A Deep Learning Approach for Featureless Robust Quality Assessment of Intermittent Atrial Fibrillation Recordings from Portable and Wearable Devices

Abstract

<jats:p>Atrial fibrillation (AF) is the most common heart rhythm disturbance in clinical practice. It often starts with asymptomatic and very short episodes, which are extremely difficult to detect without long-term monitoring of the patient’s electrocardiogram (ECG). Although recent portable and wearable devices may become very useful in this context, they often record ECG signals strongly corrupted with noise and artifacts. This impairs automatized ulterior analyses that could only be conducted reliably through a previous stage of automatic identification of high-quality ECG intervals. So far, a variety of techniques for ECG quality assessment have been proposed, but poor performances have been reported on recordings from patients with AF. This work introduces a novel deep learning-based algorithm to robustly identify high-quality ECG segments within the challenging environment of single-lead recordings alternating sinus rhythm, AF episodes and other rhythms. The method is based on the high learning capability of a convolutional neural network, which has been trained with 2-D images obtained when turning ECG signals into wavelet scalograms. For its validation, almost 100,000 ECG segments from three different databases have been analyzed during 500 learning-testing iterations, thus involving more than 320,000 ECGs analyzed in total. The obtained results have revealed a discriminant ability to detect high-quality and discard low-quality ECG excerpts of about 93%, only misclassifying around 5% of clean AF segments as noisy ones. In addition, the method has also been able to deal with raw ECG recordings, without requiring signal preprocessing or feature extraction as previous stages. Consequently, it is particularly suitable for portable and wearable devices embedding, facilitating early detection of AF as well as other automatized diagnostic facilities by reliably providing high-quality ECG excerpts to further processing stages.</jats:p>

Identification of Transient Noise to Reduce False Detections in Screening for Atrial Fibrillation

“…The CNN was better in discriminating high-quality from low-quality ECGs than the method in Clifford et al (2012). The percentage of segments labeled as AF when classified as high quality were presented in Huerta-Herraiz et al (2020), but no information on AF detection performance.…”

“…CNNs have found their way into various ECG applications, including arrhythmia detection (Rubin et al, 2018;Yıldırım et al, 2018;Hannun et al, 2019;Niu et al, 2020), AF detection (Andersen et al, 2019;Dang et al, 2019;Fujita and Cimr, 2019), heartbeat classification (Kiranyaz et al, 2016), QRS detection (Silva et al, 2020), and signal quality assessment (Huerta-Herraiz et al, 2020). Concerning the approach taken to signal quality assessment in Huerta-Herraiz et al (2020), consecutive 5-s ECG segments were inputted to the CNN which assigned a label (high-or low-quality) to each segment; similar segment-based approaches were also investigated in Clifford et al (2012) and Behar et al (2013), but then based on traditional machine learning.…”

“…Concerning the approach taken to signal quality assessment in Huerta-Herraiz et al (2020), consecutive 5-s ECG segments were inputted to the CNN which assigned a label (high-or low-quality) to each segment; similar segment-based approaches were also investigated in Clifford et al (2012) and Behar et al (2013), but then based on traditional machine learning. In the present study, a CNN interleaved with an AF detector is proposed for quality control, differing from Clifford et al (2012), Behar et al (2013), andHuerta-Herraiz et al (2020) insofar as the CNN operates on an event-to-event basis.…”

“…The earlier mentioned CNN-based approach for detecting poor-quality segments (Huerta-Herraiz et al, 2020) analyzed three different databases with ECGs from wearable devices. The CNN was better in discriminating high-quality from low-quality ECGs than the method in Clifford et al (2012).…”

“…The review load can be reduced by identifying and excluding noisy signals segments before further processing is performed. The identification problem has been addressed from a general ECG analysis perspective in many studies (Ghaffari et al, 2010;Clifford et al, 2012;Hayn et al, 2012;Behar et al, 2013;Orphanidou et al, 2015;Daluwatte et al, 2016;Abdelazez et al, 2017;Orphanidou and Drobnjak, 2017;Yaghmaie et al, 2018;Moeyersons et al, 2019;Huerta-Herraiz et al, 2020;Smital et al, 2020), however, only a few studies have done so in relation to AF detection (Oster and Clifford, 2015;Taji et al, 2018;Bashar et al, 2019). Then, the methods for identifying poor-quality segments have been based on comparing the output of two different QRS detectors (one being more sensitive to noise than the other) (Oster and Clifford, 2015), deep belief networks (Taji et al, 2018), and time-frequency analysis combined with subband decomposition of the ECG signal (Bashar et al, 2019); the latter two studies did not rely on QRS detection.…”