Fast ultrasound beam prediction for linear and regular two- dimensionalarrays

Mario Hlawitschka, Robert J. McGough, Katherine W. Ferrara, Dustin E. Kruse

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Scopus citations


Real-time beam predictions are desirable for ultrasound therapy guidance,as treatment planning requires individualized computations. To address thelong-standing issue of the computational burden associated with calculating theacoustic field in large volumes, we use modern graphics processing units toaccelerate the calculation of near-field pressure fields for therapeuticultrasound arrays. First, we accelerate field computations for singlerectangular pistons by employing the Fast Near-field Method (FNM) to accuratelyand efficiently estimate the complex near-field wave patterns in homogeneousmedia, and results are compared to the Rayleigh-Sommerfeld method. We thencompute beam patterns of 1D and 2D piston arrays using pre-calculated beampatterns from a single piston. Our results show that these algorithms canbenefit greatly from GPU computing hardware, exceeding a modern CPU by a factorof over 100. For a single rectangular piston, the FNM is able to calculate a0.01% accurate field within 30 ns per field point. Furthermore, we demonstratecalculation speeds for arrays of up to 11.5 109 field points persecond. Beam volumes containing 2563 field points are calculatedwithin one second for arrays containing up to 512 pistons, thus, facilitatingfuture real-time thermal dose predictions.

Original languageEnglish (US)
Title of host publicationProceedings - IEEE Ultrasonics Symposium
Number of pages4
StatePublished - 2010
Event2010 IEEE International Ultrasonics Symposium, IUS 2010 - San Diego, CA, United States
Duration: Oct 11 2010Oct 14 2010


Other2010 IEEE International Ultrasonics Symposium, IUS 2010
Country/TerritoryUnited States
CitySan Diego, CA

ASJC Scopus subject areas

  • Acoustics and Ultrasonics


Dive into the research topics of 'Fast ultrasound beam prediction for linear and regular two- dimensionalarrays'. Together they form a unique fingerprint.

Cite this