A deep learning artifact removal method for CPR continuity throughout the shock decision in AEDs - PROJECT SUMMARY Out-of-hospital cardiac arrest (OHCA) affects more than 475,000 people in the United States each year. Major determinants of survival from OHCA include continuous high quality cardiopulmonary resuscitation (CPR) coupled with timely defibrillatory shock, if warranted, based on accurate shock decisions. According to the American Heart Association (AHA), delivering continuous CPR along with rapid shock by automated external defibrillator (AED) are two main interventions that are likely to restart the arrested heart. However, the mechanical activity of chest compressions during CPR performance induces severe motion artifact in the electrocardiogram (ECG) signal. Thus, an accurate rhythm analysis and pulse checks require pauses in chest compressions. Unfortunately, frequent or prolonged interruptions in chest compressions lead to adverse outcomes after cardiac arrest. Despite previous attempts to work with CPR-contaminated ECG in AEDs, resuscitation performance has been hampered by two major factors: 1) there is no AHA-approved validated shock advisory algorithm during CPR performance, and 2) there is insufficient cardiac arrest data with CPR, which has hampered advancements using deep learning approaches to remove CPR artifacts. To overcome these limitations, we propose to develop a database of CPR artifact data already collected from commercial (Defibtech) defibrillators, and inpatient data from the University of Michigan and Massachusetts General Hospital. Using this database, we will evaluate and optimize the performance of our promising CPR artifact removal algorithm based on deep learning. Once the CPR artifact removal and shockable rhythm detection models are validated, we will integrate our algorithms into the existing ECG monitoring system at the University of Michigan medical center and evaluate their performance in real-world scenarios in real time using new CPR data. Successful completion of our aims will allow continuous evaluation of cardiac rhythms during AED implementation, reduce interruptions in CPR, and improve clinical outcomes. Additionally, a large CPR ECG database will catalyze research focused on improving defibrillator performance, identifying key factors for a successful resuscitation, and reducing morbidity and mortality from cardiac arrest.
