A Doppler signal converter has been developed to facilitate cardiovascular and exercise physiology research. This device directly converts audio signals from a clinical Doppler ultrasound imaging system into a real-time analog signal that accurately represents blood flow velocity and is easily recorded by any standard data acquisition system. This real-time flow velocity signal, when simultaneously recorded with other physiological signals of interest, permits the observation of transient flow response to experimental interventions in a manner not possible when using standard Doppler imaging devices. This converted flow velocity signal also permits a more robust and less subjective analysis of data in a fraction of the time required by previous analytic methods. This signal converter provides this capability inexpensively and requires no modification of either the imaging or data acquisition system.
- audio signal processing
- flow measurement
- analog circuit
- frequency conversion
investigators studying cardiovascular physiology have long been interested in measuring arterial blood flow. Initially, this was done using strain-gauge plethysmography, a technique limited to intermittently measuring flow in distal extremities (5, 28, 30–3215, 16, 19, 23, 27, 29). While nonimaging devices can measure flow velocity in vessels inaccessible to strain-gauge plethysmography, they cannot measure arterial diameter, an essential element in calculating arterial flow (20). An Interspec XL B-mode imaging system (Interspec, Conshohocken, PA) was subsequently acquired to obtain these diameter measurements (27, 29). At specified times during a study, the sonographer removed the nonimaging probe and applied the Interspec probe to image the artery. The repeated interchange of these probes required considerable skill to achieve a successful study and also raised the question of whether both probes consistently interrogated the same arterial site. In addition, since both nonimaging and imaging devices must be immediately available, one must provide space for both in a crowded clinical physiology laboratory.
A more desirable approach is to measure both arterial diameter and blood velocity with one machine without moving the probe from the selected arterial site. Toward this end, HDI 5000 Doppler ultrasound systems (ATL Ultrasound, Bothell, WA) were acquired as the standard imaging platforms used at our General Clinical Research Center (15, 16, 21–23). These machines provide excellent vascular images for selecting arterial sites and measuring arterial diameters and, in Doppler mode, can record flow velocity waveforms as video images. By the incorporation of both modalities, the HDI 5000 appeared to solve our blood flow measurement problem, but major operational issues emerged involving the intermittent character of data acquisition during long study paradigms and procedural bottlenecks in data management and analysis.
Operational issues aside, the primary limitation of measuring blood flow velocity with a standard Doppler imaging system such as the HDI 5000 is that it does not deliver a real-time velocity signal that can be simultaneously recorded with other physiological parameters. This seriously impedes an investigator's ability to continuously observe flow in response to infusions, exercise, or other experimental interventions.
Although the HDI 5000 lacks an analog flow velocity output, it does port two analog Doppler signals to a pair of RCA phono jacks (2
Designing the Doppler signal converter began with characterizing its input signals. The investigator verified that one audio output encodes for flow toward the transducer (positive velocity) and the other flow away from the transducer (negative velocity) by rapidly cycling an ultrasound probe by hand toward and away from the skin while listening to the Doppler signals through stereo headphones. The resulting sound abruptly shifted from one channel to the other as the probe changed direction.
It was initially thought that Chart's data processing functions might implement a software-based Doppler signal converter requiring no hardware except cabling. This approach was abandoned, however, when FFT analysis of Doppler audio signals from brachial artery flow in a resting subject clearly showed frequencies exceeding the 2-kHz limit of Chart's frequency-to-voltage (F/V) conversion function (1), the crucial element in this concept.
The project focus then shifted to designing a hardware signal converter (Fig. 1) with two channels, each using a zero-crossing detector, a pulse conditioning circuit, and an F/V converter to process its Doppler signal (6, 23a, 36). The resulting positive and negative velocity signals would then be combined to produce the desired composite flow velocity signal.
The HDI 5000 has recorded peak arterial flow velocities of ∼300 cm/s that correspond to Doppler frequency shifts of ∼12 kHz when using a maximum probe frequency of 6 MHz and insonation angle of 60°. Since the HDI 5000 also uses a high-pass filter with a corner frequency (fc) of 200 Hz to remove the artifact produced by low velocity arterial wall motion, the initial design specified a bandwidth of 200 Hz–15 kHz.
The converter's integrated circuits (ICs) include the following: LM358 op-amps for input and output signal conditioning (23b), LM393 comparators for signal detection (33), CD4013 flip-flops for pulse conditioning (8), CD4052 data selectors for signal switching (9), and LM2907 F/V converters for converting Doppler audio signals into flow velocity waveforms (23a). Additional ICs low-pass (LP) filter the Doppler signals and generate calibration signals as described in Testing and Refinement. The single-voltage operation of these ICs simplified selecting a power supply and facilitated packaging a compact unit (18 × 14 × 4 cm). To simplify the converter's wiring and reduce system noise, the signals were routed onboard by data selectors controlled by front panel switches (5a). The overall circuit was designed to optimize the performance of the F/V converters, the “linchpin” ICs in the converter.
Doppler signal processing.
Figure 2 diagrams the signal processing performed by the converter. An RCA-type stereo patch cable delivers two Doppler audio signals from the HDI 5000 to the input signal processors. Each input processor incorporates input protection, level shifting, and bandpass filtering. A pair of diodes limits signal amplitude to protect the input circuitry (7). Essential to a single-voltage analog circuit, level shifting establishes the signal's reference voltage (23b). The bandpass filter incorporates high-pass filtering to remove direct current signals that could saturate the input amplifier and LP filtering to limit high-frequency interference (35). Both Doppler signals pass through a data selector that, if needed, switches each signal to the other's pathway to invert the output flow velocity signal. A variable gain amplifier determines each signal's detection threshold. Each Doppler signal passes through a programmable LP filter and then a comparator where it is converted from a sinusoid into a 0–5-V square wave. Another data selector steers either the two Doppler signals or a calibration signal to a pair of flip-flop ICs. Each flip-flop halves the frequency of its signal, producing the necessary square wave clock signal for its F/V converter. Each F/V converter transforms the instantaneous frequency of its clock signal into a linearly proportional voltage representing flow velocity. At this point, both flow velocity signals have the same polarity; so, to recover directional information, one signal is inverted and combined with the other via a summing circuit. This composite flow velocity signal is LP filtered (fc
Doppler frequency (fD) is typically related to flow velocity (u) by Eq. 1, where fP is the excitation frequency of the ultrasound probe, θ is the insonation angle, and c is the average velocity of sound in tissue (36). (1) Solving Eq. 1 for u yields Eq. 2, a more useful calibration tool. (2) Since c = 1,540 m/s, a value first determined by Ludwig (17) and adopted as a standard by ultrasound device manufacturers, and fP and θ are determined and displayed by the HDI 5000, assigning a known calibration frequency (fcal) to fD establishes the relationship between fD and u.
The converter provides a 6-, 10-, or 15-kHz calibration signal that represents the Doppler frequency shift associated with a flow velocity of ∼380 cm/s for probes operating at 2.5, 4.0, or 6.0 MHz, respectively, at a maximum θ of 60°. A data selector directs the selected fcal
With the use of Eq. 2, where fD = fcal, an Excel spreadsheet (Microsoft, Redmond, WA) calculated u for θ between 0° and 60° in 1° increments for these three probe frequencies and presented u, fP, and θ in a lookup table. During a study, when a satisfactory arterial site has been located, the HDI 5000 displays fP and θ, which are then used to select the associated value of u from the table. Chart's two-point calibration function uses this flow velocity to rescale the recorded calibration signal and subsequent flow velocity signals, thus providing direct velocity measurements from the recorded waveforms.
Testing and Refinement
Even when a nonaliasing velocity range was selected, however, low or even negative velocity spikes continued to appear within systolic peak flow signals. These artifacts were attributed to the reflection of rapidly transmitted systolic pressure waves from downstream arterial structures such as bifurcations, stenoses, or vessel tapering (4). This idea was supported by a study where a SphygmaCor arterial tonometry system (AtCor Medical, Sydney, Australia) measured the augmentation component of systolic pressure caused by a reflected pulse wave that arrived just after peak systolic pressure, i.e., at the same time as low or negative velocity artifacts appeared in the systolic flow velocity signal (18).
After Chart's digital LP filter (fc = 1.0 kHz) was applied to the negative velocity Doppler signal, FFT analysis showed that the 4-kHz artifact was removed without affecting the diastolic relaxation signal.
Following this experiment, a two-channel Krohn-Hite variable LP filter was inserted between the HDI 5000 and the converter and each filter's fc was initially set at 10 kHz (Krohn-Hite, Cambridge, MA). During subsequent brachial and femoral flow studies, reflection artifacts appearing on the flow velocity signals could be reproducibly observed or removed by switching fc of the negative velocity filter between 10 and 1.0 kHz. These studies also showed the need to scale fc of both filters in proportion to the Doppler probe frequency to ensure that only the reflection artifact was removed from the negative flow velocity signal. Filter fc values were then defined for each probe frequency: 15 and 1.5 kHz at 6 MHz, 10 and 1.0 kHz at 4 MHz, and 6 and 0.6 kHz at 2.5 MHz.
Although it performed well, this variable filter-converter configuration was cumbersome to use, so the converter circuitry was redesigned to incorporate two programmable MAX293 LP filter ICs (20). A square-wave clock signal sets each filter's fc where the ratio of clock frequency (fclock) to fc is 100:1, so one simply changes fclock to change fc. A 6.0-MHz oscillator generates the master clock signal used to derive both filters' clock signals and the calibration signal (Fig. 2) (10). A switch-programmed 74HC161 “divide by N” IC initially divides the oscillator's frequency by 2, 3, or 5 (11). A flip-flop halves this frequency, producing the 1.5, 1.0, or 0.6 MHz clock signal, fLPhigh, that sets fc of the positive velocity channel's LP filter to 15, 10, or 6 kHz. A CD4017 IC divides fLPhigh by 10, concurrently producing a 150, 100, or 60 kHz clock signal, fLPlow, that sets fc of the negative velocity channel's LP filter to 1.5, 1.0, or 0.6 kHz. A second CD4017 divides fLPlow by 10 to yield the 15, 10, or 6 kHz calibration signal. The circuit schematic is available with the online version of this article.
In Vitro Study
The in vitro study employed a mock circulatory loop adapted from a system developed by Tschakovsky and colleagues (26). A dilute solution of cornstarch, by providing scattering particles that ultrasonically resemble red blood cells, served as the “blood” in this system. This solution, typically filling a 3-liter reservoir, was magnetically stirred to maintain the cornstarch in suspension. A Master-Flex roller pump (Masterflex, Vernon Hills, IL) circulated this solution to provide quasi-steady flow within the loop. A test chamber was fabricated from a transparent rectangular plastic container where lengths of silastic tubing (internal diameter = 2.8–8.9 mm) were passed through the container's sides and across the container to simulate arteries. Water was added to the test chamber to cover the tubing to a depth of 5 cm. The solution was drawn from the reservoir through a loop of Tygon tubing and pumped through a compliance chamber to filter out high-frequency pulses from the pump rollers just before passing through the silastic “artery” and returning to the reservoir.
Doppler video clips were loaded into the HDI 5000, and a video clip was selected and displayed. The operator identified a region of interest on the flow velocity waveform and manually traced it using a trackball. A program within the HDI 5000 automatically calculated and displayed peak, time-averaged peak, and TAM velocity for the traced waveform, and the operator manually logged the TAM velocity. This process was repeated for each clip recorded during the study.
The annotated timing marks were used to identify each flow velocity data segment coincident with an HDI 5000 video clip. The cursor within Chart was used to select all data within that segment. Chart then computed the mean (TAM) velocity of the selected data and automatically logged it into the Data Pad, a user-configured table within Chart that was saved as a text file. This was repeated for all three runs at each flow rate within each experiment.
In Vivo Study
The converter was evaluated during a clinical research study that measured brachial artery flow velocity over a wide range of flow rates in 10 normal volunteers (5 men and 5 women). Each volunteer provided written, informed consent. This protocol was reviewed and approved by the Institutional Review Board within the Penn State College of Medicine.
All 13 Doppler video clips from a selected arm elevation trial within a given experiment were retrieved, and a clip was selected and displayed. Guided by the ECG, the analyst located the first complete cardiac cycle and traced its flow velocity signal. As previously described, peak, time-averaged peak, and TAM velocities for that cycle were automatically computed and displayed and TAM velocity was manually logged. This process was repeated for successive cardiac cycles during the clip. The average of all TAM velocities measured within the clip was computed and reported as the representative TAM velocity for that clip (time point). This analysis was performed on the remaining 12 video clips recorded within the selected arm elevation trial and repeated for the other two trials.
The flow velocity waveform from a selected arm elevation trial was displayed. The analyst then used the annotated timing marks to locate flow velocity data segments coincident with the associated HDI 5000 video clips. Within each velocity data segment, the region spanning the largest number of complete cardiac cycles was selected. Chart then computed the mean (TAM) value of the selected velocity data and logged it into its Data Pad, a user-configured table. This procedure was repeated on the velocity data associated with each remaining time point within the selected trial. All three arm elevation trials were processed in like manner, and when all TAM values were determined, the Data Pad was saved as a text file for further processing by Excel.
The scatterplots in Fig. 3 visually compare the TAM flow velocities from the two devices for both the in vitro (Fig. 3A) and in vivo (Fig. 3B) validation studies. While each scatterplot closely approaches its ideal “y = x” line, the high correlation alone suggests but does not signify agreement, so these two data sets were examined using the Bland-Altman procedure (3). The mean difference line within each Bland-Altman plot in Fig. 4 shows minimal deviation from the ideal “zero difference” line, and the small range of the 95% limits of agreement support the claim of agreement between these two measurement systems. These data sets were also tested with the concordance correlation method developed by Lin (12) since this metric quantifies both precision and accuracy when comparing measurement systems. The resulting concordance correlation coefficients for the in vitro data, 0.997 (95% confidence interval: 0.996 and 0.997, P < 0.001) and in vivo data, 0.992 (95% confidence interval: 0.990 and 0.993, P < 0.001), confirm the Bland-Altman results and further support the claim that flow velocity signals produced by the Doppler signal converter are comparable with those produced by the HDI 5000 and can be reliably used to measure flow velocity.
Historically, to measure blood flow noninvasively, clinical investigators have had to accommodate their experimental designs to the limitations of the available technology such as strain-gauge plethysmography and nonimaging Doppler ultrasound. It is not surprising, therefore, that Doppler ultrasound imaging technology has revolutionized the noninvasive assessment of blood flow by incorporating high-quality vascular imaging for estimating lumenal area and accurate flow velocity measurements within the same machine. Present Doppler imaging systems efficiently gather, analyze, and report clinical data, typically by using a menu of preconfigured and optimized study protocols. However, these protocols may not be readily applied to clinical research studies where biomedical signals from a variety of instruments are recorded simultaneously to observe the body's response to an experimental intervention. These machines provide limited capacity to incorporate additional signals, and data management is geared toward archiving images and generating standardized reports rather than providing Doppler information in a format easily integrated into a multichannel data stream. While older Doppler imaging systems, e.g., Flo-Map (Cardiometrics, Mountain View, CA), provide an analog flow velocity signal for external recording, using such legacy devices for clinical research may not be a viable solution. Assuming working devices are even available, their imaging technology is much less capable than that of newer digital systems.
Given the excellent imaging and flow measurement capabilities of a system such as the HDI 5000, why develop a Doppler signal converter? To address this, consider how a standard HDI 5000 is typically used in clinical research. Initially, it is operated in imaging mode to locate an arterial site of interest. During study paradigms lasting ≥30 min, while ECG, blood pressure, and other signals are continuously monitored, the HDI 5000 is switched between Doppler mode to record flow velocity waveforms and imaging mode to record arterial images for diameter measurements. These waveforms and images are saved as short (<20 s) video clips. During a long study paradigm, therefore, flow velocity is not continuously recorded but rather sporadically sampled, leaving gaps in the data. This forces the investigator to guess which are the most fruitful time points in the study to record these video clips and complicates the process of synchronizing flow velocity data with other biomedical signals. (21–23). Poststudy, the analyst measures flow velocity from individual video clips by tracing Doppler flow velocity waveforms beat by beat and manually logging the data, a subjective process requiring up to 10 h/study. Data analysis with the HDI 5000 is frequently delayed, however, by the need to archive video clips as soon as possible after each study to recover space on the HDI 5000's small 17-Gb hard disk before the next study. This archiving requires 2 to 3 h to transfer data from the hard disk to the magneto-optical disk to CD-R, and processing an archived study typically requires a day to reload and analyze the video data. Since several research groups share the HDI 5000, the mutually exclusive nature of the data acquisition and analysis frequently creates scheduling issues affecting the efficient use of these machines.
As the signal converter processes the audio output of a Doppler imaging system, its performance is necessarily dependent on the quality of that output; i.e., the stronger the Doppler signal, the clearer are both the video display and audio signals from the HDI 5000 and the more reliable the signal detection and processing by the converter. At present, since the HDI 5000 most easily and reliably detects Doppler signals originating from the brachial, femoral, and carotid arteries, the Doppler signal converter most easily and reliably converts these signals as well. Doppler signals from renal and coronary arteries are more technically challenging to detect and acquire by both the Doppler imaging system and signal converter. Continued efforts are needed to improve both the technique of acquiring Doppler signals from these arteries as well as the signal detection capability of the Doppler signal converter. Although the HDI 5000 was exclusively used during the development of the signal converter, the converter is currently transitioning to service with the newer Philips iE33 and Acuson Sequoia 512 systems (Siemens Medical Systems, Iselin, NJ).
This project was supported by the National Heart, Lung, and Blood Institute Grant P01-HL-077670.
No conflicts of interest are declared by the author(s).
We acknowledge and thank Dr. Lawrence Sinoway for encouragement and continued support of this project.
- Copyright © 2010 the American Physiological Society