Neurocrit Care DOI 10.1007/s12028-015-0118-9
ORIGINAL ARTICLE
Cranial Accelerometry Can Detect Cerebral Vasospasm Caused by Subarachnoid Hemorrhage Wade S. Smith • Janet L. Browne • Nerissa U. Ko
Ó Springer Science+Business Media New York 2015
Abstract Background We previously reported the presence of a cranial ‘‘bruit’’ in patients with cerebral vasospasm by signal processing cranial accelerometry signals time locked to the cardiac cycle. This shift to higher frequencies is likely related to the turbulence of blood flow produced by vascular narrowing. We sought to build a more quantitative model to predict cerebral vasospasm then test the accuracy of this technique to detect cerebral vasospasm in a prospective blinded study. Methods Skull accelerometry was performed using an array of 6 highly sensitive accelerometers placed in contact with the scalp. Paired transcranial Doppler (TCD) recordings and accelerometry epochs were obtained in consecutive patients with subarachnoid hemorrhage undergoing TCD recordings for surveillance of cerebral vasospasm. The energy of rectified acceleration measurements within systolic and diastolic bands of the cardiac cycle were measured and correlated with TCD-defined spasm. This model was then tested prospectively in a blinded consecutive sample of subarachnoid hemorrhage patients to determine accuracy of the technique. Results We developed a model predicting cerebral vasospasm from analysis of 14 unblinded subjects with varying degrees of cerebral vasospasm as detected by TCD. We then recorded from 58 subjects obtaining 125-paired recordings of accelerometry and TCD to test this model in Electronic supplementary material The online version of this article (doi:10.1007/s12028-015-0118-9) contains supplementary material, which is available to authorized users. W. S. Smith (&) J. L. Browne N. U. Ko Department of Neurology, University of California, San Francisco, San Francisco, USA e-mail: [email protected]
a blinded analysis. Accelerometry detection of any spasm versus non-spasm correlated with TCD-defined vasospasm (P < 0.001). The model was 81 % sensitive for detecting any cerebral vasospasm in patients, while the negative predictive value was 61 %. Conclusion Highly sensitive skull accelerometry can detect cerebral vasospasm with clinically meaningful accuracy. This tool holds promise in the ICU environment to detect as well as reject cerebral vasospasm as the cause of neurological deficits in subarachnoid hemorrhage. Keywords Windkessel effect Vascular oscillation Turbulence Brain motion Brain pulse
Introduction Delayed cerebral ischemia (DCI) following subarachnoid hemorrhage (SAH) remains a significant cause of brain injury and disability [1]. Cerebral vasospasm remains a treatable cause of DCI. Mitigation of brain ischemia from cerebral vasospasm is possible by increasing arterial blood pressure, infusing vasodilators, and performing cerebral angioplasty [1]. If cerebral vasospasm is detected prior to the development of cerebral ischemia, morbidity can be reduced, or eliminated. Therefore, the reduction in morbidity is dependent on the accurate and early detection of cerebral vasospasm. The detection of cerebral vasospasm is currently problematic. Although cerebral vasospasm may be suspected by change in clinical status alone—namely, decrease in level of consciousness or new focal neurological deficit—these changes may be non-specific resulting in high false positive rate for predicting true cerebral vasospasm. Transcranial Doppler (TCD) [2], CT angiography and CT perfusion [3],
123
Neurocrit Care
Cerebral accelerometry was performed using a proprietary device (Nautilus BrainPulse, Jan Medical, Mountain View California) shown in Fig. 1. The device has an array of 6 accelerometers that contact a patient’s skull affixed to a system of adjustable plastic bands. An ambient sound detector and an accelerometer are located at the vertex. Additionally, accelerometers are placed bi-frontally, bitemporally, and one over the occipital region. Each accelerometer is sensitive to 0.1 lg. The electrocardiogram
is captured with bipolar chest leads, or via pulse oximeter, and these signals are converted to digital waveforms using a custom-designed interface and laptop computer. Accelerometer signals were then signal averaged with the ECG or pulse oximeter as a trigger, and the resulting waveforms were converted to the frequency domain using a Fast Fourier Transform algorithm (Matlab, Natick, MA). A proprietary algorithm was generated from 14 subarachnoid patients with varying degrees of cerebral vasospasm in Phase I of this study (64 recordings: 47 % with no spasm, 19 % with mild, and 34 % with moderate to severe vasospasm). This model was built on the hypothesis that proximal arterial narrowing would reduce cranial acceleration more in diastole (when the driving force of blood is less) than in systole. Accelerometry waveforms were signal averaged to the cardiac cycle, rectified, and then integrated during systolic and diastolic phases. The ratio of these areas (diastolic/systolic) was termed the ‘‘SD ratio’’. For the 14 cases of known degrees of vasospasm, the TCDdefined category of spasm (see below) was submitted along with the SD ratios of each waveform to Categorization and Regression Tree analysis (CART software by Salford Systems, Inc, San Diego, CA) that defined SD cut points for each TCD category. Further analysis of these time signals was aided by conversion to the frequency domain using a standard FFT algorithm and plotting waterfall plots (Matlab, Natick, MA). In Phase II, the SD algorithm was tested in a prospective, consecutive, blinded group of patients with SAH who had concomitant TCD recordings. A paired recording of accelerometry and TCD was considered for further analysis if both middle cerebral arteries were insonated, and the TCD recording was within 8 h of the accelerometry
Fig. 1 Prototype cranial accelerometer. Each of 6 highly sensitive (±0.1 lg) accelerometers is held in place by an adjustable head holder placing them in contact with the scalp. An ambient sound
detector is located at the vertex. Signals are amplified and converted to digital using a standard analog to digital converter and sampled at high frequency using a laptop computer
and MR perfusion [4] have ability to document vessel caliber change either directly or indirectly, but each technique is not ideal. For imaging-based techniques (CT and MRI) both contrast administration and the necessary transport of the patient make such recordings challenging and only provide episodic measure of vascular status. TCD—although a portable technique—is operator dependent and is unable to interrogate vessels beyond the Circle of Willis. The creation of a non-invasive, portable method that samples more vascular territory is needed. Cerebral accelerometry is a newly introduced technology that measures the motion of flowing blood within the brain at the scalp surface. We have previously reported that this technique can reveal a ‘‘cranial bruit’’ from cerebral vasospasm [5], and the technique holds promise in detecting cerebral concussion [6]. We embarked on a prospective study of this technology to detect cerebral vasospasm in a TCD-defined cerebral vasospasm population to determine accuracy of this new technique compared to TCD as the gold standard.
Methods
123
Neurocrit Care
recording. TCD-defined cerebral vasospasm was defined as none, mild, moderate, or severe using the criteria in Tables 1 and 2, which has been used as a laboratory standard in our vascular laboratory. For the anterior circulation segments, cerebral vasospasm was categorized by considering the mean velocity of the insonated intracranial segment and the ratio of that velocity to the mean velocity of the extracranial internal carotid artery (‘‘ratio’’ in Table 1). The final category of cerebral vasospasm was assigned considering both mean velocity and ratio as shown in Table 2. Posterior circulation cerebral vasospasm was defined by consideration of the ratio of the insonated intracranial segment of the basilar artery to the mean of the two extracranial vertebral arteries as shown in Table 1. Vascular segments were grouped into right and left frontal (MCA, ACA) and posterior circulation (basilar and vertebral arteries) for correlation with accelerometry signals. Accelerometers were grouped similarly with right and left frontal (paired recordings of the ipsilateral frontal and temporal sensors) and posterior (the occipital sensor). The accelerometer-defined degree of cerebral vasospasm was compared with the TCD category of cerebral vasospasm using the v2 test for significance. The specificity and sensitivity of the accelerometer test was reported considering categories of none-mild versus moderate to severe TCD-defined cerebral vasospasm, and none versus any cerebral vasospasm (mild, moderate. and severe) as the gold standard. All patients or their surrogates provided informed consent and the study was approved by the UCSF Committee on Human Research (ClinicalTrials.gov # NCT01595802). In Phase II, all data were monitored by an independent data monitor. The sponsor had no access to the TCD data or
knowledge of the clinical status of any patient in the Phase II cohort until the study was completed and all analyses of accelerometry data were performed blinded to TCD data. The database of accelerometry results was locked following monitor review of data completeness. The authors had full access to the TCD data and the analysis data of accelerometry signals and were not restricted in writing this manuscript.
Table 1 TCD criteria for severity of cerebral vasospasm
Phase II Results
Category
Anterior circulation Mean velocity (cm/sec)
Ratio
Posterior circulation Ratio
None
ORIGINAL ARTICLE
Cranial Accelerometry Can Detect Cerebral Vasospasm Caused by Subarachnoid Hemorrhage Wade S. Smith • Janet L. Browne • Nerissa U. Ko
Ó Springer Science+Business Media New York 2015
Abstract Background We previously reported the presence of a cranial ‘‘bruit’’ in patients with cerebral vasospasm by signal processing cranial accelerometry signals time locked to the cardiac cycle. This shift to higher frequencies is likely related to the turbulence of blood flow produced by vascular narrowing. We sought to build a more quantitative model to predict cerebral vasospasm then test the accuracy of this technique to detect cerebral vasospasm in a prospective blinded study. Methods Skull accelerometry was performed using an array of 6 highly sensitive accelerometers placed in contact with the scalp. Paired transcranial Doppler (TCD) recordings and accelerometry epochs were obtained in consecutive patients with subarachnoid hemorrhage undergoing TCD recordings for surveillance of cerebral vasospasm. The energy of rectified acceleration measurements within systolic and diastolic bands of the cardiac cycle were measured and correlated with TCD-defined spasm. This model was then tested prospectively in a blinded consecutive sample of subarachnoid hemorrhage patients to determine accuracy of the technique. Results We developed a model predicting cerebral vasospasm from analysis of 14 unblinded subjects with varying degrees of cerebral vasospasm as detected by TCD. We then recorded from 58 subjects obtaining 125-paired recordings of accelerometry and TCD to test this model in Electronic supplementary material The online version of this article (doi:10.1007/s12028-015-0118-9) contains supplementary material, which is available to authorized users. W. S. Smith (&) J. L. Browne N. U. Ko Department of Neurology, University of California, San Francisco, San Francisco, USA e-mail: [email protected]
a blinded analysis. Accelerometry detection of any spasm versus non-spasm correlated with TCD-defined vasospasm (P < 0.001). The model was 81 % sensitive for detecting any cerebral vasospasm in patients, while the negative predictive value was 61 %. Conclusion Highly sensitive skull accelerometry can detect cerebral vasospasm with clinically meaningful accuracy. This tool holds promise in the ICU environment to detect as well as reject cerebral vasospasm as the cause of neurological deficits in subarachnoid hemorrhage. Keywords Windkessel effect Vascular oscillation Turbulence Brain motion Brain pulse
Introduction Delayed cerebral ischemia (DCI) following subarachnoid hemorrhage (SAH) remains a significant cause of brain injury and disability [1]. Cerebral vasospasm remains a treatable cause of DCI. Mitigation of brain ischemia from cerebral vasospasm is possible by increasing arterial blood pressure, infusing vasodilators, and performing cerebral angioplasty [1]. If cerebral vasospasm is detected prior to the development of cerebral ischemia, morbidity can be reduced, or eliminated. Therefore, the reduction in morbidity is dependent on the accurate and early detection of cerebral vasospasm. The detection of cerebral vasospasm is currently problematic. Although cerebral vasospasm may be suspected by change in clinical status alone—namely, decrease in level of consciousness or new focal neurological deficit—these changes may be non-specific resulting in high false positive rate for predicting true cerebral vasospasm. Transcranial Doppler (TCD) [2], CT angiography and CT perfusion [3],
123
Neurocrit Care
Cerebral accelerometry was performed using a proprietary device (Nautilus BrainPulse, Jan Medical, Mountain View California) shown in Fig. 1. The device has an array of 6 accelerometers that contact a patient’s skull affixed to a system of adjustable plastic bands. An ambient sound detector and an accelerometer are located at the vertex. Additionally, accelerometers are placed bi-frontally, bitemporally, and one over the occipital region. Each accelerometer is sensitive to 0.1 lg. The electrocardiogram
is captured with bipolar chest leads, or via pulse oximeter, and these signals are converted to digital waveforms using a custom-designed interface and laptop computer. Accelerometer signals were then signal averaged with the ECG or pulse oximeter as a trigger, and the resulting waveforms were converted to the frequency domain using a Fast Fourier Transform algorithm (Matlab, Natick, MA). A proprietary algorithm was generated from 14 subarachnoid patients with varying degrees of cerebral vasospasm in Phase I of this study (64 recordings: 47 % with no spasm, 19 % with mild, and 34 % with moderate to severe vasospasm). This model was built on the hypothesis that proximal arterial narrowing would reduce cranial acceleration more in diastole (when the driving force of blood is less) than in systole. Accelerometry waveforms were signal averaged to the cardiac cycle, rectified, and then integrated during systolic and diastolic phases. The ratio of these areas (diastolic/systolic) was termed the ‘‘SD ratio’’. For the 14 cases of known degrees of vasospasm, the TCDdefined category of spasm (see below) was submitted along with the SD ratios of each waveform to Categorization and Regression Tree analysis (CART software by Salford Systems, Inc, San Diego, CA) that defined SD cut points for each TCD category. Further analysis of these time signals was aided by conversion to the frequency domain using a standard FFT algorithm and plotting waterfall plots (Matlab, Natick, MA). In Phase II, the SD algorithm was tested in a prospective, consecutive, blinded group of patients with SAH who had concomitant TCD recordings. A paired recording of accelerometry and TCD was considered for further analysis if both middle cerebral arteries were insonated, and the TCD recording was within 8 h of the accelerometry
Fig. 1 Prototype cranial accelerometer. Each of 6 highly sensitive (±0.1 lg) accelerometers is held in place by an adjustable head holder placing them in contact with the scalp. An ambient sound
detector is located at the vertex. Signals are amplified and converted to digital using a standard analog to digital converter and sampled at high frequency using a laptop computer
and MR perfusion [4] have ability to document vessel caliber change either directly or indirectly, but each technique is not ideal. For imaging-based techniques (CT and MRI) both contrast administration and the necessary transport of the patient make such recordings challenging and only provide episodic measure of vascular status. TCD—although a portable technique—is operator dependent and is unable to interrogate vessels beyond the Circle of Willis. The creation of a non-invasive, portable method that samples more vascular territory is needed. Cerebral accelerometry is a newly introduced technology that measures the motion of flowing blood within the brain at the scalp surface. We have previously reported that this technique can reveal a ‘‘cranial bruit’’ from cerebral vasospasm [5], and the technique holds promise in detecting cerebral concussion [6]. We embarked on a prospective study of this technology to detect cerebral vasospasm in a TCD-defined cerebral vasospasm population to determine accuracy of this new technique compared to TCD as the gold standard.
Methods
123
Neurocrit Care
recording. TCD-defined cerebral vasospasm was defined as none, mild, moderate, or severe using the criteria in Tables 1 and 2, which has been used as a laboratory standard in our vascular laboratory. For the anterior circulation segments, cerebral vasospasm was categorized by considering the mean velocity of the insonated intracranial segment and the ratio of that velocity to the mean velocity of the extracranial internal carotid artery (‘‘ratio’’ in Table 1). The final category of cerebral vasospasm was assigned considering both mean velocity and ratio as shown in Table 2. Posterior circulation cerebral vasospasm was defined by consideration of the ratio of the insonated intracranial segment of the basilar artery to the mean of the two extracranial vertebral arteries as shown in Table 1. Vascular segments were grouped into right and left frontal (MCA, ACA) and posterior circulation (basilar and vertebral arteries) for correlation with accelerometry signals. Accelerometers were grouped similarly with right and left frontal (paired recordings of the ipsilateral frontal and temporal sensors) and posterior (the occipital sensor). The accelerometer-defined degree of cerebral vasospasm was compared with the TCD category of cerebral vasospasm using the v2 test for significance. The specificity and sensitivity of the accelerometer test was reported considering categories of none-mild versus moderate to severe TCD-defined cerebral vasospasm, and none versus any cerebral vasospasm (mild, moderate. and severe) as the gold standard. All patients or their surrogates provided informed consent and the study was approved by the UCSF Committee on Human Research (ClinicalTrials.gov # NCT01595802). In Phase II, all data were monitored by an independent data monitor. The sponsor had no access to the TCD data or
knowledge of the clinical status of any patient in the Phase II cohort until the study was completed and all analyses of accelerometry data were performed blinded to TCD data. The database of accelerometry results was locked following monitor review of data completeness. The authors had full access to the TCD data and the analysis data of accelerometry signals and were not restricted in writing this manuscript.
Table 1 TCD criteria for severity of cerebral vasospasm
Phase II Results
Category
Anterior circulation Mean velocity (cm/sec)
Ratio
Posterior circulation Ratio
None
