Mohammad I Daoud and James C Lacefield 2009 Phys. Med. Biol. 54 5173 doi:10.1088/0031-9155/54/17/007
Mohammad I Daoud1,2 and James C Lacefield2,3
Show affiliationsComputational modeling is an important tool in ultrasound imaging research, but realistic three-dimensional (3D) simulations can exceed the capabilities of serial computers. This paper uses a 3D simulator based on a k-space method that incorporates relaxation absorption and nonreflecting boundary conditions. The simulator, which runs on computer clusters, computes the propagation of a single wavefront. In this paper, an allocation algorithm is introduced to assign each scan line to a group of nodes and use multiple groups to compute independent lines concurrently. The computational complexity required for realistic simulations is analyzed using example calculations of ultrasonic propagation and attenuation in the 30–50 MHz band. Parallel efficiency for B-mode imaging simulations is evaluated for various numbers of scan lines and cluster nodes. An aperture-projection technique is introduced to simulate imaging with a focused transducer using reduced computation grids. This technique is employed to synthesize B-mode images that show realistic 3D refraction artifacts. Parallel computing using 20 nodes to compute groups of ten scan lines concurrently reduced the execution time for each image to 18.6 h, compared to a serial execution time of 357.5 h. The results demonstrate that fully 3D imaging simulations are practical using contemporary computing technology.
General scientific summary. Three-dimensional (3D) simulations of ultrasonic wave propagation are valuable tools for ultrasound imaging research, but detailed full-wave simulations can exceed the capabilities of contemporary serial computers. This paper demonstrates the feasibility of imaging simulations using a k-space (spatial frequency domain) method implemented on parallel computer clusters to compute linear 3-D propagation through weakly scattering inhomogeneous media without employing simplifying assumptions. An analysis of computational complexity is performed to guide trade-offs between the computational grid size, number of time steps, and accuracy of the simulation results. A processor allocation algorithm is introduced to efficiently partition independent scan lines among computer cluster nodes. These steps lead to substantial improvement in the practicality of these simulations: the computation time for an example B-mode image is reduced from 357.5 h with a serial computer to 18.6 h by using 20 processor nodes to compute groups of 10 scan lines concurrently.
Issue 17 (7 September 2009)
Received 1 January 2009, in final form 12 July 2009
Published 11 August 2009
Mohammad I Daoud and James C Lacefield 2009 Phys. Med. Biol. 54 5173
Adam C Waspe et al 2007 Phys. Med. Biol. 52 1863
E B Mpemba and D G Osborne 1969 Phys. Educ. 4 172
G. Iori et al 1994 Europhys. Lett. 25 491
L Maxim and J P van der Sluijs 2010 Environ. Res. Lett. 5 014006
Yuping He et al 2008 Nanotechnology 19 465602
J-G Fan and Y-P Zhao 2008 Nanotechnology 19 155707
J-G Fan et al 2008 Nanotechnology 19 045713
J-G Fan et al 2004 Nanotechnology 15 501
Kazuyuki Tanaka and D M Titterington 2007 J. Phys. A: Math. Theor. 40 11285