Abstract 156: Enhancing QRS Detection Using the Automated Electrocardiogram Selection of Peaks (AESOP) Algorithm
Background: Real-time heart-rate-complexity and -variability (HRC, HRV) analysis in critically ill or injured patients will require accurate, automated detection of QRS complexes. Despite the fact that numerous QRS detection algorithms now exist, accurate detection remains a practical challenge. Through machine learning and fusion, we combined the strengths of several best QRS detection algorithms in order to develop a more optimal real-time solution to the QRS detection problem.
Methods: We implemented several best QRS detection algorithms — Hamilton-Tompkins, Afonso-Tompkins-Nguyen-Luo, Suppappola-Sun, and Christov algorithms — in software, and applied machine learning in order to fuse their outputs. Fusion commences with these 4 algorithms as inputs and returns final decisions as outputs in approximate real time. Central to the fusion scheme, the algorithm selects the mode R-R interval (RRI), or, alternatively, the RRI closest to the mean value of the previous 7 RRIs within a given time frame. We call this fusion process and resulting algorithm the Automated Electrocardiogram Selection of Peaks (AESOP) algorithm. We validated all results by testing records against human-verified RRIs, and consequently, R wave peaks.
Results: Our algorithm required less than 6 seconds to analyze one half-hour record of the MIT-BIH Arrhythmia Database on an Intel® Core™ Duo CPU E7500 at 2.93 GHz. For 48 records in this database, the AESOP algorithm achieved a sensitivity (Se) of 93.8% and a positive predictive value (+P) of 97.5%, thereby outperforming each of its component algorithms. In addition, for 60 trauma patient records produced at the U.S. Army Institute of Surgical Research (USAISR), the AESOP algorithm achieved an overall Se of 99.2% and a +P of 99.4%.
Conclusions: By fusing several best QRS detection algorithms, AESOP employs the strengths of each algorithm and develops a more robust and optimal real-time solution to the problem of detecting QRS complexes within an ECG signal, especially in an ambulatory or hospital environment. Following validation in larger patient datasets, AESOP will now be integrated into a real-time HRC and HRV software program for decision support and triage in critically ill and trauma patients.
- © 2010 by American Heart Association, Inc.