Parallel computing method for HRV time-domain based on GPU

Abstract

HRV (Heart rate variability, which has a function of prediction for cardiovascular disease) contains a wealth of medical information, rapid extraction and procession of these signals will bring an important meaning for the prevention of heart diseases. Physionet open source project provides a good platform for the research and development of HRV, which also provides demonstration tools for the calculation of HRV. The characteristics of medical signal are real-time and have large volume of data. Conventional serial methods are difficult to meet the requirements of biomedicine, and the parallel method based on multi-core CPU is larger communication overhead. In this paper, we designed some parallel algorithms for the calculation of HRV in time-domain based on the strategy of parallel reduction, compared and analyzed the various optimization methods, and received the highest 38 times speedup compared with serial method.