Quality-Based ECG Classification: Deep Learning Approach for Long-Duration Holter Recordings

In this contribution, we propose a deep learning-based method for automatically segmenting and classifying electrocardiogtam (ECG) signal quality from extended Holter recordings. Without relying on accelerometer data, we developed a custom dataset of 187,000 samples, where ECG signals were split into small segments and manually labeled for quality. A convolutional-recurrent neural network was trained to differentiate usable ECG segments, achieving an F1 score of 90.06 %. This automated approach streamlines the processing of long-duration Holter data, allowing clinicians to focus on high-quality, clinically relevant information.