Automated ECG Quality Assessment Using Deep Learning

Thumbnail Image
Issue Date
Journal Title
Journal ISSN
Volume Title
Cardiovascular diseases are becoming more prevalent with the increase in the global population. Due to this, there is a greater need for diagnostic techniques like ECGs. ECG recordings need to be checked for quality to ensure that only non-noisy recordings are used for analysis. As this is a time-consuming process, automated ECG quality assessment techniques have been developed. Previous research on automated ECG quality assessment techniques used unbalanced datasets or noise datasets to create balanced datasets to train their models. Our model was trained on balanced datasets generated without artificially added noise to generate extra noisy recordings. The proposed CNN-LSTM model has an accuracy of 97.27% on the test set and an accuracy of 98.04% on the external test set. The model was trained using real noise sources which likely results in better real-life performance and usability of the models compared to models trained and tested on simulated data. The proposed CNN-LSTM was incorporated into a toolbox for easy usage. The layout of the toolbox was changed to be more user-friendly. Furthermore, the toolbox is fullscreen and is equipped with faster datastructures.
Faculteit der Sociale Wetenschappen