Transcranial Doppler (TCD) ultrasound is an essential tool for the detection of cerebral vasospasm after subarachnoid hemorrhage (SAH) but is limited by the availability of skilled operators. We examined the clinical feasibility and concordance of a robotically assisted TCD system with artificial intelligence with routine handheld TCD after SAH.
We evaluated TCD velocities in the anterior cerebral artery (ACA) and middle cerebral artery (MCA) of two patients with high-grade SAH and angiographic evidence of vasospasm. A single channel TCD device with a handheld diagnostic probe as well as a robotically assisted TCD device was used, the relationship of the two tests was evaluated using the bootstrap method of resampling for the concordance correlation coefficient (CCC) paired with a Pearson’s correlation analysis, followed by a Bland-Altman plot.
Patient 1 developed angiographic and TCD evidence of vasospasm in the proximal right MCA, but except for periods of disorientation remained neurologically intact. Angiographic, TCD and clinical evidence of ACA spasm occurred 6 days after ictus in patient 2. Robotically measured mean flow velocities were comparable to manual TCDs in the MCAs (CCC=0.83; 95% confidence interval [CI], 0.42 to 0.96;
Robotically assisted TCD system with artificial intelligence provides an alternative to manual TCD for assessment of MCA velocities in patients with SAH, expanding the availability of TCD to settings in which specialized clinicians are not available. Further studies for validation of this technology are warranted.
Delayed cerebral ischemia (DCI), defined as cerebral infarction or neurological deterioration caused by cerebral vasospasm, is a significant cause of mortality and poor neurological outcome after nontraumatic subarachnoid hemorrhage (SAH) [
Early diagnosis of vasospasm is critically dependent on frequent and high-quality neurological examinations, but the noninvasive assessment of the blood flow velocity in the basal cerebral arteries or angiographic studies are also important [
A significant limitation of the routine use of TCD is that it is time-consuming and highly operator dependent [
In this study, we aimed to examine the feasibility and accuracy of robotically assisted TCD for measuring cerebral blood flow velocities in the anterior cerebral artery (ACA) and MCA in patients with high grade SAH and angiographic vasospasm.
After approval from Institutional Review Board at Beth Israel Deaconess Medical Center (Protocol number: 2019P001001), we reviewed the clinical data, the TCD results, and imaging reports from two patients, with high-grade SAH and intraventricular blood, requiring placement of external ventricular drain (EVD) for hydrocephalus management. Written informed consent by the patients was waived by the board. Both patients received standard monitoring and therapies, including nimodipine prophylaxis, optimization of their hemodynamics and volume status, as well as fever control and correction of metabolic and electrolyte disturbances.
A 47-year-old male developed a thunderclap headache without any known triggering factors. He awoke the next morning with a severe headache and mental status changes and was brought to the hospital by his family. A computed tomography (CT) scan of his head showed a modified Fischer 3 SAH (
A 60-year-old previously healthy male presented after acute onset of severe headache followed by vomiting and unresponsiveness (Hunt and Hess Grade V). A head CT showed extensive and thick SAH with intraventricular hemorrhage, a modified Fischer 4 SAH (
Manual TCD imaging was completed by vascular neurologists using a single channel TCD device with a handheld diagnostic probe (ST3 Transcranial Doppler, Spencer Technologies, Redmond, WA, USA). Once applied to the head, the Lucid® Robotic System (Neural Analytics) with AI and machine learning algorithm automatically searched for and detected the bilateral MCAs independent of an operator. The device also allowed for identification and measurement of TCD velocities in other cerebral vessels, but required control and adjustment of the TCD probes through a computer screen by a vascular technologist. The ACA measurements were, hence, not fully automated. We compared results of attempted insonation of the terminal internal carotid artery (ICA), MCA, ACA via the transtemporal window. We did not compare attempted insonation of the posterior cerebral artery, also visible via the transtemporal window. We recorded all suitable quality waveforms along the terminal ICA and MCA. The robotic system consistently automatically located at least one (and often several) depths of the MCA. Manual manipulation of the robotic TCD probe was then performed to optimize signals, check additional depths, and to locate ACA signals.
The dependent concordance correlation coefficient (CCC) was calculated to measure the agreement between robotic and manual TCD methods. A resampling method was used for interference on dependent CCCs. The bootstrap method was used as the choice of resampling approach [
For patient 1, manual and robotic TCD ultrasonographic studies were performed on the same day on three occasions, on postictus days 10, 11, and 12 (
For patient 2, manual and robotic TCD ultrasonographic studies were performed on the same day on two occasions, on postictus days 15 and 17 (
Representative plot of the robotic and manual TCD are presented in
Patient 1 was extubated soon after the coiling of his aneurysm with his neurological examination returning rapidly to his baseline, apart from intermittent confusion and delirium. He remained neurologically intact and was discharged to his home in stable condition 17 days after his bleeding (modified Rankin Scale score=0). Patient 2 improved gradually after intraarterial vasospasm therapy and initiation of milrinone. He was successfully extubated 11 days after admission. Milrinone was weaned as his vasospasm improved, and his EVD was removed 1 week later (on hospital day 18). The patient was discharged to a rehabilitation facility in stable condition with moderate cognitive deficits and moderate left hemiparesis (modified Rankin Scale score=4).
We report that robotically assisted TCD with AI is a feasible alternative to the standard handheld technique for obtaining flow velocities in patients with cerebral vasospasm after high grade SAH. Despite significant vasospasm in both patients, MCA and ACA velocities could be identified without significant delay, and the AI guided MCA velocities were comparable to those obtained with the manual technique by expert clinicians.
The importance of early TCD monitoring in patients with high-grade SAH is demonstrated by the previous findings that elevated velocities may precede clinical symptoms by 24 to 48 hours [
The Lucid Robotic System combines TCD with a headset containing robotic wands and uses machine learning to find the best cranial window and insonation angle based on patterns in the data it gathers (
It is important to note that although the two methods were performed on the same day, the discrepancies in the mean flow velocities in ACAs were higher than MCAs. In addition to underlying physiologic changes, the discrepancy in ACA flow velocities can indicate an inherent difference in training and technical skills between the two operators: a vascular neurologist versus a vascular technologist. It can also imply technical limitations with the manually controlled ACA measurements with the robotic TCD: the AI machine learning technology was only available and used for the MCAs, whereas the ACA measurements remained operator dependent.
Limitations of the TCD imaging in SAH patients include an inability to insonate intracranial vessels in 10% to 20% of patients. Indeed, Seidel et al. [
The robotically assisted TCD system used in this study has additional limitations such as inability to insonate posterior circulation with its current headset, as well as its large size and weight (
Although our results are compelling, it is important to discuss the limitations of our study. First, we have a very small sample size in this study which may affect our results and findings. Second, in this study the robotic and manual TCD measurements were often conducted many hours apart, albeit on the same day. Although unlikely, any physiological changes over the course of the day may have affected the velocities and lead to the discrepancy between data. These and other limitations described above will need to be addressed in future studies.
In our two patients, robotically assisted TCD with AI was feasible for evaluation of MCA waveforms in SAH-associated vasospasm and provided results that were overall comparable to manual TCD performed the same day. Further studies to assess the validity of this technology in SAH patients are warranted.
No potential conflict of interest relevant to this article.
Conceptualization: VAL, CSO, and AN. Data curation: SE, CMH, ABB, JW, and SGP. Formal analysis: SE, CRF, and AN. Visualization & Writing–original draft: SE and AN. Writing–review editing: CMH, JW, KAH, VAL, CSO, AJT, SS, and CRF.
Computer tomography imaging of (A) patient 1 and (B) 2 showing diffuse thick subarachnoid hemorrhage with intraventricular blood.
Angiographic findings in patient 2, (A) initially demonstrating normal calibers in all cerebral vessels and (B) significant vasospasm in bilateral anterior cerebral arteries 6 days after ictus.
(A) Robotic and (B) manual transcranial Doppler waveform of the right middle cerebral artery (MCA) of patient 1 on postictus day 10 demonstrating elevated right MCA mean flow velocity, consistent with vasospasm in both techniques.
(A) Moderate agreement between transcranial Doppler (TCD) findings using robotic technique and TCD findings using manual technique for the middle cerebral artery (concordance correlation coefficient=0.83; 95% confidence interval [CI], 0.42 to 0.96). (B) Poor agreement between TCD findings using robotic technique and TCD findings using manual technique for the anterior cerebral artery (concordance correlation coefficient=0.26; 95% CI, –0.01 to 0.71).
Bland-Altman plot shows no proportional bias, indicating agreement between the two methods for the (A) middle cerebral artery and (B) anterior cerebral artery.
The robotically assisted transcranial Doppler system used in this study (Neural Analytics Lucid Robotic System. Credit: Neural Analytics).
Mean flow velocities in anterior circulation reported by manual and robotic TCD imaging
Patient 1 (days after SAH) |
Patient 2 (days after SAH) |
|||||||||
---|---|---|---|---|---|---|---|---|---|---|
Manual TCD |
Robotic TCD |
Manual TCD |
Robotic TCD |
|||||||
Days after SAH | 10 | 11 | 12 | 10 | 11 | 12 | 15 | 17 | 15 | 17 |
RMCA velocity (cm/sec) | 153 at 55 mm | 149 at 48 mm | 140 at 49 mm | 159 at 59 mm | 130 at 59 mm | 154 at 50 mm | 80 at 51 mm | 73 at 62 mm | 78 at 51 mm | 71 at 59 mm |
LMCA velocity (cm/sec) | 142 at 45 mm | 139 at 51 mm | 147 at 41 mm | 203 at 54 mm | 119 at 54 mm | 140 at 51 mm | 75 at 54 mm | 81 at 56 mm | 89 at 55 mm | 69 at 56 mm |
RACA velocity (cm/sec) | –49 at 63 mm | –40 at 61 mm | –39 at 63 mm | –87 at 79 mm | NR | –65 at 71 mm | –27 at 70 mm | –23 at 62 mm | NR | NR |
LACA velocity (cm/sec) | –45 at 64 mm | –104 at 66 mm | –101 at 64 mm | –121 at 66 mm | –100 at 70 mm | –99 at 68 mm | –38 at 66 mm | –31 at 66 mm | –83 at 79 mm | –58 at 75 mm |
Duration of the imaging (min) | 52 | 43 | 57 | 40 | 33 | 36 | 24 | 17 | 32 | 17 |
TCD, transcranial Doppler; SAH, subarachnoid hemorrhage; RMCA, right middle cerebral artery; LMCA, left middle cerebral artery; RACA, right anterior cerebral artery; NR, not reported; LACA, left anterior cerebral artery.
Comparison between the mean flow velocities in anterior circulation reported by manual and robotic transcranial Doppler imaging
Average bootstrap estimate of concordance coefficient | Bootstrap standard error | Bootstrap 95% BCa confidence interval | Bootstrap estimates of bias | |
---|---|---|---|---|
MCA | 0.83 | 0.12 | 0.42 to 0.96 | –0.01 |
ACA | 0.26 | 0.20 | –0.01 to 0.71 | –0.001 |
Bootstrapping was applied with n=1,000 to obtain concordance coefficient results.
BCa, bias-corrected and accelerated; MCA, middle cerebral artery; ACA, anterior cerebral artery.