Design of Low Power Algorithms for Automatic Embedded Analysis of Patch ECG Signals

The diagnosis of cardiac arrhythmias often depends on information from long-term ambulatory electrocardiographic (ECG) monitoring. For several decades, these recordings have been obtained by wired Holter recorders. However, to overcome some of the known disadvantages of the old technologies, several different cable-free wireless patch-type ECG recorders have recently reached the market. One of these recorders is the ePatch designed by the Danish company DELTA. The extended monitoring period available with the patch recorders has demonstrated to increase the diagnostic yield of outpatient ECG monitoring. Furthermore, the patch recorders facilitate the possibility of outpatient ECG monitoring in new clinically relevant areas, e.g. telemedicine monitoring of cardiac patients in their homes. Some of these new applications could benefit from real-time embedded interpretation of the recorded ECGs. Such algorithms could allow the real-time transmission of clinically relevant information to a central monitoring station. The first step in embedded ECG interpretation is the automatic detection of each individual heartbeat. An important part of this project was therefore to design a novel algorithm that was optimized for heartbeat detection in ePatch ECGs and embed the algorithm in the ePatch sensor for realtime analysis. We designed the algorithm based on a novel cascade of computational efficient filters and adaptive thresholds. We evaluated the algorithm on both standard databases and three different manually annotated ePatch databases. We found a very high detection performance with respect to both normal and abnormal beats as well as different types of artifacts arising from different daily life activities (average detection performance on 952,632 manually annotated beats obtained from 198 different patients: Se = 99.86% and P⁺ = 99.74%). This shows the possibilities for the embedded analysis of the ePatch ECGs, and the designed algorithm thus provides a platform for further research in this area. The expected advantages of the patch recorders are, however, unconditionally limited by their ability to record high-quality diagnostic ECGs throughout the recording period. Another main focus of this thesis was therefore to investigate different important clinical aspects of the novel ePatch recorder. To achieve this, we designed two pilot studies that were intended to provide information about the clinical usability of the ePatch ECGs for heart rhythm analysis. In the first pilot study, two medical doctors were asked to provide an individual assessment of the usefulness of 200 ePatch ECG segments for heart rhythm analysis. They found that more than 98% of the segments were useful. The second pilot study was designed as a high level comparison between the diagnostic information that could be extracted from simultaneous recordings obtained with the ePatch recorder and the traditional telemetry equipment. This comparison was conducted by a cardiologist on 11 admitted patients. He found no clinically relevant differences between the information extracted from the two systems. Both pilot studies thus indicate a high potential for the clinical application of ECGs recorded with the ePatch system. To further investigate the general signal quality obtained by the ePatch, we designed a novel algorithm for the automatic estimation of the overall percentage of analyzable time (PAT) in ECGs. The algorithm obtained very high classification performance and is therefore expected to provide a reliable estimation of the overall PAT. We then applied the algorithm to investigate the PAT in 250 different ePatch recordings. We found that 10% of the recordings obtained less than 10% analyzable time, and they were considered as incorrect measurements. For the remaining 90% of the recordings, we found a very high PAT (median: 100% (interquartile range: 97.9% to 100%); mean: (92.4 ± 18.8)%). We therefore didn’t find indications of problems related to the general signal quality obtained by the ePatch recorder. Overall, we thus find a high potential for the application of the ePatch recorder in many different clinical settings in the future.