ECG-based pulse detection during cardiac arrest using random forest classifier.

Area of Science:

• Biomedical Engineering
• Cardiology
• Medical Devices

Background:

• Sudden cardiac arrest (SCA) is a major cause of death globally.
• Accurate pulse detection is critical for SCA management and patient survival.
• Current methods may require manual intervention or are less integrated into resuscitation devices.

Purpose of the Study:

• To develop and validate an automated pulse detection method using only electrocardiogram (ECG) signals.
• To assess the algorithm's accuracy and efficiency during cardiopulmonary resuscitation (CPR).
• To enable integration into automated external defibrillators (AEDs) for improved clinical practice.

Main Methods:

• Utilized a random forest classifier trained on ECG features (time, frequency, slope, regularity).
• Analyzed ECG data from 191 cardiac arrest patients (1177 segments: 796 with pulse, 381 without).
• Employed a leave-one-patient-out cross-validation and patient-wise bootstrap for robust performance estimation.

Main Results:

• Achieved a mean sensitivity of 88.4% (±1.8%) and specificity of 89.7% (±1.4%) for pulse detection.
• The algorithm requires only 4-second ECG segments for accurate analysis.
• Demonstrated high accuracy in distinguishing pulse presence/absence during simulated CPR.

Conclusions:

• The developed ECG-based algorithm offers accurate, automated pulse detection during CPR.
• Its ability to use short ECG segments and potential for AED integration can minimize critical therapy interruptions.
• This technology holds promise for improving survival rates in patients experiencing sudden cardiac arrest.