Assessing Electrocardiogram Quality: A Deep Learning Framework For Noise Detection And Classification
Year: 2025
Type: conference-paper
Abstract: The electrocardiogram (ECG) is essential for diagnosing cardiovascular conditions, yet ECG signals are highly prone to noise, which reduces their reliability, especially in wearable devices and long-term monitoring applications. Traditional noise detection methods achieve limited accuracy, and deep learning (DL) has emerged as a promising solution. Current DL approaches are focused on binary classification (noisy versus clean) and lack more detailed quality information. This paper explores these challenges by developing a DL model capable of performing a comprehensive signal quality assessment (SQA) through the detection of noisy segments and the classification of specific noise types in the ECG data. To accomplish this, a labeled dataset is generated by injecting controlled noise into clean signals. A model based on gated recurrent units (GRUs) is developed to handle the time-series nature of the data, identifying and classifying common noise types. The model achieved a high accuracy of 92.86 % for electrode motion (EM) noise followed by an accuracy of 92.05 % and 81.55 % for baseline wander (BW) and muscle artifacts (MA), respectively. Despite challenges with overlapping noise, the model consistently maintains high recall and precision rates, proving effective in identifying noise when present. Compared to state-of-the-art DL approaches, the proposed model significantly outperforms previous methods, achieving 99.72 % accuracy when conducting a binary classification of the signals. This thesis offers a significant advancement in ECG signal processing by providing a more detailed understanding of noise beyond binary classification. Future improvements comprise adding the quantification of noise to predict the level of contamination. The work has the potential to impact clinical practice, long-term monitoring, and self-monitoring technologies, benefiting medical professionals, researchers, and wearable device users alike.
Authors: Márcia Monteiro, Mariana Dias, Hugo Gamboa
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W4407929973