Graphic Processing Unit Acceleration of an Electrocardiogram Denoising Process Using a Hybrid Approach

The electrocardiogram (ECG) analysis applications necessitate real-time processing due to the vast volume of data that must be analyzed at high frequencies. Thus, this presents a considerable challenge for researchers. The aim of this study is to employ a heterogeneous architecture in order to parallelize the filtering process of an ECG signal. This involves integrating a central processing unit (CPU) with a graphic processing unit (GPU) in a unified computing framework (CPU-GPU) to decrease computational time. Adapted ECG signal filtering algorithms include discrete wavelet transform (DWT) and adaptive double threshold filter (ADTF) with DWT-ADTF hybridization techniques combining both techniques to provide accurate and better results. The evaluation of our method based on Open Computing Language (OpenCL) to exploit the parallelism and compared to the CPU-based sequential architecture using C/C++, was validated by employing various ECG signals sourced from the MIT-BIH Arrhythmia database, which were sampled at a frequency of 360 Hz. The results indicate a mean execution time of 0.27ms on the CPU-GPU-based parallel architecture and 28ms on the CPU-based sequential architecture. Real-time performance can be attained by utilizing effective parallelization techniques on the CPU-GPU heterogeneous architecture.