The objective of this work was to benchmark different deep learning architectures for noise detection against cardiac arrhythmia episodes recorded by pacemakers and implantable cardioverter-defibrillators (PM/ICDs) and transmitted for remote monitoring. Up to now, most signal processing from ICD data has been based on classical hand-crafted algorithms, not AI or DL-based ones. The database consist of PM/ICD data from 805 patients representing a total of 10471 recordings from three different channels: ...
The objective of this work was to benchmark different deep learning architectures for noise detection against cardiac arrhythmia episodes recorded by pacemakers and implantable cardioverter-defibrillators (PM/ICDs) and transmitted for remote monitoring. Up to now, most signal processing from ICD data has been based on classical hand-crafted algorithms, not AI or DL-based ones. The database consist of PM/ICD data from 805 patients representing a total of 10471 recordings from three different channels: the right ventricular (RV), the right atria (RA), and the shock channel. Four deep learning approaches were trained and optimized to classify PM/ICDs' records as actual ventricular signal vs noise episodes. We evaluated the performance of the different models using the F2 score. Results show that the use of 2D representations of 1D signals led to better performances than the direct use of 1D signals, suggesting that the detection of noise takes advantage of a spectral decomposition of the signal, which remains to be confirmed in other contexts. This study proposes deep learning approaches for the analysis of remote monitoring recordings from PM/ICDs. The detection of noise allows efficient management of this large daily flow of data.
+