Multi-class Detection of Arrhythmia Conditions Through the Combination of Compressed Sensing and Machine Learning

• Published in: Biomedical Engineering Systems and Technologies Vol. 1710; pp. 213 - 235
• Main Authors: Rosa, Giovanni, Russodivito, Marco, Laudato, Gennaro, Colavita, Angela Rita, Vito, Luca De, Picariello, Francesco, Scalabrino, Simone, Tudosa, Ioan, Oliveto, Rocco
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2023; Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Arrhytmia; Automatic detection; Compressed sensing; ECG; Machine learning
• ISBN: 9783031206634; 3031206630
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-031-20664-1_12

Medical technologies in the form of wearable devices are an integral part of our daily lives. These devices are devoted to acquire physiological data to provide personal analytics and to assess the physical status of assisted individuals. Nowadays, thanks to the research effort and to the continuously evolving technologies, telemedecine plays a crucial role in healthcare. Electrocardiogram (ECG) is one of the source signal that has been widely involved in telemedicine and therefore the need for a quick and precise screening of ECG pathological conditions has become a priority for the scientific community. Based on the above motivation, we present a study aimed at evaluating the applicability of an highly accurate detector of arrhythmia conditions to be used in combination of a compressed version of the ECG signal. The advantage of using a technique of Compressed Sensing (CS) relies on a faster detection of the approach, due to the lower complexity of the method’s workflow. We conducted an experimental study to determine if such a detector, working on compressed ECG signal, can achieve comparable results with the original approach applied to the uncompressed signal. The results demonstrated that with a Compression Ratio equal to 16 it is possible to achieve classification metrics around 99 $$\%$$ , therefore showing a high suitability of the approach to be involved in contexts of Compressed ECG.