Abstract

This article discusses the capabilities of layered ‘‘2.5D’’ schemes in ultrasound tomography, in which the emitters and detectors are located in the same plane. The primary application of this method is medical imaging technology for early-stage breast cancer diagnosis. The inverse problem of tomographic image reconstruction is posed as a coefficient inverse problem for the wave equation, where the wave propagation velocity is an unknown function of two coordinates. In this formulation, the problem has much less computational complexity than in a 3D formulation, in which the wave velocity is reconstructed as a function of three coordinates. Nevertheless, the inverse problem in the layered ‘‘2.5D’’ formulation is non-linear and requires the use of supercomputers. A practically feasible parallel implementation of the image reconstruction algorithms on a GPU cluster is proposed. This study shows that the layered model is applicable if the imaged objects are close to cylindrical. Refraction of ultrasound waves in the vertical direction cannot be taken into account by the layered model. The more significant is the refraction, the more the reconstructed image differs from reality. The experiments were carried out using the test bench developed at the Scientific Research Computing Centre of Lomonosov Moscow State University.

中文翻译：

---

分层超声层析成像的能力

摘要

本文讨论了超声层析成像中分层的“ 2.5D”方案的功能，其中发射器和检测器位于同一平面上。该方法的主要应用是用于早期乳腺癌诊断的医学成像技术。断层图像重建的反问题是波动方程的系数反问题，其中波传播速度是两个坐标的未知函数。在此公式中，该问题的计算复杂度比在3D公式中要小得多，在3D公式中，波速是根据三个坐标重建的。但是，分层的“ 2.5D”公式中的反问题是非线性的，需要使用超级计算机。提出了一种在GPU集群上实际可行的图像重建算法并行实现方法。这项研究表明，如果成像对象靠近圆柱，则分层模型是适用的。分层模型无法考虑超声波在垂直方向上的折射。折射值越大，则重建图像与现实的差异就越大。实验是使用罗蒙诺索夫莫斯科国立大学科学研究计算中心开发的测试台进行的。重建的图像与现实的差异越大。实验是使用罗蒙诺索夫莫斯科国立大学科学研究计算中心开发的测试台进行的。重建的图像与现实的差异越大。实验是使用罗蒙诺索夫莫斯科国立大学科学研究计算中心开发的测试台进行的。