Coherence-based measure of instantaneous ECG noise

This paper presents a coherence-based method for estimation of spatial or temporal variability of leads quality in a multichannel ECG record. The method is dedicated to stress test or Holter analyzers and aimed at providing an objective criterion for local assessment of data reliability (e.g. ST-segment elevation or depression) in presence of variable noise. The procedure starts with heartbeat segmentation followed by Fast Fourier Transform to the frequency domain. Consequently, autospectra and cospectra for each pair of signals are determined and the resulted coherence function is normalized and weighted by the noise spectrum (NS). Finally, a triangular matrix summarizes the coherence power and the ECG sections (beats or channels) are sorted accordingly to the value of Noise Estimate. With average accuracy of noise estimate of 11.23% for regular and of 3.0% for NS-weighted coherence, the method is accurate enough for beat-to-beat noise tracking and for a reliable selection of best channel in a multichannel ECG record.