Yeung F, Levinson S F, Parker K J
Department of Electrical Engineering, University of Rochester, NY 14642, USA.
Ultrasound Med Biol. 1998 Mar;24(3):427-41. doi: 10.1016/s0301-5629(97)00281-0.
A multilevel motion model-based approach to ultrasonic speckle tracking has been developed that addresses the inherent trade-offs associated with traditional single-level block matching (SLBM) methods. The multilevel block matching (MLBM) algorithm uses variable matching block and search window sizes in a coarse-to-fine scheme, preserving the relative immunity to noise associated with the use of a large matching block while preserving the motion field detail associated with the use of a small matching block. To decrease further the sensitivity of the multilevel approach to noise, speckle decorrelation and false matches, a smooth motion model-based block matching (SMBM) algorithm has been implemented that takes into account the spatial inertia of soft tissue elements. The new algorithms were compared to SLBM through a series of experiments involving manual translation of soft tissue phantoms, motion field computer simulations of rotation, compression and shear deformation, and an experiment involving contraction of human forearm muscles. Measures of tracking accuracy included mean squared tracking error, peak signal-to-noise ratio (PSNR) and blinded observations of optical flow. Measures of tracking efficiency included the number of sum squared difference calculations and the computation time. In the phantom translation experiments, the SMBM algorithm successfully matched the accuracy of SLBM using both large and small matching blocks while significantly reducing the number of computations and computation time when a large matching block was used. For the computer simulations, SMBM yielded better tracking accuracies and spatial resolution when compared with SLBM using a large matching block. For the muscle experiment, SMBM outperformed SLBM both in terms of PSNR and observations of optical flow. We believe that the smooth motion model-based MLBM approach represents a meaningful development in ultrasonic soft tissue motion measurement.
已开发出一种基于多级运动模型的超声散斑跟踪方法，该方法解决了与传统单级块匹配（SLBM）方法相关的内在权衡问题。多级块匹配（MLBM）算法在粗到细的方案中使用可变的匹配块和搜索窗口大小，在保持与使用大匹配块相关的相对抗噪声能力的同时，保留与使用小匹配块相关的运动场细节。为了进一步降低多级方法对噪声、散斑去相关和误匹配的敏感性，已实现了一种基于平滑运动模型的块匹配（SMBM）算法，该算法考虑了软组织元素的空间惯性。通过一系列实验将新算法与SLBM进行了比较，这些实验包括软组织体模的手动平移、旋转、压缩和剪切变形的运动场计算机模拟，以及一项涉及人体前臂肌肉收缩的实验。跟踪精度的度量包括均方跟踪误差、峰值信噪比（PSNR）和光流的盲观测。跟踪效率的度量包括平方差和计算的次数以及计算时间。在体模平移实验中，SMBM算法在使用大匹配块和小匹配块时都成功地匹配了SLBM的精度，同时在使用大匹配块时显著减少了计算次数和计算时间。对于计算机模拟，与使用大匹配块的SLBM相比，SMBM产生了更好的跟踪精度和空间分辨率。对于肌肉实验，SMBM在PSNR和光流观测方面均优于SLBM。我们认为，基于平滑运动模型的MLBM方法代表了超声软组织运动测量中有意义的进展。
