基于压缩感知的超声成像方法研究

To increase the frame rate (temporal resolution) of medical ultrasound imaging has important significance for the implementation of three-dimensional imaging, novel elasticity imaging and blood flow imaging. Most of the current methods increase the frame rate but sacrificing the spatial resolution and contrast resolution. This project proposes to use the compressed sensing theory in information sciences combined with the beamforming technique in acoustics, to increase the frame rate of medical ultrasound imaging and obtain better tradeoffs among different imaging performance indices. First, based on compressed sensing and acoustical theory, the theoretical model of compressed sensing based ultrasound imaging will be established, which is suitable for different ultrasound transducers including linear array, convex array, ring array, phased array and two-dimensional matrix array. Based on the model, the ultrasound data acquisition method and image reconstruction method based on compressed sensing will be developed, and the corresponding parameters will be optimized. Last, the method will be applied to two- and three-dimensional ultrasound imaging to obtain good temporal resolution, spatial resolution and contrast resolution simultaneously. The project will perform systematic theoretical analysis and validation using computer simulations, phantom experiments and in vivo experiments. This project will generate new idea in obtaining high-frame-rate ultrasound imaging, which is promising to be applied to different medical ultrasound imaging modes with high frame rate and imaging quality.

提高医学超声成像的帧频（时间分辨率），对于实现三维成像、新型弹性成像与血流成像、具有重要的意义。目前的大部分方法在提高帧频的同时，都将牺牲空间与对比度分辨率。本项目提出采用信息科学的压缩感知理论，结合声学的波束合成技术，提高医学超声成像的帧频，并获得不同成像性能指标之间的更好平衡。首先，基于压缩感知理论与声学理论，建立压缩感知的超声成像理论模型，适用于线阵、凸阵、环阵、相控阵与二维面阵等探头；在此基础上，建立基于压缩感知的超声数据采集方法与图像重建方法，并进行参数优化；最后，将该方法应用于二维及三维超声成像，同时获得高的时间、空间与对比度分辨率。项目将进行系统的理论分析、计算机仿真、仿体实验与在体实验的验证。本项目的实施，将为高帧频超声成像带来新的思路，可望应用于多种医学超声成像模式，获得更高的帧频和成像质量。

为了提高超声成像的帧频与质量，提出了基于压缩感知的合成发射孔径成像方法（CS-STA），包括成像理论、成像序列、重建算法，缩短了超声成像的采集时间，提高了成像帧频，同时获得了高的时间分辨率、空间分辨率、对比度、信噪比和穿透力。该方法适用于不同的超声探头，包括线阵、凸阵、相控阵与二维面阵探头，具有普遍的意义。对CS-STA的成像序列与重建算法进行了系统的优化，提出基于同相/正交数据处理的重建算法、基于极小2范数最小二乘求解的重建算法与基于自监督深度学习的重建算法，实现了合成发射孔径数据的高质量、高速度重建。另外，提出了自监督深度学习的超声平面波图像重建算法，同时获得了超高帧频的实时、高质量超声成像。项目提出的理论和方法还可望应用于雷达、声纳、超声无损探伤等与医学超声密切相关的领域。项目取得丰富的研究成果。合计发表论文61篇，其中SCI论文30篇，包括医学成像领域的顶级期刊《IEEE Trans Medical Imaging》2篇，医学图像处理领域的顶级期刊《Medical Image Analysis》2篇，医学超声领域的主流期刊《IEEE Trans Ultrasonics, Ferroelectrics, and Frequency Control》8篇，生物医学工程领域的主流期刊《Physics in Medicine and Biology》3篇。获得美国发明专利授权1项、欧盟发明专利授权1项、中国发明专利授权3项，申请中国发明专利6项。研究生2人获得电气电子工程师学会（IEEE）国际超声年会（IUS）的Student Travel Support Award，1人获得IEEE IUS最佳学生论文提名。