Session PC6.1

Performance and Productivity Benefits Using Multi-Core Processors for the Analysis of Digital Long-Term ECG Recordings

T Hilbel*, RL Lux, J Dietzsch, HA Katus

University of Heidelberg
Heidelberg, Germany

The modern Holter recorders allowing the acquisition of 12 lead ECGs with a sampling rate of 1KHz or higher and a resolution of 16bit over more than 24h. While the large volume of data can be easily stored on flash memory cards the analysis of these biosignals requires a lot of calculation power and network bandwidth. In general, processing time is important when hundreds of digital recordings from large study groups need to be analyzed. The objective of the following investigation is to address the question: Can performance and productivity benefits in ECG analysis be achieved using multi-core processor technology? Because these processors have two or more processing cores they can perform parallel processing under standard operating systems without special programming requirements.
Methods: The relative performance changes in execution time of an ECG analysis program that calculates RR-Intervals, HRV parameters, QT-Duration and Eigenvectors from 1KHz, 16 bit 24 hour Holter data was compared when analyzing 24h of ECG as a single program on a dual-core machine or as several parallel program instances running at the same time under Windows XP®. In the single program test the program analyzed hours 1 – 24 of a high resolution 12-lead Holter recording. In the two program test, the same program was started twice. One program was set to analyze hour 1 – 12 of the data while the second was set to analyze hour 13 – 24 of the biosignals at the same time. So each of the two programs could use one of the processor cores at the same time. The performance was also measured on a Windows 2003® server with two separated Intel XEON® dual-core processors. In the server setting the same program was started up to four times under the assumption that the server provides up to 4 processor cores.
Results: An ECG analysis program can take advantage of a dual-core processor and ran 44% faster when it was running twice while analyzing the first 12h in one program instance and hour 13h – 24h in a second program instance. Running the program in parallel up to 4 times on a server with 2 dual-core processors. Two (2*12h of recording time(RT)), three (3*8h RT) or four (4*6h RT) parallel instances of the same program shortened the time of the biosignal analysis by 44 %, 46 % and 54 % respectively. It is notable that the relative effect of acceleration of this test was greatest when running two programs in parallel, while an additional third or forth process, does not contribute more than about 10% in shortening of scientific calculation.
Conclusion: Running the same program multiple times parallel on modern multi-core processors can accelerate the scientific processing speed considerable. This can be a great benefit for the analysis of long-term ECG recordings with high data volumes. In the near future the benefits of multi core and modern IT techniques like parallel computing, grid computing or distributed computing should be considered for large volume ECG data system.

(Abstract Control Number: 351)