Avive’s AED Machine Learning Algorithm Exceeds AHA’s Performance Recommendations
Avive was selected to present a study on a novel machine-learning algorithm for arrhythmia classification via a podium presentation at the American College of Cardiology Conference 2020.
The presentation and audio can be found here.
SAN FRANCISCO, CA – (March 30, 2020) – Avive Solutions, Inc (“Avive” or the “Company”), a developer of a next-generation Automated External Defibrillator (“AED”) and innovative cardiac arrest response solutions, was selected to present a study on a novel machine-learning algorithm for arrhythmia classification via a podium presentation at the American College of Cardiology Conference 2020.
The study is entitled, “Development and Validation of a Convolution Neural Network Algorithm for Shockable Arrhythmia Classification Within A Next-Generation Automated External Defibrillator,” and the published abstract can be found at onlinejacc.org.
Authors of this study assembled one of the world’s largest databases of electrocardiogram (ECG) rhythms from both adult and pediatric patients, procured from 13 separate sources, including publicly available databases, healthcare centers, and privately-owned databases. ECGs were adjudicated by varying sets of 3 cardiologists, with strict criteria for the required agreement on rhythm classification, and were subsequently divided into distinct sets of data used for development and testing of the algorithm.
“The primary objective of the study was to evaluate the performance of the novel algorithm for the detection of shockable arrhythmias. A secondary objective was to show that the algorithm’s accuracy meets the American Heart Association’s performance recommendation for arrhythmia analysis algorithms,” states Dr. Sanjeev Bhavnani, Division of Cardiology and Healthcare Innovation at Scripps Clinic, the study’s principal investigator.
The authors found that the neural network’s performance exceeded the American Heart Association’s (AHA) performance recommendations for sensitivity and specificity for each of the 12 specific rhythm categories, and achieved promising results on metrics used to evaluate machine learning algorithms.
“Overall, the convolution neural network demonstrated greater than 99% diagnostic accuracy for shockable and non-shockable rhythms. The direct application of this innovation embedded in a new AED device is unique, and offers significant implications to improve sudden cardiac arrest process of care,” states Bhavnani.
Avive’s machine learning approach to arrhythmia classification is a promising development for rapid, versatile, and accurate rhythm diagnoses.
“While already exceeding the AHA’s performance criteria, we are continuing to optimize our algorithm even further,” states Rory Beyer, Co-Founder, and CEO of Avive. “We believe that this novel framework will allow us to continually improve performance over the wide range of patients and situations that our AED will need to handle, especially as we collect additional data in the field. In the end, our hope is that these improvements will save more lives.”
The study’s collaborators were across several institutions, including: Scripps Clinic, University of Pittsburgh, University of Kentucky, University of Utah, Massachusetts Institute of Technology, Imperial College of London, and the University of San Diego.