Bigated Transcranial Doppler for the Detection of Clinically Silent Circulating Emboli in Normal Persons and Patients With Prosthetic Cardiac Valves
Background and Purpose Detection of clinically silent circulating microemboli by transcranial Doppler sonography is now being widely investigated in the hope of identifying patients at increased risk for stroke. Automatic detection by bigated Doppler, which uses sampling from two different depths in the artery under study and considers the motion of the embolus, may help to define “periods of interest” that can be evaluated off-line.
Methods In 12 normal volunteers and 10 patients with prosthetic aortic valves, we performed 1-hour recordings from one middle cerebral artery. In the normal subjects, we produced additional artifacts to use them as false-positives. Detection of microemboli was done off-line from digital audiotapes by an experienced blinded investigator (used as the gold standard) and was compared with on-line detection using specially designed software.
Results With the setting used, 91.5% of all recorded artifacts could correctly be identified as such with the software. Embolic signals were detected by the software with a specificity of 59.9% and a sensitivity of 74.3%.
Conclusions Bigated Doppler adds a new dimension to the definition and detection of microembolic signals. It constitutes an important step forward toward automatic screening of stroke-prone patients. Assessing on-line periods of interest during the recording and going over the recorded data again off-line helps to save time for the discrimination of embolic signals from both the normal Doppler spectrum background and artifacts.
Detection of clinically silent circulating microemboli by transcranial Doppler sonography is now being widely investigated in the hope of detecting patients at increased risk for stroke and taking preventive measures in those presenting with a striking number of embolic signals.1 2 3 4 5 With use of endarterectomy of high-grade ICA stenoses, both the number of embolic signals and the risk of stroke can be reduced.6 7 8 9 So far, however, this concomitant decrease is the only valid indication that the number of microembolic signals and the risk of stroke are causally related. A few further reports are also suggestive of such a relationship but need to be validated in larger studies.4 10 Circulating silent microemboli may be an indicator of an otherwise unnoted stroke risk, similar to transient ischemic attacks and amaurosis fugax attacks during sleep.
Embolus detection is very time consuming and laborious. Several approaches to automated embolus detection have been proposed. The use of a trained neural network is an attractive development for semiautomated embolus detection.6 11 The most promising development at present is the bigated transcranial Doppler approach, which takes into account the fact that an embolus moves proximally to distally inside the artery, thus appearing in two sample volumes sequentially, ie, with a certain time delay. This device traces the moving embolus at two different depths of the same artery and takes the time delay of its appearance as the crucial criterion. This is contrary to an artifact that affects the flow signal at both depths simultaneously. So far, three reports on bigated transcranial Doppler ultrasound and embolus detection using a different software have been published.12 13 14
Patients with mechanical prosthetic cardiac valves have been shown to have numerous embolic events, thus making them an ideal model to test a new embolus detection software.15 16 In this study, we evaluated the bigate technique in normal subjects and in patients with prosthetic cardiac valves and tested its reliability in rejecting artifacts.
Subjects and Methods
All subjects received a full color duplex investigation of their neck arteries (documented on videotape) and a continuous-wave Doppler investigation of the periorbital arteries. The subjects were also examined with transcranial Doppler ultrasound, including the intracranial segments of the ICA, the MCA, and the anterior and posterior cerebral arteries.
The MCA was insonated through the temporal window.17 A 2-MHz probe was secured in a head ribbon. A setting guaranteeing optimal embolus discrimination from the background spectrum was used with a small sample volume of 5 mm in length and a low gain; ultrasound intensity was 20 mW/cm2.18 This setting was maintained throughout the recordings. The investigations were well tolerated by the subjects without side effects or discomfort.
The Multi-Dop X transcranial pulsed Doppler ultrasound device (DWL) was used for all studies. It used a 64-point FFT and a graded color scale to display the intensity of the received Doppler signal. The Doppler signal was recorded onto a digital audiotape deck recorder (DTC-690; Sony Germany GmbH) with normal speed. The tapes (DM120; Maxell Europe Ltd) were given numbers and mixed with tapes from other recordings. No patient details could be deduced from these numbers. The blinded analysis was done off-line with use of a specially designed device to introduce the flow signals under study into the FFT processor. The experienced observer’s analysis of embolic signals comprised (1) listening to the signal through a headphone (MDR CD250; Sony Germany GmbH) and (2) watching the signal on the screen at highest speed. The following definition for embolic signals was used: typical visible and audible (click, chirp, whistle) short-duration high-intensity signal within the Doppler flow spectrum, occurring at random in the cardiac cycle.5 7 19 We also included bidirectional signals if they fulfilled the other criteria and if their maximum was within the spectrum. The code of the tapes was not broken before having completed the whole analysis.
In addition to the above-mentioned off-line analysis, the bigated embolus detection software TCD-8 for MDX, version 8.00 K, was used on-line. The distance between the two sample volumes was set at 5 mm. The relative intensity increase of events (microemboli and artifacts) relative to background noise was calculated by the machine from the deeper sample volume, ie, the one first passed by an embolus during its intra-arterial migration to the periphery. It should be noted that the decibel values of this software differed from those of other software. In contrast to the manual calculation of the relative intensity increase of the embolic signal compared with the adjacent background spectrum intensity, or compared with the background spectrum in the same position in an adjacent cardiac cycle, this algorithm uses the whole screen as a background. This background includes spectrum-free (“black”) areas of low intensity. Consequently, the decibel values of the microembolic signals detected during this latter approach are higher than the ones calculated in previous studies. Relying on recent unpublished observations of our group, we could readily detect spontaneous spotlike intensity fluctuations within the normal Doppler spectrum in artifact-free periods of normal control subjects with a relatively low detection threshold of 6 dB. Thus, a detection threshold of 6 dB was used in the present study to detect as many events as possible (high sensitivity) and to postpone the definition of an appropriate intensity cutoff level for microembolic signals. The software recorded all events at and above the preset threshold of 6 dB. It also indicated the calculated spatial distance between the two sample volumes (Fig 1⇓).
The software used the raw Doppler values before they were subjected to an FFT. An illustrative example is given in Fig 1⇑. In this patient with a mechanical prosthetic cardiac valve, an embolic signal with a relative intensity increase of 31 dB was detected at a depth of 52 mm. The relative background intensity increase over the entire screen including the embolic signal was 46 dB. The software also calculated the velocity of the particle, which was 80 cm/s. The embolus was identified a second time at a depth of 47 mm. The time lag in the appearance of the embolus could clearly be demonstrated by the bichannel recordings of the pre-FFT signal (Fig 1⇑, right). There was a lag of 7 milliseconds between the peaks of the two signals. This difference would indicate that the embolus had moved 5.6 mm from the first to the second sample volume [(7 ms)*(800 mm/1000 ms)=5.6 mm]. The expected preset difference with respect to the two sample volumes was 5 mm. The term “fictive sample volume distance” (FSVD) is used to indicate the spatial distance of the two sample volumes calculated by the machine from the velocity and the time of occurrence of the embolus in the two sample volumes as opposed to the expected preset distance of the two sample volumes of 5 mm. This FSVD appears somewhat unusual at first sight; however, compared with the parameter “time lag in occurrence,” the FSVD has the advantage of being independent from the velocity of the moving particle because a slowly moving embolus is more likely to produce a time delay than a fast one. By contrast, an ideal probe motion artifact has an FSVD of 0 mm, since the artifact occurs simultaneously in both sample volumes. One aim of this study was to define the FSVD range within which an embolus could be identified as such as opposed to artifacts.
Twelve young medical students with normal color duplex findings of their cervical and intracranial arteries and without any cardiovascular or cerebrovascular disease in the past participated in the study after having given informed consent. These subjects were taking no medication. There were 6 women and 6 men. In 3 women and 3 men, we recorded blood flow from the left MCA, in the other subjects from the right MCA. The more peripheral recording depth varied from 43 to 53 mm, the deeper one from 48 to 58 mm. As mentioned above, the spatial distance between the sample volumes was kept at 5 mm.
These normal control subjects first underwent a 7-minute series of provoked artifacts. In the first minute they opened and closed their mouths 10 times, in the second minute they coughed 10 times, in the third minute they clicked their teeth together 10 times, in the fourth minute they swallowed 10 times, in the fifth minute they moved their jaw 10 times horizontally to the side of the ultrasound probe and away from it, in the sixth minute the investigator tapped 10 times against the probe, and in the seventh minute the subjects read a text from Goethe’s Faust aloud. Second, during the following 53 minutes (indicated below as the “silent period”) the subjects were asked not to move and not to speak.
Patients With Prosthetic Cardiac Valves
Seven men and 3 women, aged 50 to 82 years, were investigated 3 months to 18 years after cardiac surgery. One patient had a Björk-Shiley and 9 had Carbomedics prosthetic aortic valves. In 4 patients, MCA flow signals were recorded on the left and in 6 on the right. Insonation depth of the deeper sample volume varied from 56 to 41 mm, and the more peripheral sample volume was set 5 mm apart. Six patients had minor carotid artery plaques on the investigated side that were not expected to produce microemboli; 4 had a normal ultrasound examination. All patients were taking oral anticoagulation.
No embolic signals were found in normal control subjects by the investigator’s acoustic and visual off-line evaluation of both the artifact periods as well as the artifact-free silent period. By contrast, the software recorded 501 events in the artifact-free period with a fictive distance of the two sample volumes of >0 mm, ie, meeting features of moving emboli. The relative intensity increases of these events, which corresponded to spontaneous fluctuations of the Doppler spectrum (Doppler speckle background), were normally distributed (bell-shaped distribution). Fig 2⇓ shows this distribution in percentage of the total number of 501. Of these events, 6% (30/501) had a relative intensity increase of 11 dB, and only 1.6% of all events (7/501) were 12 or 13 dB above background. With the setting we used, a relative intensity increase of 12 dB and more seemed to be a reasonable cutoff value to differentiate between embolic signals verified by an experienced investigator and arbitrary events within the background Doppler spectrum.
A total of 720 artifacts (plus 1 minute of continuous-reading artifacts in each subject) were produced. The software recorded 259 of these artifacts as events. There were 39 recorded artifacts during mouth opening, 34 during coughing, 40 during teeth clicking, 3 during swallowing, 13 during jaw movement, 93 during tapping against the probe, and 37 during reading. The distribution of the artifacts with respect to FSVD and intensity increase is given in Fig 3⇓.
The artifacts during tapping against the probe, mouth opening, and teeth clicking had higher intensity increases and were closer to 0 mm of FSVD than the other artifacts. There were 199 artifacts of ≥12 dB. Thus, the intensity increase of an event is not a sufficient parameter to differentiate artifacts from embolic signals. This is why we focused on the FSVD as an additional parameter to differentiate microemboli from artifacts. The white columns in Fig 4⇓ illustrate the distribution of the FSVD of the artifacts of ≥12 dB.
With a threshold of an FSVD of ≥0.5 mm, 22.0% of the 259 recorded artifacts (or 12.4% of 259 if additionally only artifacts with a relative intensity increase of ≥12 dB were considered) would be included as embolic in nature. If only events of an FSVD of ≥1 mm were included, 14.7% of the 259 recorded artifacts (or 8.5% of 259 when additionally considering only artifacts with a relative intensity increase of ≥12 dB) would be classified as embolic signals. When signals of an FSVD of ≥2.0 mm were considered, 8.5% of the 259 recorded artifacts (or 3.9% of 259 when considering additionally only artifacts with a relative intensity increase of ≥12 dB) would be included. Therefore, an FSVD threshold of ≥1 mm seemed to be a reasonable threshold.
Patients With Prosthetic Cardiac Valves
A total of 268 embolic signals was detected off-line on the digital audiotapes by the investigator’s acoustic and visual evaluation. All patients had embolic signals the frequency of which varied from 2/h to 84/h (15, 26, 2, 13, 2, 12, 35, 3, 76, and 84 in the 10 patients, respectively). All of these 268 events except one were correctly identified by the software, with varying intensity increases and distances.
Mean relative intensity increase of all the 267 recorded signals was 24.5±7.0 dB; mean FSVD was 2.3±4.4 mm.
Thirteen embolic signals had a relative intensity increase of <12 dB. Six of these 13 events showed a FSVD of <1 mm, and 6 ranged from 1 to 10 mm in FSVD. One embolic signal revealed an FSVD of >10 mm. Only 3 of all embolic signals (1.1%) had an FSVD of >10 mm; 1.1% of the embolic signals and 0.7% of the artifacts had an FSVD of >10.0 mm. Thus, an upper threshold of 10 mm of FSVD seemed reasonable.
Fifty-nine embolic signals (22.1%) had an FSVD of <1 mm; 205 (76.8%) of all the 267 embolic signals were within the FSVD limits of 1 to 10 mm, corresponding to 199 (78.3%) of the 254 recorded embolic signals of ≥12 dB. Another 28 (10.5%) of all the 267 embolic signals could have been detected if an FSVD of ≥0.5 mm had been accepted, corresponding to 28 (11.0%) of the 254 recorded embolic signals of ≥12 dB. Only 31 (11.6%) of all the 267 embolic signals or 25 (9.8%) of the 254 ones with ≥12 dB had an FSVD of <0.5 mm. The black columns in Fig 4⇑ illustrate the distribution of the FSVD of embolic signals of ≥12 dB.
In the patients with prosthetic cardiac valves, the software used in this study did not only detect (1) embolic signals as events but also (2) artifacts, (3) faint signals we would not consider to correspond to emboli (but which could result from very tiny emboli), and (4) signals we could not confirm visually or acoustically. With a relative intensity increase of ≥12 dB and with an FSVD from 1 to 10 mm, a total of 332 events including 199 genuine embolic signals was identified. This yielded a specificity of 59.9% and a sensitivity of 74.3%. We went over the digital audiotapes again and listened to each period identified by the software to contain a moving intensity increase of ≥12 dB. In 70 such cases, nothing was heard or seen that would match the investigator’s experience of an embolic event or that could be identified as circumscribed short-lasting transient at all. In 32 additional such periods of interest, a very weak signal was noted; it was too faint, however, to be acceptable as embolic. In 31 further cases, we found a clear-cut artifact, falsely detected by the software.
Our study demonstrates that automatic embolus detection by bigated transcranial Doppler ultrasound, although not yet perfect, may serve as a tool to identify clinically silent circulating microemboli and to reject transient intensity increases produced by artifacts. For the instrumentation and setting we used, we defined a range of relative intensity increase of ≥12 dB to discriminate events (true embolic signals and artifacts) from the Doppler speckle background and an FSVD of 1 to 10 mm for a preset sample volume distance of 5 mm to discriminate embolic signals from artifacts. With the setting used, 91.5% of all recorded artifacts could be identified as such. A specificity of 59.9% and a sensitivity of 74.3% for embolic signals verified by an experienced investigator could be obtained. Assessment during the recording periods of interest by the software and off-line review of these periods of interest by the investigator helps to save time in the discrimination of embolic signals from the Doppler speckle background and from artifacts.
In the study by Georgiadis et al,12 with use of a different software, 98.1% of the artifacts and 98.1% of the microembolic signals were identified correctly. The method they used, however, was different, and their terms “sensitivity” and “specificity” are not comparable with ours. They first assessed the position of an event off-line and then assessed the characteristics of the event using the software. Their terms sensitivity and specificity refer to events preselected by a human investigator and not to the raw data. By contrast, we wanted to assess the value of the software processing the raw data by itself without any preselection. Using the same software as Georgiadis et al, Molloy and Markus14 found a sensitivity of 75.2% for microemboli in valve patients and a sensitivity for microemboli of 92.6% in patients with carotid artery stenoses. Specificity was 99.6%.
The values calculated for sensitivity and specificity during this investigation are also worse than the values obtained by a neural network (both >99%)6 and less favorable than the off-line algorithm used by Markus et al (>97%).20 However, the above neural network had first been trained according to manmade decisions. We suspect that the neural network will repeat its teacher’s mistakes when evaluating the Doppler signal. Because bigated Doppler adds a new dimension to the definition of embolic signals (namely, the time lag of moving particles when passing two gates arranged in sequence), it would be interesting to combine the bigated Doppler technique with a neural network and to perform interobserver studies between these methods.
Our finding of natural fluctuations (Doppler speckle) of the intensity increase distribution in the artifact-free periods in healthy volunteers is similar to the one described by Markus et al.3 According to them, only 3% of episodes of Doppler speckle had a relative intensity increase of more than 4 dB. They chose this latter value as a cutoff for the identification of embolic signals. Their decibel values, however, were calculated using 5-millisecond time frames and comparing manually the maximum relative power amplitude of the events to the average of three amplitude measurements taken out of the preceding or following cardiac cycle at the same velocity. Our decibel values were three to four times higher due to the fact that our software uses the average relative power amplitude of the whole screen as a background. This calculation is technically easier and faster, since the software does not need to pinpoint a comparable section in the previous or following cardiac cycle. This approach is also less time consuming than manual assessment of the relative power amplitudes of the events. In analogy to the study by Markus et al,3 we chose an intensity increase cutoff of 12 dB in our study. The relative intensity increase thresholds for embolic signals vary considerably in the literature because of the different techniques used: 3 dB,19 4 dB,3 9 dB,6 11 and 12 dB (the present study) have been proposed. Future definitions should take into account these differences and should also provide intensity values of normal Doppler speckle background in the absence of embolic signals.
Some questions remain open and need to be addressed in future studies. The expected distance of the two sample volumes calculated from embolic signals was 5 mm. However, the distance most frequently found was between 1 and 2 mm. The unknown insonation angle cannot account for this discrepancy. The sonographically measured velocities of the embolus and the blood column change concomitantly with the cosine of the insonation angle, thus neutralizing each other because the angle is set constant for both sample volumes. There is no reason to assume that the insonation angle is constantly different for the two samples. Both beam width and axial sample volume are, however, greater for embolic signals than for the normal Doppler spectrum. This is due to the high relative intensity increases produced by embolic signals, making them detectable even in the margins of the sample volume and possibly outside its 5-mm diameter given by the manufacturer. It may be that the effective axial volume for high-intensity transient signals of the deeper sample extended into regions where no embolic signals could be detected and that the more peripheral (ie, more shallow) sample volume extended into the other sample volume, thus shortening the effective distance between the two samples. Here again, more refinement of the technology in good models is necessary.
A 5-mm distance of the sample volumes was chosen in this study, taking into account the normal anatomic length and course of the M1 segment of the MCA. With a longer distance, one might have been faced with curves in the artery course or might have erroneously insonated MCA branches or the distal ICA. The velocity of the embolus, which probably does not move constantly in the axial flow cylinder, changes permanently. Thus, the variability in embolus velocity will be less pronounced with a shorter distance of the two sample volumes than with a larger one, making calculations more accurate. Whether enlarging or reducing the distance of the two sample volumes would result in a higher specificity and sensitivity needs to be investigated.
Selected Abbreviations and Acronyms
|FFT||=||fast Fourier transform|
|FSVD||=||fictive sample volume distance|
|ICA||=||internal carotid artery|
|MCA||=||middle cerebral artery|
We are very grateful to A. Jahn for technical assistance and to Dr H.S. Markus (London, England) and Professor Dr E.B. Ringelstein (Münster, Germany) for their helpful comments.
- Received September 19, 1996.
- Revision received November 7, 1996.
- Accepted December 16, 1996.
- Copyright © 1997 by American Heart Association
Markus HS, Thomson N, Brown MM. Asymptomatic cerebral embolic signals in symptomatic and asymptomatic carotid artery disease. Brain. 1995;118:1005-1011.
Babikian VL, Hyde C, Pochay V, Winter MR. Clinical correlates of high-intensity transient signals detected on transcranial Doppler sonography in patients with cerebrovascular disease. Stroke. 1994;25:1570-1573.
Spencer MP. Detection of cerebral arterial emboli. In: Newell DW, Aaslid R, eds. Transcranial Doppler. New York, NY: Raven Press Publishers; 1992:216-230.
Siebler M, Sitzer M, Rose G, Bendfeldt D, Steinmetz H. Silent cerebral embolism caused by neurologically symptomatic high-grade carotid stenosis. Brain. 1993;116:1005-1015.
Spencer MP, Thomas GI, Nicholls SC, Sauvage LR. Detection of middle cerebral artery emboli during carotid endarterectomy using transcranial Doppler ultrasonography. Stroke. 1990;21:415-423.
van Zuilen EV, Moll FL, Vermeulen FE, Mauser HW, van Gijn J, Ackerstaff RG. Detection of cerebral microemboli by means of transcranial Doppler monitoring before and after carotid endarterectomy. Stroke. 1995;26:210-213.
Siebler M, Sitzer M, Rose G, Steinmetz H. Cerebral microembolism and the risk of ischemia in asymptomatic high-grade internal carotid artery stenosis. Stroke. 1995;26:2184-2186.
Siebler M, Sitzer M, Steinmetz H. Detection of intracranial emboli in patients with symptomatic extracranial carotid artery disease. Stroke. 1992;23:1652-1654.
Georgiadis D, Goeke J, Hill M, König M, Nabavi DG, Stögbauer F, Zunker P, Ringelstein EB. A novel technique for identification of Doppler microembolic signals based on the coincidence method: in vitro and in vivo evaluation. Stroke. 1996;27:683-686.
Nabavi D, Georgiadis D, Mumme T, Zunker P, Ringelstein EB. Detection of microembolic signals in patients with middle cerebral artery stenosis by means of a bigate probe. Stroke. 1996;27:1347-1349.
Molloy J, Markus HS. Multigated Doppler ultrasound in the detection of emboli in a flow model and embolic signals in patients. Stroke. 1996;27:1548-1552.
Georgiadis D, Mackay TG, Kelman AW, Grosset DG, Wheatley DJ, Lees KR. Differentiation between gaseous and formed embolic materials in vivo: application in prosthetic heart valve patients. Stroke. 1994;25:1559-1563.
Braekken SK, Russell D, Brucher R, Svennevig J. Incidence and frequency of cerebral embolic signals in patients with a similar bileaflet mechanical heart valve. Stroke. 1995;26:1225-1230.
Droste DW, Markus HS, Brown MM. The effect of different settings of ultrasound pulse amplitude, gain and sample volume on the appearance of emboli studied in a transcranial Doppler model. Cerebrovasc Dis. 1994;4:152-156.
Consensus Committee of the 9th International Cerebral Hemodynamic Symposium. Basic identification criteria of Doppler microembolic signals. Stroke. 1995;26:1123. Special report.
Markus H, Loh A, Brown MM. Computerized detection of cerebral emboli and discrimination from artifact using Doppler ultrasound. Stroke. 1993;24:1667-1672.