Adaptive sampling of the ECG, yet attractive as digital crowd reducing technique, involves a compromise of diagnostic quality and data size. Statistics-based approaches (i.e. compressed sampling [1]) are chal-lenged by the noise and ECG model-based approaches poorly correspond to inter-individual variability. In this work we propose a data resampling method based on local minimization of reconstruction error able to flexi-bly assign a given data stream.
Our method works by establishing fixed number N-1 of samples in tar-get non-uniform representation. Extrema in the original signal are identi-fied and sorted by descending amplitude. Then N/2 samples are assigned to extrema time points in the original ECG what results in N/2-1 non-uniform intervals between them. Next, the remaining N/2-1 samples are assigned one to each interval so as to minimize the distance collected in the interval between real data points and their linear interpolation The algorithm iterates through a loop 20 times, updating either even- or odd-indexed points based on minimum distances and handling edge cases effectively. The proposed method was tested with MIT-BIH Arrhythmia Database (360 sps) and data streams were reduced from 120 to 20 sps. Our results show that with as few as 60 sps the error minimization procedure guaran-tees acceptable distortion level of 0.1%. The distortion level, however, besides unwanted alteration of medical finding also results from reduc-tion of high frequency noise always present in real-life records. Our research also proves that non-uniform ECG sampling may effi-ciently adapt to the local features of the signal with no a priori model. The data stream can be then freely adjusted depending on medical con-tent or occurrence of important cardiac events.[1] Craven, D.; McGinley, B.; Kilmartin, L.; Glavin, M.; Jones, E. Adaptive Dictionary Reconstruction for Compressed Sensing of ECG Signals. IEEE J. Biomed. Health Inform. 2017, 21, 645–654.