Embolus Detection and Differentiation Using Multifrequency Transcranial Doppler
To the Editor:
In their article on the detection and discrimination of cerebral microemboli using dual-frequency transcranial Doppler (TCD), Markus and Punter write that this method is not accurate enough for use in clinical or research studies.1 Furthermore, that “It is possible that further refinement of the dual-frequency approach may improve discrimination.” They unfortunately are not aware that further refinement of this method has in fact been carried out since our original description of the first multifrequency TCD in 2002.2
In our original description of multifrequency TCD using in vitro and limited in vivo data, we recognized 3 potential problems when using this method to differentiate cerebral microemboli.3,4 Firstly, that small variations in the Doppler background signal and changes in reflected Doppler power measurements attributable to beam distortion could affect sensitivity when detecting small solid microemboli. We proposed therefore a horizontal lower dEBR (reflected ultrasound power at 2.5-MHz compared with that at 2.0-MHz) detection limit for small solid microemboli of −0.83 dB. However, if the dEBR was within 0.2 dB of this limit (ie, between −0.63 and −0.83 dB), it was classified as uncertain solid. Subsequent clinical experience showed, however, that this dEBR detection limit sometimes had problems detecting small solid microemboli, such as those which can be present in patients with carotid stenosis. This clinical experience enabled us to improve sensitivity for solid microemboli without influencing specificity by changing lower dEBR detection limit from a horizontal line at −0.83 dB to a dEBR detection limit with a slight slope of y=−0.1, x=−0.12 dB, where y=dEBR and x=2.0 MHz EBR.2
Markus and Punter did not use this refined lower dEBR for the detection of solid microemboli in their carotid stenosis patients, which can explain why many of the solid microemboli were not correctly classified in their study.
The second problem which we described in our original description of multifrequency TCD was resonance caused by very small gas bubbles (<3 μm; <8 dB). This was first apparent when ultrasound contrast was injected into the circulation but was subsequently observed in all clinical situations where air may be introduced into the circulation. Very small gas bubbles which resonate at 2.5-MHz ultrasound frequency will give a larger Doppler power measurement attributable to resonance and be incorrectly classified as solid. This is because embolus differentiation is based on the principle that solid microemboli normally reflect more ultrasound at 2.5-MHz frequency compared with 2.0-MHz, whereas the opposite is the case for gaseous microemboli. Multifrequency TCD can therefore not correctly differentiate very small microbubbles where there is a possibility of resonance effects. This potential error can be reduced by using multifrequency TCD to discriminate only those emboli which cause a Doppler signal enhancement ie, embolus-blood-ratio (EBR) of >28 dB/ms simultaneously in both 2.0-MHz and 2.5-MHz channels.2
These limits were not used by Markus and Punter which can explain why some of the microbubbles which were introduced into their patent foramen ovale (PFO) patients were wrongly classified as solid. This can especially be the case for ≈4% of the gaseous emboli shown in their plot.1 These emboli gave a Doppler power increase at 2.0 MHz of 5 to 7dB which corresponds to bubble sizes <3μm.
It is also impossible to exclude the possibility that very small blood clots were introduced during their experiments in these patients even though they tried to avoid this by flushing the intravenous cannula.
The third problem which we acknowledged in our original publications was the fact that multifrequency TCD cannot accurately count all of the emboli when several emboli enter the sample volume at the same time. This is especially relevant when air bubbles are introduced into the cerebral circulation as is the case when an agitated air and saline solution is introduced into PFO patients.
Markus and Punter1 also assessed a simple intensity threshold at 2.0 MHz to differentiate solid and gaseous microemboli. This detection threshold is only appropriate for their very special data set where ≈88% of their solid microemboli gave a reflected Doppler power at 2.0 MHz of <13 dB, whereas ≈88% of the gaseous gave a reflected Doppler power at 2.0 MHz of >13 dB. This is normally not the case in routine clinical situations when gaseous and solid emboli are present.
In conclusion, we completely agree with Markus and Punter that it is extremely important to have a TCD method which can differentiate between solid and gaseous cerebral microemboli if we are going to understand the clinical significance of these events.
We feel that the multifrequency TCD method has considerable potential and our subsequent clinical experience with this method in the past 3 years has allowed us to further increase its clinical accuracy in different patient groups. Our revised criteria have been described previously in Stroke2 and in this letter, but to our knowledge have not been incorporated into any commercially available multifrequency TCD. These refinements can, however, be very easily made on the multifrequency instrumentation used by Markus and Punter.1 They should now use a clinically more representative set of data to assess the accuracy of multifrequency criteria for embolus differentiation and, if necessary, propose further refinements which may help toward achieving the important goal of embolus differentiation.
Markus HS, Martin P. Can Transcranial Doppler Discriminate Between Solid and Gaseous Microemboli?: Assessment of a Dual-Frequency Transducer System. Stroke. 2005; 36: 1731–1734.
Russell D, Brucher R. Embolus Detection and Differentiation Using Multifrequency Transcranial Doppler. Stroke. 2005; 36: 706.
Brucher R, Russell D. Automatic online embolus detection and artifact rejection with the first multifrequency transcranial Doppler. Stroke. 2002; 33: 1969–1974.
Russell D, Brucher R. Online automatic discrimination between solid and gaseous cerebral microemboli with the first multifrequency transcranial Doppler. Stroke. 2002; 33: 1975–1980.
In our article1 we applied the default software settings recommended by the manufacturer of this commercially available system, resulting in a sensitivity and specificity for solid emboli of 50.3% and 94.2%, respectively. There was some improvement in specificity when considering only those embolic signals (ES) classified as definite and those with an intensity increase >7 db when insonated with the 2.0 MHz transducer (Table 2), but this was at the expense of sensitivity. The sensitivities were 44.8% and 48.4%, respectively, and specificities were 96.91% and 97.4%, respectively.
We are grateful to Russell and Brucher for their suggestions and have reanalyzed our dataset with the further refinements as they describe. Results of the revised categorization of ES are shown in Table 1 and the revised sensitivities and specificities in Table 2. Firstly, we excluded ES which were classified as <7 db intensity at either frequency. This resulted in an improvement in sensitivity to 59.4% for all solid and 96.5% for all gaseous emboli, and specificity to 96.5% and 59.4%, respectively. However, this was at the expense of exclusion of a significant number of ES: 49/145 (33.8%) of the ES in patients with carotid stenosis and 80/648 (12.3%) of those in the PFO group. There was further improvement in sensitivity for solid emboli to 63.6%, with no change in specificity, when only ES categorized as definite were considered; this resulted in an additional 8 possible solid ES being excluded.
We performed a second analysis using the suggested lower dEBR detection limit y=−0.1, x=−0.12. When considering all ES this resulted in increased sensitivity for solid emboli (59.3%) but reduced specificity (86.9%) and reduced sensitivity for gaseous emboli (86.9%).
Implementing both revisions together resulted in increased sensitivity for solid emboli (74.2%) but reduced specificity (85.4%) and reduced sensitivity for gaseous emboli (85.4%).
In conclusion, the refinements as suggested by Russell and Brucher do improve sensitivity for solid ES. However, this is at a moderate loss of specificity and an exclusion of approximately a third of solid ES. ES >7 dB at 2 Mz have been found to be independently predictive of stroke, and may be infrequent in patients with carotid stenosis with a median per hour of only 4 in those subjects in whom ES are present during an hours recording.2 Therefore, exclusion of a third of these ES will convert some ES positive individuals to ES negative; it remains to be determined what effect this would have on the predictive value of the technique.
Markus HS, Punter M. Can Transcranial Doppler Discriminate Between Solid and Gaseous Microemboli? Assessment of a Dual-Frequency Transducer System. Stroke. 2005; 36: 1731–1734.
Markus HS, MacKinnon A. Asymptomatic embolisation, detected by Doppler ultrasound, predicts stroke risk in symptomatic carotid artery stenosis. Stroke. 2005; 36: 971–975.