Reliable ECG analysis using recognition scores from multiple deep neural networks

The adoption of contactless patient monitoring has surged in response to the COVID-19 pandemic. After implementing remote electrocardiogram monitors, systematic management techniques are necessary to analyze extensive data and identify measurement errors. During the electrocardiogram measurement process, motion artifacts that may occur are often challenging to distinguish from the electrocardiogram itself because they share similar characteristics in terms of amplitude and frequency. To address these challenges, six deep neural networks capable of recognizing the features of normal electrocardiograms were developed. An algorithm that determines parameters representing the characteristics of electrocardiograms and identifies abnormal waveforms was created by combining the results of six deep neural networks. Waveforms from 10 patients were analyzed, and the differences between leads and individuals were quantified. Additionally, a recognition score was introduced to distinguish highly reliable electrocardiogram parameters while filtering out unreliable values caused by motion artifacts and measurement errors.