Abstract 14065: Computational Intelligence: A Novel Way of Uncovering Hidden Information

• Published in: Circulation (New York, N.Y.) Vol. 144; no. Suppl_1; p. A14065
• Main Authors: Bender, Dieter, Morgan, Ryan W, Nadkarni, Vinay M, Berg, Robert A, Kilbaugh, Todd J, Sutton, Robert M, Nataraj, C
• Format: Journal Article
• Language: English
• Published: Lippincott Williams & Wilkins 16.11.2021
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.144.suppl_1.14065

BackgroundThis study describes a novel approach to identify diagnostic properties in an electrocardiogram (ECG) waveform to classify cardiac arrest etiologies in human patients. Through an interdependent optimization algorithm linking machine learning and signal processing, the technique allows extensive electrocardiogram analysis, retrieving physiologically relevant and demonstrative characteristics embedded in the waveforms across a population. MethodsSecondary analysis of electrocardiography data collected from swine models of either primary VF (n=17) or secondary asphyxia-associated VF (7 minutes of asphyxia prior to VF induction; n=13). Carrying out the analysis, we develop a mutually related algorithm utilizing wavelet synchrosqueezed transform (WSST) and Machine Learning Tree-Classifier (ML-TC). The extraction of the waveform characteristics (WSST) and the learning stage of the ML algorithm (ML-TC) are mutually related through a genetic optimization method to extract the most effective signal characteristics (and their attributes) with respect to the classification performance. This method thus focuses the WSST analysis on a particular frequency bandwidth of the ECG waveform and previously unknown time-dependent frequency patterns that were consistent across all subjects. ResultsThe innovative approach uncovers previously unknown frequency patterns in the ECG waveform. The proposed method achieves an average classification accuracy of 100% when using 35-second ECG segments and 93% while using 5-second ECG segments, this being the best results ever obtained for this specific problem. ConclusionThis kind of analytical research demonstrates the benefits of computational intelligence, such as machine learning and optimization. It presents a novel method of viewing frequently used data in electrocardiography and illustrates how new hidden information can be obtained to assist in patient diagnosis and prognosis.