Quantification of speed-up and accuracy of multi-CPU computational flow dynamics simulations of hemodynamics in a posterior communicating artery aneurysm of complex geometry. (12th March 2013)

• Record Type: Journal Article
• Authors: Karmonik, Christof; Yen, Christopher; Gabriel, Edgar; Partovi, Sasan; Horner, Marc; Zhang, Yi J; Klucznik, Richard P; Diaz, Orlando; Grossman, Robert G
• Is Part Of: Journal of neurointerventional surgery. Volume 5(2013)Supplement 3
• Page Start: iii48
• Page End: iii55
• Publication Date: 2013-03-12
• Subjects: Aneurysm -- MRI -- Technique; Nervous system -- Surgery -- Periodicals; Cerebrovascular disease -- Surgery -- Periodicals; 617.48
• DOI: 10.1136/neurintsurg-2012-010586
• Languages: English
• ISSNs: 1759-8478

Abstract

Abstract : Background: Towards the translation of computational fluid dynamics (CFD) techniques into the clinical workflow, performance increases achieved with parallel multi-central processing unit (CPU) pulsatile CFD simulations in a patient-derived model of a bilobed posterior communicating artery aneurysm were evaluated while simultaneously monitoring changes in the accuracy of the solution. Methods: Simulations were performed using 2, 4, 6, 8, 10 and 12 processors. In addition, a baseline simulation was obtained with a dual-core dual CPU computer of similar computational power to clinical imaging workstations. Parallel performance indices including computation speed-up, efficiency (speed-up divided by number of processors), computational cost (computation time × number of processors) and accuracy (velocity at four distinct locations: proximal and distal to the aneurysm, in the aneurysm ostium and aneurysm dome) were determined from the simulations and compared. Results: Total computation time decreased from 9 h 10 min (baseline) to 2 h 34 min (10 CPU). Speed-up relative to baseline increased from 1.35 (2 CPU) to 3.57 (maximum at 10 CPU) while efficiency decreased from 0.65 to 0.35 with increasing cost (33.013 to 92.535). Relative velocity component deviations were less than 0.0073% and larger for 12 CPU than for 2 CPU (0.004±0.002%, not statistically significant, p=0.07). Conclusions: Without compromising accuracy, parallel multi-CPU simulation reduces computing time for the simulation of hemodynamics in a model of a cerebral aneurysm by up to a factor of 3.57 (10 CPUs) to 2 h 34 min compared with a workstation with computational power similar to clinical imaging workstations.