perspective that the influence of external nonlinear factors in EGG recordings cannot be entirely compensated and spectral components remain distributed outside the normogastric range even if the internal signals record entirely normal and coupled electrical activity. Labeling the presence of such EGG spectral components “bradygastrias” and “tachygastrias” is therefore misleading. As a result of this study a set of nonlinear manipulators was obtained for each dog, which when related with a particular channel pair in an adaptive compensation setup could maximize the percent distribution of spectral components in the normogastric range. These sets of nonlinear manipulators were not statistically different for the different dogs, implying that the given channel pair was subjected to similar nonlinear and noise factors in different animals.
V. CONCLUSION In this study, we describe an adaptive compensation method for canine EGG’s which takes into account the nonideal noise
247
environment and the nonlinear influences on some EGG channels. A quantitative methodology for tuning a real-time nonlinear multiplier in the reference channel aimed at maximizing the percent distribution of the EGG spectral components in the normogastric range is also discussed. The tuning procedure was applied to healthy dogs, in which this percent distributions were expected to be maximal. With this methodology we were able to obtain a set of nonlinear manipulators, which when used in the reference channels of adaptive compensators provided significant improvement in the normogastric spectral distributions of some poor-quality EGG channels. We conclude that substantial external nonlinear influences can contribute to the poor quality of some EGG channels leading to erroneous interpretation of the EGG’s as abnormal.
REFERENCES [1] M. A. Kentie, E. J. Van Der Schee, J. L. Grashuis, and A. J. P. M. Smout, “Adaptive filtering of canine electrogastrographic signals. Part 1: System design,” Med. Biol. Eng. Comput., vol. 19, pp. 759–764, 1981. [2] M. A. Kentie, E. J. Van Der Schee, J. L. Grashuis, and A. J. P. M. Smout, “Adaptive filtering of canine electrogastrographic signals. Part 2: Filter performance,” Med. Biol. Eng. Comput., vol. 19, pp. 765–769, 1981. [3] J. Chen, J. Vanderwalle, W. Sansen, G. Vantrappen, and J. Janssens, “Adaptive method for cancellation of respiratory artifact in electrogastric measurements,” Med. Biol. Eng. Comput., vol. 27, pp. 57–63, 1989. [4] B. Widrow, J. R. Glover, J. M. McCool, J. Kaunitz, C. S. Williams, R. H. Hearn, J. R. Zeidler, E. Dong Jr., and R. C. Goodlin, “Adaptive noise cancellation: Principles and applications,” in Proc. IEEE, vol. 63, 1975, pp. 1692–1716. [5] W. Philips, “Adaptive noise removal from biomedical signals using warped polynomials,” IEEE Trans. Biomed. Eng., vol. 43, pp. 480–492, 1996. [6] M. P. Mintchev, Y. J. Knigma, and K. L. Bowes, “Accuracy of cutaneous recordings of gastric electrical activity,” Gastroenterology, vol. 104, pp. 1273–1280, 1993. [7] J. G. Proakis and D. G. Manolakis, Digital Signal Processing. Englewood Cliffs, NJ: Prentice-Hall, 1996. [8] M. P. Mintchev and K. L. Bowes, “A new look at the amplification of gastric electrical signals,” presented at the 17 Int. Conf. IEEE–EMBS, Montreal, P.Q., Canada, Sept. 1995. [9] M. P. Mintchev and K. L. Bowes, “Impact of external factors on the stability of human electrogastrograms,” Med. Biol. Eng. Comput., vol. 34, pp. 270–272, 1996. [10] A. J. P. M. Smout, E. J. Van der Schee, and J. L. Grashuis, “What is measured in electrogastrography?,” Dig. Dis. Sci., vol. 25, pp. 179–188, 1980. [11] M. P. Mintchev and K. L. Bowes, “Do increased electrogastrographic frequencies always correspond to internal tachygastria?,” Ann. Bio. Eng., vol. 25, pp. 1052–1058, 1997. [12] M. P. Mintchev, A. Stickel, S. J. Otto, and K. L. Bowes, “Reliability of percent distribution of power of the electrogastrogram in recognizing gastric electrical uncoupling,” IEEE Transactions on BME, vol. 44, pp. 1288–1291, 1997. [13] M. P. Mintchev and K. L. Bowes, “Extracting quantitative information from digital electrogastrograms,” Med. Biol. Eng. Comput., vol. 34, pp. 244–248, 1996. [14] J. Chen and R. McCallum, “Electrogastrographic parameters and their clinical significance,” in Electrogastrography: Principles and Applications, J. Z. Chen and R. W. McCallum, Eds. New York: Raven, 1994, pp. 45–73. [15] K. L. Koch, R. M. Stern, M. Vasey, J. J. Botti, G. W. Creasy, and A. Dwyer, “Gastric dysrhythmias and nausea of pregnancy,” Dig. Dis. Sci., vol. 35, pp. 961–968, 1990. [16] S. Cucchiara, G. Riezzo, and R. Minella et al., “Reversal of gastric electrical dysrhythmias by cisapride in children with functional dyspepsia. Report of three cases,” Dig. Dis. Sci., vol. 37, pp. 1136–1140, 1992. [17] A. Girard and M. P. Mintchev, “Tuning nonlinear adaptive compensators in nonideal noise environments,” Med. Eng. Phys., vol. 20, pp. 689–696, 1999.
248
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 47, NO. 2, JANUARY 2000
Martin P. Mintchev received the M.Sc. degree in electronics from the Technical University of Sofia, Sofia, Bulgaria, in 1987, and the Ph.D. degree in electrical engineering from the University of Alberta, Edmonton, Alta., Canada, in 1994. After a brief postdoctoral fellowship in experimental surgery he became an Assistant Professor of Electrical Engineering and Surgical Research in Edmonton. Presently, he is an Associate Professor in the Department of Electrical and Computer Engineering at the University of Calgary, Calgary, Alta., Canada, and Adjunct Professor of Surgical Research at the University of Alberta.
Arnaud Girard received the M.Sc. degree in electronics from the Engineering School in Electronics and Optics, University of Orléans, Orléans, France, in 1999. In 1998, he was a Research Assistant with the Bioinstrumentation Laboratory at the University of Calgary, Calgary, Alta., Canada. He is currently a Design Engineer in a multinational company.
Kenneth L. Bowes received the M.D. degree from Queen's University, Kingston, Ont., Canada, in 1962, and M.Sc. degree in experimental surgery from the University of Alberta, Edmonton, Alta., Canada, in 1965. In 1967, he became a Fellow with the Royal College of Surgeons of Canada in 1967. He was a Medical Research Council of Canada Fellow in 1968 and 1969. He was a Research Fellow of the McLaughlin Foundation in 1969 and 1970. Since 1971 he has worked in the Department of Surgery at the University of Alberta, and is currently Professor of Surgery in that Department.