Reducing Acoustic Inhomogeneity Based on Speed of Sound Autofocus in Microwave Induced Thermoacoustic Tomography
Reducing Acoustic Inhomogeneity Based on Speed of Sound Autofocus in Microwave Induced Thermoacoustic Tomography
复制标题
DOI：
10.1109/tbme.2019.2957535
复制
发表时间：
2019-12
影响因子：
4.6
通讯作者：
Shuangli Liu;Zhu Zheng;Xiaoxuan Sun;Zhiqin Zhao;Yuanjin Zheng;Huabei Jiang;Xiaozhang Zhu;Q. Liu
中科院分区：
文献类型：
--
作者：
Shuangli Liu;Zhu Zheng;Xiaoxuan Sun;Zhiqin Zhao;Yuanjin Zheng;Huabei Jiang;Xiaozhang Zhu;Q. Liu
微波热声层析成像是一种新兴的无创、无电离成像技术。在实际应用中，如乳腺肿瘤检测和脑成像等，被检测组织的声学特性通常是未知的，并且在空间上是不均匀的，这导致了被埋目标的失真和模糊。本文提出了一种基于声速自动对焦的图像重建方法，以降低不同软组织的声不均匀性对图像的影响。根据这种方法，可以通过决策图自动确定组织类型的数量，在本工作中称为聚类。为了区分不同组织的边界，对得到的图像数据进行高斯混合模型（Gaussian Mixture Model， GMM）的拟合进行软聚类，取代传统的硬聚类。通过将具有相应数据密度峰特征的组织中心固定为高斯参数的均值，而不是随机选择，可以保证自适应和鲁棒的重建性能。在执行迭代GMM优化后，实现了SoS自动对焦。采用更新后的SoS分布重构图像的精度高于采用齐次假设重构图像的精度。与现有的同类方法相比，该方法避免了额外的实验费用，并且在介质相对复杂时对硬赋值模型误差具有较好的鲁棒性。结合琼脂幻影和猪脑的实验，给出了真实的乳房模型和脑模型仿真，验证了该方法的有效性。
Microwave induced thermoacoustic tomography is a newly developing non-invasive and non-ionizing modality. In practical applications, such as breast tumor detection and brain imaging, the acoustic properties in the tissue to be detected are usually unknown and spatially non-uniform, which results in distortion and blurring of the buried targets. In this paper, a reconstruction method based on speed of sound (SoS) autofocus is proposed to reduce the effect of acoustic inhomogeneity in different soft tissues. According to this method, the number of tissue types, which are referred to as clusters in this work, can be automatically determined by a decision graph. To distinguish the boundaries of different tissues, a Gaussian Mixture Model (GMM) is fitted to the obtained image data for soft clustering instead of traditional hard clustering. Through fixing the tissue centers which are characterized by corresponding data density peaks as the means of Gaussian parameters rather than choosing them randomly, adaptive and robust reconstruction performance can be guaranteed. After performing an iterative GMM optimization, the SoS autofocus is achieved. Image reconstructed by using the updated SoS distribution is with higher accuracy than that with homogeneous assumption. Compared with the existing similar methods, the proposed method strategy obviates the need of extra experiment costs, and possesses good robustness with respect to hard assignment model errors when the medium is relatively complex. Realistic breast model and brain model simulations combined with experiments of agar phantom and pig's brain are provided to demonstrate the effectiveness of the proposed method.