high performance computing on graphics processing units: hgpu.org

New elastographic techniques such as poroelastography and viscoelasticity imaging aim at imaging the temporal mechanical behavior of tissues. These techniques usually involve the use of curve fitting methods being applied to noisy data to estimate new elastographic parameters. As of today, however, current elastographic implementations of poroelastography and viscoelasticity imaging methods are in general too slow and not optimized for clinical applications. Furthermore, image quality performance of these new elastographic techniques is still largely unknown due to a paucity of data and the lack of systematic studies that analyze their performance limitations. In this paper, we propose a new elastographic time constant (TC) estimator, which is based on the use of the least square error (LSE) curve-fitting method and the Levenberg-Marquardt (LM) optimization rule as applied to noisy elastographic data obtained from a material in a creep-type experiment. The algorithm is executed on a massively parallel general purpose graphics processing unit (GPGPU) to achieve real-time performance. The estimator’s performance is analyzed using simulations. Experimental results obtained from poroelastic phantoms are presented as a proof of principle of the new estimator’s technical applicability on real experimental data. The results of this study demonstrate that the newly proposed elastographic estimator can produce highly accurate and sensitive elastographic TC estimates in real-time and at high signal-to-noise ratios.