CAJ | 학술논문

研究超声弹性成像零相位估计算法的并行化计算问题。针对标准的零相位估计算法不易并行实现的特性，提出零相位估计算法的并行计算框架。首先通过互相关算法计算初步的位移并作为零相位估计算法引导位移，然后再使用二维零相位估计算法计算最终每一个估计点位移。仿真结果显示，该并行计算框架在生成的弹性图像的信噪比和对比度噪声比上能获得与标准方法非常接近的性能评价，同时与标准方法的CPU实现相比，该方法的GPU实现有效地提高了零相位估计算法的计算速度，其加速比可达7倍。研究表明，该算法框架不仅能有效地并行计算，同时能够保证得到高质量的弹性图。
연구초성탄성성상령상위고계산법적병행화계산문제。침대표준적령상위고계산법불역병행실현적특성，제출령상위고계산법적병행계산광가。수선통과호상관산법계산초보적위이병작위령상위고계산법인도위이，연후재사용이유령상위고계산법계산최종매일개고계점위이。방진결과현시，해병행계산광가재생성적탄성도상적신조비화대비도조성비상능획득여표준방법비상접근적성능평개，동시여표준방법적CPU실현상비，해방법적GPU실현유효지제고료령상위고계산법적계산속도，기가속비가체7배。연구표명，해산법광가불부능유효지병행계산，동시능구보증득도고질량적탄성도。
This paper presents a parallel phase zero estimation framework to compute the axial tissue displacements. The framework includes two independent steps: coarse motion estimates and motion estimate refinement. For two frames of baseband signals, coarse motion estimates are first computed by using cross correlation function. The obtained motion estimates are used as initial guess for phase zero estimation. Then, the 2D phase zero estimation method is employed to refine the motion estimates. The experimental results illustrated that the two-step parallel estimation strategy is capable of generating fine motion estimation and is also suitable for parallel computation. Compared with the standard phase zero estimation method based on CPU, the proposed parallel estimation framework based on GPU achieved 7 times speed-up. In conclusion, the proposed parallel estimation framework not only obtains the fine elastogram, but also improves the computational efficiency.