A Novel Technique for Identification of Doppler Microembolic Signals Based on the Coincidence Method
In Vitro and In Vivo Evaluation
Background and Purpose The applicability of a novel differentiation technique in embolus detection based on the coincidence principle and using a multigate probe was evaluated in this study.
Methods According to the coincidence method, high-intensity transients should only be classified as microembolic signals if they appear sequentially in the two sample volumes monitored and within a defined time window calculated from the blood velocity and the spatial distance between the insonation depths. Part A: microbubbles were introduced in a continuous-flow bench model of the middle cerebral artery to evaluate the accuracy of the multigate probe in embolus detection. Part B: in the subjects and patients, the minimal and maximum time delays in the appearance of microembolic signals in the two middle cerebral artery sample volumes were calculated as 0.01 second and set at 0.1 second, respectively. The multigate probe was used to monitor (1) 5 normal volunteers in whom 1008 artifact signals were produced, (2) 2 patients undergoing aortic valve replacement surgery, and (3) 12 patients with potential cardiac or carotid embolic sources.
Results In the bench model, 95.5% of microembolic signals produced by microbubbles appeared in the two sample volumes with a time delay between 0.02 and 0.05 second, while in the remaining 4.5% a shorter passage time of 0.01 second was measured. A total of 1968 high-intensity signals were recorded in subjects and patients. All but 20 of these (99%) appeared in both monitoring channels within the above time frame. To summarize, 996 (98.8%) of the 1004 artifact signals and 943 (98.1%) of the 961 microembolic signals were correctly classified.
Conclusions Application of the coincidence theory to distinguish microembolic signals from artifacts provides a promising new technique with high sensitivity and specificity that could decisively improve the validity of embolus detection.
Since the detection of intracranial microembolic signals (MES) by means of transcranial Doppler ultrasonography (TCD), several methods have been proposed to improve the differentiation between true embolic signals and artifacts. These approaches were based on the audiovisual characteristics of MES,1 2 spectral analysis,3 automated detection software,4 or a trained neuronal network.5 While a differentiation between MES and artifacts with these methods was principally feasible, they could not provide definitive evidence on the nature of individual signals. In particular, signals overloading the system were equivocal, since the above techniques could no longer define them as a result of overload.6
The purpose of this study was to evaluate the feasibility of the differentiation between microembolic and artifact signals with the use of the coincidence method, originally described by Bothe and Kohlhörster.7 They used this approach to investigate cosmic radiation, ie, to distinguish real energetic particles from random artifacts. The occurrence of two or more events caused by the same physical quality (for example, an excited physical state, a beam of laser light, or a current fluid) strongly correlates with time. The occurrence of “true” events should therefore take place within a predefined time window.
To apply this approach in embolus detection, simultaneous monitoring of spatially separate segments of a vessel is required. “Multigate” monitoring of blood flow velocity within two or more spatially separated segments of the same vessel and its simultaneous analysis through separate Doppler channels was originally described as a method to assess blood velocity profiles.8 The use of a multigate device for embolus detection was first proposed by Rune Aaslid during the 8th International Symposium on Cerebral Hemodynamics in Münster, Germany, in September 1994. Passage of emboli through the artery under study would theoretically result in the detection of high-intensity signals in all channels in a certain sequence and with a certain time delay between them defined by the direction and flow velocity of the blood column and the distance between the separate sample volumes. In contrast, artifacts should appear in all channels simultaneously or in only one channel.
Subjects and Methods
The following theoretical considerations should facilitate the understanding of the technique used in this study: A current fluid transporting particles travels from a given point A to a point B within a time δt, which depends on its velocity and the distance between the two points. δt is also the minimal time required for a transported particle to cover the same distance. The maximum time required (δt′) depends on a number of parameters, including the flow profile of the fluid and the surface characteristics of the particle. Assuming that a particle is registered at point A at the time tA and another particle is registered at point B at the time tB, we regard these events as “coincidental” if the condition
is satisfied. In this case it is clear that the same particle was registered consecutively at both points. The restriction of the elapsing time after time tA (δt′) is essential to distinguish the particle under observation from random or subsequent particles.
The same principles apply for monitoring of the middle cerebral artery (MCA) for the detection of MES with TCD if two spatially separate segments A and B of the artery are insonated simultaneously and the data recorded on two separate Doppler channels (Fig 1⇓). The same current blood passes the sample volume A at a time tA and the sample volume B at a time tB. The distance between the sample volumes and the blood flow velocity determine the length of δt. Simultaneous recording of echoes on both channels allows the identification of artifact signals, since they are recorded at either time tA or tB on both channels, or on one channel only. Consequently, coincidental signals can be interpreted as being caused by true particles transported by the streaming blood.
The passage time δt was calculated from both the peak systolic blood flow velocity (Vs1-2) in the arterial segment under study and the spatial distance (s1-2) between the insonation depths 1 and 2 of the sample volumes according to the formula
where Vs1-2=Vs1+Vs2/2 and Vs1 and Vs2 are the peak systolic flow velocities at insonation depths 1 and 2, respectively. δt also reflects the minimal time required for an embolus to travel from depth 1 to depth 2. A longer time interval than δt may be necessary for this passage, since the route of the embolus through the artery is not predictable, eg, it is unclear whether it travels the shortest way in the middle of the blood column or occasionally bumps on the wall of the vessel, thus traveling a longer distance. In our paradigm, a strict limitation of the appropriate length of the time window (t+δt′) within which the emboli should have passed the arterial segment of interest is less important, since microemboli very rarely appear only milliseconds after each other. For this reason, the parameter t+δt′ was set at 0.1 second.
A prototype multigate probe was used (Eden Medical Electronics). This was a 2-MHz pulsed-wave probe consisting of one transmitting and two receiving channels. The burst duration of this probe was 13 microseconds (when the sample volume was set at a length of 10 mm), and the pulse repetition frequency was 6.25 kHz. Measurement of the blood flow velocities at both insonation depths was performed by splitting the reflected echoes and feeding them into separate receiving channels. The probe was used with a standard TCD device (Pioneer TC 4040, Eden Medical Electronics). Signal analysis was based on a 128-point fast Fourier transform. Data from both channels were recorded on digital audiotapes using a two-channel digital audiotape recorder (Sony 59ES), whereby only one of the quadrature audio signals was recorded per channel. The time of the appearance of MES and artifacts was documented based on the time display of the digital audiotape equipment. An experienced observer was present during all monitoring sessions.
In Vitro Studies
A bench model of the MCA was devised. This consisted of a continuous-flow diaphragm pump (PAR-MAX 3, ITT Jabsco), plastic tubing 3.5 mm in diameter, and a solution of cellulose with an average particle size of 20 μm (Sigmacell type 20, Sigma Chemical Co) mimicking human blood. The plastic tubing was inserted in a specially constructed chamber filled with ultrasonic gel. Rigid tubing was used, since the flexibility of the tube does not influence flow dynamics in a constant flow situation. The probe was mounted in the chamber in a position providing a 10° insonation angle of the simulated MCA. Since the power of the ultrasonic signal was not of interest in this setting, no shell imitating the temporal bone was intersected between the probe and the ultrasonic gel. Microbubbles were produced by agitating water in a syringe. Subsequently, the solution containing microbubbles was injected into the system through an 80-μm filter via a syringe through a side port. Before all experiments, the model was degassed for 30 minutes with the use of a vacuum pump (R Halm Elektromotorenfabrik). Fifteen minutes of tentative TCD monitoring was performed after degassing to ensure the absence of microbubbles in the circuit. All TCD settings were kept constant throughout the study (sample volume, 8 mm; gain, 14).
Two hundred separate microbubbles were introduced into the circulation of the model. MES were recorded from paired sampling depths of 50 and 68 mm and 50 and 62 mm, respectively, with a constant flow velocity of 60 cm/s. The calculated passage time for emboli was 0.03 for the first and 0.02 second for the second combination.
The MCA was identified according to generally accepted criteria9 and the probe subsequently fixed on the temporal skull with an elasticized band. Monitoring was performed at two different insonation depths with a constant distance of 10 mm between them, usually at 56 and 46 mm. The whole length of the main stem of the MCA was investigated in 2-mm steps to exclude the presence of significant velocity changes between the two sample volumes, eg, due to MCA stenosis or hyperperfusion. After obtaining a satisfactory signal, the sample volume was gradually reduced in both channels as far as possible provided that a sufficient background MCA signal could be maintained. The final sample volumes used varied between 6 and 10 mm in length.
Five healthy volunteers, 2 patients undergoing cardiac surgery for aortic valve replacement, and a total of 12 patients with potential cardiac or carotid embolic sources were used to investigate the feasibility of the new TCD technique. Each control subject underwent a 30-minute monitoring procedure. A total of 1008 artifact signals were evaluated; they occurred either naturally by proband movements, slight probe dislocation, accidental hitting of the probe by the proband, chewing, or speaking, or they were actively produced by the observer by lightly touching the probe or its fixation.
Two patients undergoing cardiac surgery for aortic valve replacement (Tecna, Baxter) were continuously monitored during the surgical procedure. A total of 862 MES were detected according to conventional criteria, ie, (1) short duration (<0.1 second for signals appearing in systole, <0.3 second for signals appearing in diastole), (2) characteristic sound, (3) random appearance in the cardiac cycle, and (4) a minimum intensity of 4 dB above the background MCA signal.2
Additionally, 12 patients expected to reveal MES in their MCAs (mean±SE age, 51±4 years; 9 men, 3 women) were recruited from the departments of cardiothoracic surgery (n=7) and neurology (n=5). Five of them had prosthetic cardiac valves and two a left ventricular assist device (Novacor, Baxter). Neurological patients had high-grade (>70%) symptomatic carotid artery stenosis diagnosed with extracranial duplex (n=5) and additional cerebral angiography (n=2). Clinical details and monitoring data of these patients are listed in the Table⇓. The overall monitoring time was 8 hours and 15 minutes. A total of 99 MES were identified using the four conventional criteria listed above.
Off-line evaluation of both artifacts and MES consisted of measurement of the exact time of occurrence of each signal within each channel. This was done with an accuracy of 0.01 second. The maximum blood velocity (vb) in centimeters per second was also calculated. Both parameters were calculated using standard software incorporated in the Doppler device. Subsequently, the time between the appearance of the signals in the two channels was evaluated. The smallest distinguishable time difference was 0.01 second. For the purposes of this study, the channel with the greater insonation depth was always coded as “channel 1” as opposed to the channel with the more shallow insonation depth, which was coded “channel 2.” Every true MCA embolus traveling orthogradely first had to pass channel 1 and then channel 2. The distance between the two sampling depths was 1 cm, and peak systolic flow velocities ranged from 47 to 85 cm/s. The longest time delay was calculated according to δt=1 cm/47 cm · s−1=0.021 s, and the shortest one according to δt=1 cm/85 cm · s−1=0.012 s. δt was therefore set at 0.01 second and the time window for the identification of MES between 0.01 and 0.1 second.
In Vitro Studies
All 200 microbubbles that were introduced into the model circulation produced MES in both monitoring channels. When the distance between the sample volumes was 18 mm, the calculated time delays were 0.02 second in 9 of 100 MES, 0.03 second in 50 of 100 MES, 0.04 second in 38 of 100 MES, and 0.05 second in 3 of 100 MES. When the distance was 12 mm, the calculated time delays were 0.02 second in 67 of 100 MES and 0.03 second in 33 of 100 MES. In summary, all artificially produced gaseous microemboli passed the sample volumes within the expected time window, while none of them occurred in one channel only or in both channels simultaneously.
Four (0.4%) of the 1008 artifact signals recorded in the normal volunteers appeared in only one channel (one in channel 1 and three in channel 2). Of the remaining 1004 signals, 992 appeared simultaneously in both channels and 12 with a time delay of 0.01 second. Thus, the specificity of this technique was quite high (98.7%).
Of the 862 intraoperatively detected MES, 15 (1.7%) were viewed in only one channel (13 appeared only in channel 1 and two only in channel 2). This could be explained by escape of microemboli via MCA branches originating from the MCA stem between the two sample volumes, a longer time delay due to “sticking” of the emboli on the arterial wall, or a failure to detect them in the second channel as a result of insufficient time overlap of the Doppler device. Of the remaining 847 signals, 845 (99.7%) fulfilled the characteristic criteria of emboli according to the coincidence method, ie, with respect to the time delay of their appearance in the two channels (Table⇑).
Similar results were obtained in the 12 patients with potential embolic sources: of 99 MES, 1 was viewed in only one channel (1%), while the remaining 98 MES (99%) were correctly characterized as such by the multigate approach.
To summarize the results of the three groups, the described method correctly identified 996 of 1008 artifact signals and 943 (845+98) of 961 MES (Fig 2⇓). Thus, the sensitivity of the new device (true positive/all positive) was 98.1% and the specificity (true negative/all negative) 98.8%.
Our results demonstrate that a differentiation between MES and artifacts using the coincidence method is feasible. The specificity and sensitivity of this method are superior to those of previously proposed techniques.4 5 Based on the coincidence principle, our method is able to individually characterize every signal. The development of refined software analyzing the time delay on-line would therefore allow an automated and reliable identification of MES and would decisively improve the cost-efficacy ratio of embolus detection.
The failure rate of 3.1% is possibly due in part to measurement error caused by imprecise placement of the cursors. This could be avoided with the application of a more sophisticated software and a better time resolution of the Doppler device. Presumably, a small failure rate due to signals viewed in only one channel cannot be avoided, but this was minimal in our setting and accounted for only 0.4% of the artifact and 1.4% of the MES.
The main limitation of this technique is that its applicability to individual patients has not yet been evaluated. Sampling of the MCA at two different depths spatially separated by a distance of approximately 10 mm was possible in all patients in our study. Larger studies with the multigate probe are needed to determine its applicability in other settings and to develop guidelines for further refinement. The current time resolution of the Doppler device allows application of this new technique only in patients with an MCA velocity lower than 100 cm/s. While this limitation is fulfilled by most individuals,9 it prohibits the use of this method in patients with MCA stenosis or cerebral hyperperfusion. Again, improvement of the temporal resolution of the Doppler equipment is required to extend its applicability to practically every clinical setting.
The distribution of the time delay in our model was sharper than that during the in vivo studies. This could be due to the homogeneity of size and composition of the embolic material in the in vitro study. Additionally, the interactions between microbubbles and in vitro circulating fluid are thought to be quantitatively less pronounced than those between blood and formed particles.
Most MES recorded in vivo in our study (585 of 961, 60.9%) appeared with a time delay of 0.02 to 0.03 second, which suggests that the bulk of microemboli traveled with a speed corresponding to the systolic blood velocity. The speed of embolic particles is mainly influenced by their size and their surface characteristics. We hypothesize that the distribution of different flowing velocities within the laminar layers of the blood column and the heterogeneous size of the particles considerably contribute to the broad scatter of passage times between the two sample volumes.
In conclusion, our results suggest that the application of the coincidence theory to distinguish MES from artifact signals provides a promising new technique that could decisively improve the validity of embolus detection.
The authors would like to thank Eden Medical Electronics for supplying a prototype of this novel probe and the Technical University of Cologne for providing technical support throughout this project.
- Received September 25, 1995.
- Revision received December 14, 1995.
- Accepted December 15, 1995.
- Copyright © 1996 by American Heart Association
Spencer MP. Detection of cerebral arterial emboli. In: Newell DW, Aaslid R. Transcranial Doppler. New York, NY: Raven Press; 1992:215-230.
Georgiadis D, Grosset DG, Kelman AW, Faichney A, Lees KR. Incidence and characteristics of intracranial microemboli signals in patients with different types of prosthetic cardiac valves. Stroke. 1994;25:587-592.
Brucher R, Russell D. Spectral characteristics of emboli with artefact suppression. Stroke. 1993;24:510. Abstract.
Markus HS, Loh A, Brown MM. Computerized detection of cerebral emboli and discrimination from artifact using Doppler ultrasound. Stroke. 1993;24:1667-1672.
Bothe W, Kohlhörster W. Das Wesen der Höhenstrahlung. Zeitschrift für Physik. 1929;29:751-777.
Hocks APG, Reneman RS, Peronnau PA. A multi-gate pulsed Doppler system with serial processing. IEEE Trans Sonics Ultrason SU. 1981;28:242-247.
Ringelstein EB, Otis SM, Spaar-Kahlscheuer B, Niggermeier E. Transcranial Doppler sonography: anatomical landmarks and normal velocity values. Ultrasound Med Biol. 1990;8:745-761.