A Unified ECG Paper Digitization Framework with R-peak Detection for RR Interval Analysis and Deep ECG Learning Applications

Electrocardiogram (ECG) signals are commonly acquired using the ECG recording system which produces printed ECG waveforms with essential ECG parameters and patient information in most of the hospitals and clinics. To preserve old ECG records with essential clinical information (arrhythmias) and utilize large volumes of paper ECGs for deep learning based ECG analysis systems, there is a demand for an ECG paper digitization method to extract digitized ECG signals from printed ECG papers. In this paper, we present a fully-automated ECG paper digitization framework that generates one-dimensional ECG signals and detects R-peaks, which can enable automated ECG signal analysis under low-resource settings. The proposed framework includes the major steps of: preprocessing, edge detection, ECG lead segmentation, 1-D ECG signal generation, and R-peak detection for beat-to-beat interval (BBI) analysis. We evaluate the performance of the proposed framework using scanned ECG papers with different image qualities (low-quality scans, skewed ECG papers, dark colors, and different ECG paper colors), ECG papers with different kinds of PQRST complexes and numbers of leads in addition to the printed texts on the ECG paper. The performance is validated in terms of number of the R-peaks and the Pearson correlation coefficient (PCC) metric. The proposed framework achieves a Pearson correlation coefficient greater than 0.93. Evaluation results demonstrate that the proposed unified framework with R-peak detection has great potential in beat-to-beat interval analysis and deep learning based ECG analysis applications.