Digitization of ECG paper records based on density and prediction methods

Abstract

ABSTRACTTraditional 12-lead ECGs have the common problem of lead waveform crossover. Importantly, this problem leads to difficulty in extracting lead waveforms and causes signal distortion during the digitization of ECGs. In this paper, an ECG digitization method that combines density clustering and curve prediction is proposed, and it can effectively solve the issue of lead waveform crossover while also minimizing signal distortion. This method first uses the image preprocessing technique to remove the background mesh, and second, the density and prediction methods are used to solve the lead waveform crossover. Finally, morphological and vertical scanning methods are used to digitize the lead waveform. In addition, experimental verification is carried out on a large quantity of ECG records provided by Yifu Hospital, which is affiliated with Nanjing Medical University. Furthermore, five indicators are adopted for quantitative measurements and comparisons between the reconstructed signals and original waveforms. The comparison results show that the accuracy of the method is 95.5%, and this verifies the effectiveness of the algorithm.CO EDITOR-IN-CHIEF: Yuan, Shyan-MingASSOCIATE EDITOR: Yuan, Shyan-MingKEYWORDS: 12-lead ECGlead waveform crossoverdensity clusteringECG digitalization Nomenclature APR=average P-R intervalAQRS=average QRS intervalAQT=average Q-T intervalAR=average R waveD=lead waveform pixel point coordinate setECG=electrocardiogramHR=heart rateHSV=hue, saturation, valueIo=the calculated result of the original waveformIr=the calculated result of the reconstructed waveformMinPts=minimum number of samples in epsNRMSE=normalized root-mean-square errorPDVD=period distance vertical directionε=distance thresholdDisclosure statementNo potential conflict of interest was reported by the author(s).