Abstract 13968: Automated Pulse Detection is Achieved by Analysis of the Activity Level and Periodicity of Accelerometer Signals Acquired at the Carotid Site

Introduction: Manual palpation during cardiopulmonary resuscitation (CPR) is the gold standard for pulse assessment by professional rescuers. However, this method is unreliable and time-consuming. A fast, reliable, and objective assessment of pulse presence remains an unmet need. Hypothesis: Automatic pulse presence detection is feasible by assessing the activity level and periodicity of accelerometer (ACC) signals acquired at the carotid site during CPR. Methods: A 2-step classification algorithm was developed using the activity level and periodicity of ACC signals to identify 4 categories: ‘Pulse’ (low activity and periodic), ‘No-pulse’ (low-activity and aperiodic), ‘CPR’ (high activity and periodic) or ‘Artifact’ (high activity and aperiodic). This approach uses 3 features extracted from non-overlapping 3 s windows of data: signal energy, peak prominence and rate. The algorithm analyzes the norm of the 3-axis ACC signals, computed after normalization by the gravitational constant and filtering between 0.5 and 12 Hz. The classifier was trained on data from 5 male patients (age range = 45-70 years) undergoing CPR in a hospital, with ACC signals sampled at 125 Hz. The performance was assessed by benchmarking against manual classification of artefacts, chest compressions, pulses, and pulse absence by visual inspection of all signals done by experts in the field. Results: Activity classification in all cases yielded sensitivity and specificity >90.3% and >88.8%, respectively (Fig. 1). For CPR classification the sensitivity and specificity was >71.9% and >71.6%, respectively, while for pulse classification the sensitivity and specificity was >51.9% and >39.5%, respectively. Conclusions: Automated pulse presence detection based on the activity level and periodicity of ACC signals at the carotid site is feasible with reasonable sensitivity and specificity. Automatic ACC-based pulse palpation could provide objective pulse detection support during CPR.