1764
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 10, OCTOBER 2005
Pulsed Doppler Signal Processing for use in Mice: Design and Evaluation Anilkumar K. Reddy*, Member, IEEE, Alan D. Jones, Christian Martono, Walter A. Caro, Sridhar Madala, Member, IEEE, and Craig J. Hartley, Senior Member, IEEE
Abstract—We have developed and evaluated a high-frequency, real-time pulsed Doppler and physiological signal acquisition and analysis system specifically for use in mice. The system was designed to provide sampling rates up to 125 kilosamples/s (ksps) with software controlled data acquisition and analysis in real-time. Complex fast Fourier transforms are performed every 0.1 ms (or longer up to 10 ms) to provide 0.1-ms time resolution and using 64–1024 sample segments of the Doppler audio signals resulting in frequency resolution ranging from 122–1953 Hz. The system was evaluated by its response to frequency swept signals with slopes (accelerations) and magnitudes (velocities) comparable to actual blood velocity signals in mice. Signals up to a maximum frequency of 125 kHz and a maximum acceleration of 20 MHz/s were processed and displayed. This corresponds to a maximum velocity of 480 (960) cm/s and a maximum acceleration of 750 (1500) m/s2 when Doppler shifts are measured with a 20- (10-) MHz probe, thereby allowing us to measure high stenotic jet velocities. The directional transitions of the spectrogram across zero frequency and across Nyquist frequency (sampling rate/2) were smooth with no discernible artifacts. Signals with period as low as 2 ms were processed and displayed at sweep speed that is ten times that in clinical Doppler systems, so that measurements of small temporal events can be made with precision. Thus, the new system can measure higher blood velocities with higher spatial and temporal resolution than is possible using clinical Doppler systems adapted for use in mice. Index Terms—Doppler signal processing workstation, doppler spectrogram, high-frequency pulsed Doppler ultrasound, mouse cardiovascular system.
I. INTRODUCTION
S
MALL body mass (15–35 g) and high heart rates (500–700 beats/min) make the measurement of hemodynamic parameters challenging in mice. Clinical ultrasound systems cannot be used to accurately measure blood flow velocities in mice because these systems typically operate at low frequencies (poor spatial resolution), their spectrum analyzers have poor temporal resolution [1] and their probes cannot be oriented properly. Previously we had developed 10-MHz and 20-MHz pulsed Doppler probes [1], [2] and used them to measure blood velocities in mice [3]–[5]. The spatial resolution
Manuscript received August 20, 2004; revised January 9, 2005. This work was supported in part by National Institutes of Health under Grant HL-52364, Grant HL-22512, and Grant AG-13251, and in part by the Texas Advanced Technology Program. Asterisk indicates corresponding author. *A. K. Reddy is with the Baylor College of Medicine, Houston, TX 77030 USA (e-mail:
[email protected]). A. D. Jones, C. Martono, W. A. Caro, and S. Madala are with the Indus Instruments, Houston, TX 77058 USA. C. J. Hartley is with the Baylor College of Medicine and DeBakey Heart Center, Houston, TX 77030 USA. Digital Object Identifier 10.1109/TBME.2005.855710
is significantly improved at these high frequencies. However, the measured Doppler audio signals were processed using an adapted clinical system [Medasonics spectrum analyzer (MSA)] whose time resolution was limited to 8 ms. This time resolution is not adequate to measure the time of cardiac events in a mouse [3], [5]. The MSA has good frequency resolution (12.5–150 Hz) at sampling rates of (3.2–38.4 ksps), but its maximum frequency is limited to 24 kHz (or a velocity up to 220 cm/s using a 10-MHz pulsed Doppler) which is adequate to measure blood velocities in normal mouse arteries. However, in mouse models of aortic stenosis the peak blood flow velocities just past the stenosis at the arch can be on the order of 400 cm/s or more [6], [7]. Thus, the adapted clinical system does not provide adequate time resolution for the measurement of hemodynamic parameters in mice and does not have adequate bandwidth to measure high Doppler frequency shifts associated with stenotic jet velocities in mice. An important parameter that serves as an indicator of arterial compliance is pulse wave velocity (PWV) which is determined by measuring the difference in time of onset of the blood velocity signals at two sites along a vessel separated by a known distance [6]. The instrument processing the velocity signals must have good time resolution for measurement of the transit time of the foot of the velocity waveform from the first site to the second site. Since the MSA had poor temporal resolution, we previously used a zero-crossing interval histogram (ZCIH) counter to maximize the temporal resolution of the Doppler signals to determine PWV in mice [8]. Thus, the need arises for a Doppler signal processing system which can sample data at high rates, has better frequency and temporal resolution, and is capable of simultaneously acquiring and analyzing multiple cardiovascular signals from a mouse. Here, we describe the design of a high-speed, high-frequency, real-time data acquisition system [Doppler signal processing workstation (DSPW)] and evaluate its performance with known signals from a signal generator. In the succeeding article [9] the performance of the DSPW in measuring cardiovascular signals is compared to that of the MSA and ZCIH we had used previously. Also, we describe several applications of the system through the measurement of Doppler flow velocity signals and other cardiovascular signals such as electrocardiogram (ECG) and blood pressure (BP) in mice. II. METHODS A. The Doppler Spectrum Analyzer The DSPW is a real-time spectrum analyzer and data analysis system developed by Indus Instruments in collaboration with
0018-9294/$20.00 © 2005 IEEE
REDDY et al.: PULSED DOPPLER SIGNAL PROCESSING FOR USE IN MICE: DESIGN AND EVALUATION
1765
Fig. 1. Block diagram of Doppler signal processing workstation shown with the flow of transduced signals from the animal to the analog signal processing modules that generate the Doppler in-phase (I) and quadrature (Q) audio signals, and other cardiovascular signals such as ECG, BP, etc. The DSPW hardware consists of high-speed DSP and A/D boards. The DSPW software controls the hardware and allows for acquisition, analysis, display, and storage.
Baylor College of Medicine specifically for use in mice. The DSPW can accept signals from any analog ultrasonic Doppler system capable of generating in-phase (I) and quadrature (Q) demodulated audio signals. The hardware includes an analog-to-digital (A/D) board and a 40-MHz Analog Devices SHARC3000 fast floating point digital signal processing board (Spectrum Signal Processing, Canada) with both boards connected by a serial link. The A/D board has six analog input channels two of which are dedicated to Doppler I and Q audio signals, and the remaining four channels receive ECG, pressure, and/or other physiologic/cardiovascular signals. The hardware is controlled by a software program (written in Visual C++ which runs on MS Windows 2000/NT/XP operating systems) which was designed to acquire, process, and display signals in real-time. Fig. 1 shows a block diagram of data flow to the DSPW. In the offline mode the program can perform semi-automatic analysis of data, export raw and analyzed waveforms, and generate reports. The analog input channels have separate A/D converters which can sample up to 125 kilosamples/s (ksps) in each channel with a 16-bit sample size and input signal range of 3.0 V. The fourth-order low-pass filter can be programmed to cutoff at any frequency up to 25 kHz or be turned off (for high-velocity signals expected in stenosed vessels). The Doppler I and Q signals are analyzed using a complex fast Fourier transformation (cFFT) at sample sizes of 64–1024 samples. The symmetric properties of the cFFT allow for the separation of forward and reverse directions of flow velocity [10] and the Doppler frequencies can be resolved up to the sampling rate instead of the conventional limit of half the sampling rate [11]. At a sampling rate of 125 ksps these cFFT sample sizes result in 0.51–8.19-ms window lengths and 1953–122 Hz frequency resolutions. The velocity signal is calculated from using the Doppler equation the Doppler shift frequency , where is the speed of sound is the ultrasonic frequency (5-, 10-, or (1540 m/s) and 20-MHz). The Doppler angle can be entered manually into the program to correct for the angle between the sound beam and the direction of flow. The sweep speed of the display can be varied both in real-time as well as offline resulting in temporal resolutions from 0.1–10 ms. The cFFT is calculated every 0.1 ms with overlapping cFFT sample windows to achieve 0.1-ms time resolution, and the data is displayed on the screen using 10 pixels/ms (fast
Fig. 2. Calculation of a spectrogram from Doppler in-phase (I) and quadrature (Q) audio signals. The I and Q signals are sampled at a preset sampling rate and the samples are streamed into a circular buffer. A set of the acquired samples is processed using a window function (cosine, hamming, etc), window location (past, center, future: with respect to Ic and Qc), sample size (64, 128, 256, 512, 0.1 ms, slowest 10ms). The spectrum 1024), and sweep speed (fastest obtained from each cFFT calculation is plotted as a Frequency versus Time as shown above. The spectrogram shown is the aortic flow velocity (during systole) in a mouse. The signal was sampled at 125 ksps and a 512-sample cFFT was calculated and displayed at fastest sweep rate. The maximum frequency (or velocity) of each spectrum represents the bandwidth of the signal in that sample window and the intensity represents the energy of the frequency components (black—high energy; white—zero energy).
=
=
sweep) for a 100-ms full scale display. Similarly, cFFT is calculated every 10 ms (10-ms time resolution) and the data is displayed using 0.1 pixel/ms (slow sweep) for a 1-s full scale display. While all six channels are sampled at the highest rate, only signals acquired via the two channels designated for Doppler audio signals are retained at the sampled rate and all other signals (ECG and three auxiliary channels) are decimated to a rate of 4 ksps which is adequate to represent physiologic signals in mice. The record length of data to be saved can be set by the user from 1–60 s. The saved data file is in raw form, and every time the file is opened, the spectrogram is recalculated and displayed. The spectrogram is also recalculated every time the sweep speed or the cFFT sample size is changed. The analysis mode of the software allows for calculation of the maximum frequency envelope of the Doppler spectrogram and automatic detection of R-wave peaks from the ECG. The analysis procedures are more elaborately discussed in the succeeding paper (Applications) [9]. B. Spectral Analysis of Doppler Signals and Extraction of Maximum Frequency Envelope Each spectral estimate is calculated as shown in Fig. 2. The I and Q samples of the data are selected from the buffer (n1, n2, , nc, , ni, , n1—earliest sample, ni—latest sample) as per the user’s cFFT sample size (64–1024) setting. The cFFT is calculated based upon the preset sweep speed (0.1 ms at 10 pixels/ms to 10 ms at 0.1 pixel/ms) to obtain a spectral estimate of each set of I and Q samples. The display of the velocity spectrogram is optimized by adjusting the gain, contrast, and noise levels either automatically or manually (each one can be adjusted independently) before the calculation of the maximum frequency envelope. The adjustment of gain, contrast, and noise uses the concept of histogram
1766
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 10, OCTOBER 2005
Fig. 3. (a) A signal V (t) was used to control the voltage controlled oscillator to produce a 10 kHz to 115-kHz frequency sweep signal x(t). The signal x(t) was input into an analog Doppler module which contains a 90 phase-shifter [13] to produce the in-phase, x (t) and quadrature, x (t) components. Both the x (t) and x (t) signals were acquired by the DSPW at a sampling rate of 125 ksps and processed to generate an output frequency spectrogram signal F (t), (b) The sweep signal (V 1 to V 2) and the corresponding frequency signal (F 1 to F 2) it generates. V 1 and V 2 are arbitrary nonzero voltages levels.
0
equalization for image enhancement [12]. Following the adjustment of the spectrogram the maximum frequency or envelope of the spectrum is automatically calculated using the percentile method [13] by choosing the frequency below which 95% of the total power in each spectrum lies. The envelope is calculated every 0.1 ms and smoothed using a 41-point sinc function (sin thereby having the effect of subjecting the maximum frequency envelope to a noncausal (depends on past, present, and future values) low-pass filter with a cutoff frequency of about 244 Hz. After calculation of the maximum frequency envelope, the envelope, ECG, and the signals from the auxiliary channels can be exported at sampling intervals of 0.1–10 ms. Although the ECG and the signals from the auxiliary channels are saved at a maximum of 4 kHz, they can be re-sampled by interpolation to obtain sampling rates up to 10 kHz (0.1-ms sample interval). C. DSPW Testing Protocol The DSPW was tested using signals from a sweep generator with accelerations and magnitudes in the range of physiologic signals from mice. Fig. 3 shows a block diagram of a typical test procedure. The system was evaluated with arbitrary inputs of a 100-ms triangular sweep of frequencies ranging from 10 kHz to 115 kHz to demonstrate the dynamic range of the DSPW required for making blood flow velocity measurements in mice, triangle sweeps of 1–20 kHz with periods of 20, 10, 5, 2 ms to demonstrate linearity and temporal resolution, 10-ms tone burst of 2 kHz to demonstrate the effect of cFFT sample size on the spectrogram, and saw-tooth sweeps of 62.5 to 62.5 kHz with a period of 100 ms to demonstrate the smooth transition from negative frequency (reverse flow velocity) to positive frequency (forward flow velocity). All the signals were acquired at a sampling rate of 125 ksps. Also, to demonstrate normal, aliased, and unwrapping of aliased signals we used 500-ms triangular sweeps of 1–10 kHz and 1–20 kHz sampled at 31.25 ksps. III. RESULTS The user interface window of the DSPW system is shown in Fig. 4 containing the spectrogram of the 100-ms triangular
Fig. 4. DSPW interface showing the spectrogram of a signal containing frequencies from 10 kHz to 115 kHz with an arbitrary period of 100 ms. The in-phase and quadrature signals generated by the analog Doppler system were acquired by the DSPW at a sampling rate of 125 ksps, processed using 256-sample cFFT, and displayed.
0
Fig. 5. Effect of cFFT sample size on temporal width and frequency resolution of the spectrogram of a 10-ms 2-kHz tone burst. Best temporal resolution is obtained at small cFFT sample size and best frequency resolution is obtained at large cFFT sample size. Here, the in-phase and quadrature signals were sampled at 125 ksps. The sweep speed was set to 10 pixels/ms (cFFT calculated every 0.1 ms). The temporal width of each calculated spectrum depends upon the sampling rate, sample size.
sweep of frequencies ranging from 10 kHz to 115 kHz. The input signal was generated as shown in Fig. 4. The spectrogram generated is smooth without any distortions or aliases. The effect of window sample size on the spectrogram of a 10-ms 2-kHz tone burst is shown in Fig. 5. Here, the sweep speed (10 pixels/ms) was kept constant and the cFFT sample size was varied. While the 64-point cFFT generated a spectrogram with best temporal resolution the 1024-point cFFT generated a spectrogram with best frequency resolution. The response of the DSPW to 1–20 kHz triangle sweep signals each with periods of 20, 10, 5, and 2 ms is shown in Fig. 6. The spectrogram is shown in the upper panel and the voltage signal used to generate the 1–20 kHz frequency sweep is shown in the lower panel. The signals were sampled at 125 ksps and a 64-point cFFT was used to calculate the spectrogram. The spectrogram of the signal with the shortest period (2 ms) shows adequate time resolution to make temporal measurements. In response to a 62.5–62.5-kHz saw-tooth sweep signal the DSPW generated a spectrogram that demonstrated smooth
REDDY et al.: PULSED DOPPLER SIGNAL PROCESSING FOR USE IN MICE: DESIGN AND EVALUATION
1767
Fig. 6. Response of DSPW to 1–25 kHz triangular frequency sweeps of periods 20, 10, 5, and 2 ms. The signals were sampled at 125 kHz and spectrograms were generated with 64-sample cFFT. The lower panel shows the voltage signal used to generate the 1–25 kHz triangular sweep signals. Fig. 8. Response of the DSPW to (a) 1–10 kHz, (b) and (c) 1–20 kHz triangular sweeps of 500-ms period. The lower window shows the voltage signal used to generate triangular frequency sweeps. The signals were sampled at 31.25 ksps and the respective spectrograms were generated with 256-sample cFFT.
spectrum analyzers have poor temporal resolution [1]. The DSPW was specifically designed to address these issues. A. Evaluation of Performance of DSPW
0
Fig. 7. Response of the DSPW to 62.5–62.5-kHz sawtooth sweep signal with a period of 100 ms. The signal was sampled at 125 ksps and a 256-sample cFFT was used to generate the spectrogram. The spectrogram shows smooth transition at zero frequency.
directional transition from negative ( 62.5 kHz) to positive (62.5 kHz) frequency (velocity) as shown in Fig. 7. Shown in Fig. 8(a) is the response of the DSPW to triangular sweeps of 1–10 kHz (500-ms period) sampled at 31.25 ksps. Here, the spectrogram is displayed in the range of 15.625 kHz. The spectrogram of the signal ranges from 1–10 kHz without any aliasing. The spectrogram of the triangular sweep of 1–20 kHz (500 ms) sampled at 31.25 ksps is shown in Fig. 8(b). Here, it can be seen that spectrogram is aliased when displayed in the 15.625 kHz range. However, the aliased signal is unwrapped when the spectrogram is displayed in the 0–31.25 kHz range as shown in Fig. 8(c). IV. DISCUSSION Clinical ultrasound systems cannot be used to accurately measure blood flow velocities in mice because these systems typically operate at low frequencies resulting in poor spatial resolution. The clinical spectrum analyzers are designed for humans whose heart rate is a tenth of that of mice. Therefore, the mouse flow velocity spectrograms obtained with clinical
The effects of cFFT sample size on the temporal width and frequency resolution of the spectrogram are shown in Fig. 5. The temporal width of the spectrogram is always larger than the width of an input tone burst. This is because the spectrogram is obtained using 64–1024 sample cFFT calculation. For the example in Fig. 5 (10-ms 2-kHz tone burst input sampled at 125 ksps) the width of a 64-sample cFFT window was 0.51 ms resulting in a 10.51-ms-wide spectrogram. Similarly, a 1024sample cFFT (8.19 ms) would result in an 18.19-ms-wide spectrogram. The frequency resolution is high (122 Hz) with a 1024sample cFFT and poor (1953 Hz) with a 64-sample cFFT. Thus, for a given sampling rate an appropriate cFFT sample size can be chosen to obtain an optimal spectrogram. Also, since the effect of cFFT sample size on the temporal width can be quantified, corrections/adjustments can be made when necessary. Comparatively, the MSA (the adapted clinical system previously used by us) has a maximum sampling rate of 38.4 ksps. At this rate a 256-sample FFT results in a frequency resolution of 150 Hz. Since the FFT is calculated every 8 ms [3], [5] the temporal width of the spectrogram would be at least 8 ms more than that of the input signal. The clinical system Interspec Apogee X-200 ultrasonograph (Interspec-ATL) used in mice has a maximum sampling rate of 35.7 kHz and minimum computation time of 1 ms [14]. Broberg et al., [15] reported that the minimum temporal resolution of Doppler using the Acuson Sequoia 256 (Siemens Medical) was 2 ms perhaps indicating that FFT is calculated every 2 ms. Other clinical systems used to measure blood velocities in mice have maximum sweep speed of 1-s full-scale which suggests that the sample size is on the order of 2–10 ms. The timing of cardiac events such as isovolumic contraction and relaxation phases can be small (less than 10 ms). Responses in Fig. 6 show that the DSPW can measure such small temporal events. The DSPW generated a discernible spectrogram of the 0–20 kHz sweep signal with a period as small as 2 ms. The spectrograms shown were displayed at the fastest sweep speed
1768
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 10, OCTOBER 2005
Fig. 9. Response of the DSPW to (a) 1–25 kHz sweep signal with a 4-ms period and, (b) mouse aortic flow velocity signal distal to aortic stenosis with high-frequency fluctuations of about 4-ms periods. The signal was obtained with a 20-MHz esophageal pulsed Doppler probe (sampling rate 125 ksps, 256-sample cFFT) was used to generate the spectrogram. ((b) adapted with permission from ILAR journal, 43(3):147–158, 2002, Institute for Laboratory Animal Research, National Academies, 500 Fifth Street NW, Washington DC 20001 (www.national-academies.org/ilar)).
(10 pixels/ms or 100-ms full scale with a cFFT calculated every 0.1 ms). However, even at lower sweep speeds (5 pixels/ms or cFFT calculated every 0.2 ms and 2 pixels/ms or cFFT calculated every 0.5 ms) we were able to measure the time period of the 2-ms signal from the spectrogram. Also, it can be seen that the spectrograms follow the input voltage sweep signals demonstrating the linearity of the processing. On the other hand, the fastest sweep speed in the MSA is 1-s full-scale and with 8-ms FFT spectrogram calculations of the cardiac Doppler signals, the measurements of isovolumic contraction and relaxation phases would be suspect [3]. The fastest sweep speeds in clinical Doppler systems used in mice are limited to 1-s fullscale or 200 mm/s. This is not sufficient to measure cardiac timing in mice unless the heart rate is significantly slowed by using anesthetics as reported in several studies using clinical Doppler systems [14], [16]–[19]. Yang et al., have shown that the timing of cardiac events in the ascending aortic velocity signal in awake mice ( 658 beats/min) is not readily discernible compared to that in mice ( 293 beats/min) anesthetized by ketamine (150 g/g) xylazine (150 g/g) [20]. The DSPW is capable of resolving the high-frequency fluctuations of the order of 250 Hz in its spectral peaks. This is demonstrated in its response to the 1–20 kHz triangular sweep with a 4-ms period as shown in Fig. 9(a) and its capability in detecting such fluctuations (vortex shedding frequencies) in the peak aortic flow velocity signal distal to the aortic stenosis in mice [Fig. 9(b)]. The detection of such signals is possible only at high sweep speeds, high sampling rates, small FFT-sample sizes, and high temporal resolutions as afforded by the DSPW. The issue of directional transition has been addressed with the response of the DSPW to an input sawtooth sweep signal varying from 62.5 to 62.5 kHz. The spectrogram in Fig. 7 shows smooth transition across zero frequency from negative ( 62.5 kHz) to positive (62.5) frequency (velocity). This is important when measuring flow velocities which change direction during the cardiac cycle as illustrated in Fig. 10 and in cases where regurgitative flows occur due to abnormal valves. Also,
Fig. 10. Mitral inflow velocity (upward) and aortic outflow velocity (downward) signals obtained simultaneously in a mouse using a 10-MHz pulsed Doppler probe. The transition across zero is most apparent at valve clicks (arrows). Also shown is the ECG of the mouse. The signal was sampled at 125 ksps and a 512-sample cFFT was used to generate the spectrogram.
Fig. 11. Jet flow velocity across a transverse aortic stenosis in a mouse showing Doppler shifts as high as 115 kHz. This signal was obtained using a 20-MHz pulsed Doppler probe and sampled at 125 ksps. A 256-cFFT was used to generate the spectrogram. Also shown is the mouse ECG.
the transition is smooth across the Nyquist frequency (sampling rate/2) as shown in Figs. 4 and 11. This aspect of the DSPW is comparable to the MSA and other clinical Doppler systems. We have previously shown that under certain conditions it is possible to resolve frequency aliases in pulsed Doppler velocimeters beyond the normally accepted Nyquist sampling limit [11]. This feature has been incorporated into the DSPW and its capability of resolving frequency aliases has been demonstrated. The spectrogram [in Fig. 8(a)] generated by the DSPW in response to 1–10 kHz triangular sweep signal
REDDY et al.: PULSED DOPPLER SIGNAL PROCESSING FOR USE IN MICE: DESIGN AND EVALUATION
sampled at 31.25 kHz is normal (unaliased) because the input signal is sampled higher than the Nyquist sampling rate [21]. On the other hand, the spectrogram of a 1–20 kHz triangular frequency sweep signal sampled at 31.25 ksps is aliased [Fig. 8(b)] since this sampling rate is lower than the Nyquist sampling rate. However, the aliasing is unwrapped when the spectrogram is displayed in the 0–31.25 kHz range [Fig. 8(c)]. Similarly signals can be sampled at 125 ksps and processed without aliasing when they contain frequencies that range up to 125 kHz. An example of a high-frequency flow velocity signal obtained from the site of an aortic stenosis in a mouse is shown in Fig. 10 [6], [7]. This signal contains Doppler frequency shifts as high as 115 kHz and is comparable to the ( 10 to 115 kHz) signal shown in Fig. 4. Thus, signals containing frequencies up to the sampling rate can be processed and displayed without aliasing. This is a standard feature in the MSA and all clinical Doppler spectrum analyzers. B. Limitations and Recommendations Time and frequency resolutions are inversely related, that is, improvement of time resolution causes the frequency resolution to diminish, and vice-versa. We consistently use a 125 ksps sampling rate with a 256-sample cFFT calculation to obtain the velocity spectrograms. Another limitation is that the temporal width of the spectrogram is increased (as shown in Fig. 5) with increasing cFFT sample size. Typically, the Doppler signals are processed such that the additional temporal width occurs on the trailing side of the signal. This additional width depends upon the sampling rate and FFT sample size and, therefore, can be determined and corrected for if necessary.
V. CONCLUSION The DSPW performed as per design specifications to achieve high temporal resolution, fast sweep speed, and high bandwidth. The system has a frequency response of 0–125 kHz and, therefore, capable of measuring blood velocities as high as 9.6/4.8 cm/s and accelerations as high as 15.0/7.5 m/s with a 10-/20-MHz pulsed Doppler with the ultrasound beam parallel to the flow. This allows measurement of high stenotic jet velocities. The DSPW can detect high-frequency fluctuations in spectral peaks hence enabling the detection of vortex shedding frequencies in disturbed flows. The fastest sweep speed in the DSPW is ten times that in clinical Doppler systems reflecting the fact that heart rate in mice is ten times that in humans. The spectrogram is recalculated whenever the settings of frequency (velocity) scale, cFFT sample size, and sweep speed are changed. Thus, the performance of DSPW is superior to clinical Doppler systems that are currently used in mice.
ACKNOWLEDGMENT The authors wish to thank S. Ngo, A. Rizvi, R. Ganta, E. Garza, and T. Wernli for providing assistance on the project. They also wish to thank J. Brooks for editorial review.
1769
REFERENCES [1] C. J. Hartley and J. S. Cole, “An ultrasonic pulsed Doppler blood system for measuring blood flow in small vessels,” J. Appl. Physiol., vol. 27, pp. 626–629, 1974. [2] C. J. Hartley, L. A. Latson, L. H. Michael, C. L. Seidel, R. L. Lewis, and M. L. Entman, “Doppler measurement of myocardial thickening with a single epicardial transducer,” Am. J. Physiol., vol. 245, pp. H1066–H1072, 1983. [3] G. E. Taffet, C. J. Hartley, X. Wen, T. T. Pham, L. H. Michael, and M. L. Entman, “Noninvasive indexes of cardiac systolic and diastolic function in hyperthyroid and senescent mouse,” Am. J. Physiol. Heart Circ. Physiol., vol. 270, pp. H2204–H2209, 1996. [4] G. E. Taffet, T. T. Pham, and C. J. Hartley, “The age-associated alterations in late diastolic function in mice are improved by caloric restriction,” J. Gerontol. A. Biol. Sci. Med. Sci., vol. 52, pp. B285–B290, 1997. [5] C. J. Hartley, L. H. Michael, and M. L. Entman, “Noninvasive measurement of ascending aortic blood velocity in mice,” Am. J. Physiol. Heart Circ. Physiol., vol. 268, pp. H499–H505, 1995. [6] C. J. Hartley, G. E. Taffet, A. K. Reddy, M. L. Entman, and L. H. Michael, “Noninvasive cardiovascular phenotyping in mice,” Inst. Lab. Anim. Res., vol. 43, pp. 147–158, 2002. [7] Y.-H. Li, A. K. Reddy, G. E. Taffet, L. H. Michael, M. L. Entman, and C. J. Hartley, “Doppler evaluation of peripheral vascular adaptations to transverse aortic banding in mice,” Ultrasound Med. Biol., vol. 29, pp. 1281–1289, 2003. [8] C. J. Hartley, G. E. Taffet, L. H. Michael, T. T. Pham, and M. L. Entman, “Noninvasive determination of pulse-wave velocity in mice,” Am. J. Physiol. Heart Circ. Physiol., vol. 273, pp. H494–H500, 1997. [9] A. K. Reddy, G. E. Taffet, Y.-H. Li, S.-W. Lim, T. T. Pham, J. S. Pocius, M. L. Entman, L. H. Michael, and C. J. Hartley, “Pulsed Doppler signal processing for use in mice: Applications,” IEEE Trans. Biomed. Eng., vol. 52, no. 10, pp. 1771–1783, Oct. 2005. [10] N. Aydin and D. H. Evans, “Implementation of directional Doppler techniques using a digital signal processor,” Med. Biol. Eng. Comput., vol. 32, pp. S157–S164, 1994. [11] C. J. Hartley, “Resolution of frequency aliases in the ultrasonic pulsed Doppler velocimeters,” IEEE Trans. Biomed. Eng., vol. BME-28, pp. 69–75, 1981. [12] R. C. Gonzalez and R. E. Woods, Digital Image Processing, 2nd ed. Reading, MA: Addison-Wesley, 1992, ch. 4. [13] D. H. Evans and W. N. McDicken, Doppler Ultrasound: Physics, Instrumentation, and Signal Processing, 2nd ed. New York: Wiley, 2000, ch. 8, p. p. 180, 103. [14] B. D. Hoit, S. F. Khoury, E. V. Kranias, N. Ball, and R. A. Walsh, “In vivo echocardiographic detection of enhanced left ventricular function in gene-targeted mice with phospholamban deficiency,” Circ. Res., vol. 77, pp. 632–637, 1995. [15] C. S. Broberg, G. A. Pantely, B. J. Barber, G. K. Mack, K. Lee, T. Thigpen, L. E. Davis, D. Sahn, and R. Hohimer, “Validation of the myocardial performance index by echocardiography in mice: A noninvasive measure of left ventricular function,” J. Am. Soc. Echocardiogr., vol. 16, pp. 814–823, 2003. [16] A. Fard, C. Y. Wang, S. Takuma, H. A. Skopicki, D. J. Pinsky, M. R. Di Tullio, and S. Homma, “Noninvasive assessment and necropsy validation of changes in left ventricular mass in ascending aortic banded mice,” J. Am. Soc. Echocardiogr., vol. 13, pp. 582–587, 2000. [17] R. D. Patten, M. J. Aronovitz, P. Bridgman, and N. G. Pandian, “Use of pulse wave and color flow Doppler echocardiography in mouse models of human disease,” J. Am. Soc. Echocardiogr., vol. 15, pp. 708–714, 2002. [18] C. Pollick, S. L. Hale, and R. A. Kloner, “Echocardiographic and cardiac Doppler assessment of mice,” J. Am. Soc. Echocardiogr., vol. 8, pp. 602–610, 1995. [19] A. Schafer, G. Klein, B. Brand, P. Lippolt, H. Drexler, and G. P. Meyer, “Evaluation of left ventricular diastolic function by pulsed Doppler tissue imaging in mice,” J. Am. Soc. Echocardiogr., vol. 16, pp. 1144–1149, 2003. [20] X.-P. Yang, Y.-H. Liu, N.-E. Rhaleb, N. Kurihara, H. E. Kim, and O. A. Carretero, “Echocardiographic assessment of cardiac function in conscious and anesthetized mice,” Am. J. Physiol. Heart Circ. Physiol., vol. 277, pp. H1967–H1974, 1999. [21] A. V. Oppenheim and R. W. Schafer, Discrete-Time Signal Processing, 1st ed. Englewood, NJ: Prentice-Hall., 1989.
1770
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 10, OCTOBER 2005
Anilkumar K. Reddy (S’93–M’97) was born in Kalwakole, Andhra Pradesh, India, in 1963. He received the B.E. degree in electronics and communication engineering from Osmania University (Chaitanya Bharathi Institute of Technology), Hyderabad, India, in 1985, the M.S. degree in biomedical engineering from The University of Akron, Akron, OH, in 1991, and the Ph.D. degree in bioengineering from Texas A&M University, College Station, in 1996. From 1997 to 2001, he was a Postdoctoral Fellow in the Section of Cardiovascular Sciences in the Department of Medicine, Baylor College of Medicine, Houston, TX, as a Postdoctoral Associate, and worked on the development of ultrasound instrumentation for small animal research. Since 2001, has been a member of the faculty and is currently Assistant Professor of Medicine. His past experience includes working as a Teaching and a Research Assistant, an Instructor, and a Research/Design Engineer. His interests include development of instrumentation for use in mice and the study the cardiovascular system of normal, disease, and transgenic models of mice. His past work involved analysis of physiological signals. Dr. Reddy is a member of American Physiological Society, Houston Society for Engineering in Medicine and Biology, and since 2002 has served on a National Institutes of Health (NIH) study section that reviews Small Business Grants. He is an ad hoc reviewer for IEEE TBME, ABME, and AJP journals. He was an invited speaker at the 3rd International Congress on Cardiovascular Disease in Taiwan in 2004. He is a recipient of a Research Career Development Award (2005–10) from the NIH.
Alan D. Jones has received the B.S.E.E. degree from Rice University, Houston, TX, in 1977. Since 1977 he worked as a Programmer and Designer, Manager of Engineering, Manager of Research and Development, and Senior Design Engineer. From 1983 to 1989, he was Vice President of Don Quixote Windfarms, Inc. As a Senior Design Engineer at Indus Instruments from 1996 to 1998 has worked on the design of real-time processing, display, storage, and analysis of Doppler ultrasound signals obtained from mice and other small animals. In 1998, he started a company, Farpoint Software, Houston, and since then has been an independent consultant for several companies including Indus Instruments.
Christian Martono received the B.S. degree in electrical and computer engineering from University of Houston, Houston, TX, in 1996. He worked as a Hardware/Software Design Engineer at Indus Instruments, Houston, TX, from 1997 to 2001 and during this period worked extensively on the development of Ultrasound Instrumentation for Small Animal Research. From 2001 to 2003, he worked as an Embedded Software Engineer at Applied Science Fiction, Inc., Austin, TX, and since 2003 has been working as a Staff Software Engineer at National Instruments, Austin.
Walter A. Caro received the B.S.E.E. degree from Rice University, Houston, TX, in 1999. From 1999 to 2001, he worked as an Internet Applications Developer specializing in enterprise business integration software. He has been with Indus Instruments since 2001. His interests include real-time acquisition firmware development and DSP architecture.
Sridhar Madala (S’83–M’88) received the B.S.E.E. degree from the Indian Institute of Technology, Madras, India, in 1982, and the M.S. and Ph.D. degrees in electrical and computer engineering from Rice University, Houston, TX, in 1985 and 1988, respectively. He was Vice President (1988–1990) and President (1991–1992) of Coherent Systems, Inc., Houston, and since 1992 has been the President/Owner of Indus Instruments, Houston. He had served on a National Institutes of Health (NIH) study section that reviews Small Business Grants. He had been a principle investigator of NIH Phase I and Phase II SBIR grants and is currently a principle investigator on an NIH phase I STTR grant. Dr. Madala is a member Houston Society for Engineering in Medicine and Biology and of several societies of IEEE. He is a co-holder of a US patent for “Pipeline Mandrel Positioning Control System.”
Craig J. Hartley (S’64–M’66–SM’88) received the B.S.E.E. and Ph.D. degrees from the University of Washington, Seattle, in 1966 and 1970, respectively. From 1970 to 1973, he was a Postdoctoral Fellow in Bioengineering at Rice University, Houston, TX. Since 1973, he has been with Baylor College of Medicine, Houston, where he is currently a Professor of Medicine in the section of Cardiovascular Sciences. He is also an adjunct Professor of Bioengineering at Rice University and Professor of Electrical Engineering at the University of Houston. Since 1968, he has been active in the development of ultrasonic methods to measure blood flow and cardiovascular function in man and in animal models of human diseases. He is principle investigator on several research grants and has received a Research Career Development Award and a MERIT award from the National Institutes of Health (NIH). Dr. Hartley is a member of the American Institute of Ultrasound in Medicine, the American Physiological Society, the Cardiovascular Systems Dynamic Society, and since 1993 has served on an NIH study section that reviews Small Business Grants. In 1993, he received the Laufman Prize for career achievement from the Association for the Advancement of Medical Instrumentation, and in 1998 became an AIMBE fellow. He is the regional representative to the EMBS administrative committee, was treasurer of the EMBS-BMES 2002 Joint Conference in Houston, and is currently chair of the EMBS Awards Committee.