A Unique Signature of Cardiac-Induced Cranial Forces During Acute Large Vessel Stroke and Development of a Predictive Model

• Published in: Neurocritical care Vol. 33; no. 1; pp. 58 - 63
• Main Authors: Smith, Wade S., Keenan, Kevin J., Lovoi, Paul A.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2020; Springer Nature B.V
• Subjects: Accelerometers; Accelerometry; Aged; Aged, 80 and over; Ballistocardiography; Cerebral Angiography; Computed Tomography Angiography; Critical Care Medicine; Digitization; Electrocardiography; Female; Fourier transforms; Humans; Infarction, Middle Cerebral Artery - diagnosis; Infarction, Middle Cerebral Artery - physiopathology; Intensive; Internal Medicine; Ischemia; Ischemic Stroke - diagnosis; Ischemic Stroke - physiopathology; Machine Learning; Male; Medicine; Medicine & Public Health; Middle Aged; Migraine; Migraine Disorders - diagnosis; Migraine Disorders - physiopathology; Neurology; Original Work; Patients; Pulsatile Flow; Software; Stroke; Tomography, X-Ray Computed; Veins & arteries; Headpulse; Large vessel stroke; Cranial accelerometry
• ISSN: 1541-6933; 1556-0961; 1556-0961
• DOI: 10.1007/s12028-019-00845-x

Background Cranial accelerometry is used to detect cerebral vasospasm and concussion. We explored this technique in a cohort of code stroke patients to see whether a signature could be identified to aid in the diagnosis of large vessel occlusion (LVO) stroke. Methods A military-grade three-axis accelerometer was affixed to a headset. Accelerometer and electrocardiogram (ECG) outputs were digitized at 1.6 kHz. We call the resulting digitized signals the “headpulse.” Three-minute recordings were performed immediately after computed tomography (CT) angiography (CTA) and/or immediately before and after attempted mechanical thrombectomy in patents with suspected stroke. The resulting waveforms were inspected by eye and then subjected to supervised machine learning (MATLAB Classification Learner R2018a) to train a model using fivefold cross-validation. Results Of 42 code stroke subjects with recordings, 19 (45%) had LVO and 23 (55%) had normal CTAs. In patients without LVO, ECG-triggered waveforms followed a self-similar time course revealing that the headpulse is highly coupled to the cardiac contraction. However, in most patients with LVO, headpulses showed little cardiac contraction correlation. We term this abnormality “chaos” and parameterized it with 156 measures of trace-by-trace variation from the ECG-signal-averaged mean for machine learning model training. Selecting the best model, using biometric data only, we properly classified 15/19 LVOs and 20/23 non-LVO patients, with receiver operating characteristic curve area = 0.79, sensitivity of 73%, and specificity of 87%, P  < 0.0001. Headpulse waveforms following thrombectomy showed return of cardiac contraction correlation. Conclusions Headpulse recordings performed on patients with suspected acute stroke significantly identify those with LVO. The lack of temporal correlation of the headpulse with cardiac contraction and resolution to normal may reflect changes in cerebral blood flow and may provide a useful technique to triage stroke patients for thrombectomy using a noninvasive device.