Electrocardiographic Signal Quality Assessment Without Morphology Analysis

The electrocardiogram (ECG) is the primary screening method of the cardiologist and is fundamental to understand the electrical activity of the heart. Signal interference sources that are non-issues in medical recordings become significant sources of noise in wearable devices recordings using dry electrodes. It is crucial to develop methods to assess recording quality in order to minimize unreliable data and provide cleaner raw recordings to algorithms such as machine learning. In this paper a methodology for classification of the most common signal distortion sources affecting dry electrodes ECG recordings is presented; classification is not reliant on absolute signal analysis and ECG morphology, making it suitable for applications where the system cannot directly analyze the ECG due to regulatory restrictions. The methodology was successfully validated with a commonly used dataset - Computing in Cardiology Challenge 2011 - as well as with data obtained in real driving conditions using the CardioWheel system [1].