Analyzing long wearable electrocardiogram recordings: R-spike detection and cycle pattern analysis

This paper considers very long ECGs (electrocardiograms), looking mainly at leads with salient QRS complexes, to identify ECG pattern dynamics of potential interest to clinicians, especially for prevention and in the context of personalized medicine. The raw data comes from portable recording devices running for up to twenty-four hours, resulting in fairly noisy segments. Minimal intervention is applied to the data, involving only a linear convolution filter or the Fasano–Villani method to correct for baseline drift. An automatic procedure is introduced for detecting zones devoid of meaningful ECG signals, and a new methodology for R-spike detection is proposed, which leverages the narrow amplitude range of R-spikes relative to the wandering baseline. This new method shows favorable comparisons with the Pan–Tompkins algorithm. The cycles from the good zones of the ECG are grouped into coherent and non-coherent blocks, from which certain averages of coherent blocks are identified as patterns. Graphic devices are introduced to visualize large sequences of patterns that exhibit morphological changes based on their representation. The findings demonstrate an automatic procedure can effectively detect possible anomalies in portable ECG recordings.