基于低秩理论的超声血流壁滤波器

发布时间：2018-01-31 03:17

  本文关键词： 超声血流成像 壁滤波 低秩模型 FIR滤波器　出处：《中国生物医学工程学报》2016年04期 　论文类型：期刊论文

【摘要】：传统的超声血流壁滤波器采用单一的截止频率,对于心跳、呼吸等因素所产生的不同组织运动滤波效果不佳。提出一种基于低秩理论的超声血流壁滤波方法。通过分析超声血流的特点,将其公式化并建立低秩模型,模型包括一个矩阵秩的最小化和一个稀疏矩阵问题。松弛后转化为最小化核范数与1范数的线性组合,成为凸优化可解问题,从而实现低秩滤波。这种滤波的创新性在于它是自适应的,同时考虑了信号的低秩性与稀疏性两个特点,通过线性组合使得滤波效果达到最优。通过低秩滤波,超声血流仿真数据得到更加精确的血流信号。对比传统的FIR滤波,分别采用12阶、56阶和92阶的FIR滤波器,得到血流信号对比实际仿真数据的RMS误差分别约为34%、16%和12%,而低秩滤波的RMS误差则低于0.001%。低秩滤波相比传统FIR滤波,不仅提高了精度,而且滤波后的信号长度不会损失。但是低秩模型计算过于复杂,因此目前还很难应用在实时的超声系统中。
[Abstract]:The traditional ultrasonic blood flow filter uses a single cutoff frequency for the heartbeat. A method of ultrasonic blood flow wall filtering based on low rank theory is proposed. By analyzing the characteristics of ultrasonic blood flow, the paper formulates it and establishes a low-rank model. The model consists of a matrix rank minimization and a sparse matrix problem, which is transformed into a linear combination of minimized kernel norm and 1-norm, and becomes a solvable problem for convex optimization. The innovation of this filter is that it is adaptive, considering the characteristics of low rank and sparsity of signal, the filtering effect is optimized by linear combination, and by low rank filter. Compared with the traditional FIR filter, the ultrasonic blood flow simulation data obtained more accurate blood flow signal. Compared with the traditional FIR filter, the FIR filter of 12 order 56 and 92 order respectively is used. The RMS error of the actual simulation data of the blood flow signal is about 34 ~ 16% and 12% respectively. The RMS error of low rank filter is less than 0. 001. Compared with traditional FIR filter, low rank filter not only improves the accuracy. The length of the filtered signal will not be lost, but the calculation of the low-rank model is too complicated, so it is difficult to be used in the real-time ultrasonic system.
【作者单位】： 深圳大学医学部生物医学工程学院医学超声关键技术国家地方联合工程实验室;深圳迈瑞生物医疗电子股份有限公司;香港中文大学计算机科学与工程系;
【基金】：中国博士后科学基金(2014M562207) 国家自然科学基金(61372006;61427806)
【分类号】：TN713;R445.1
【正文快照】： 3(香港中文大学计算机科学与工程系,香港)引言超声血流成像中最关键的一步是提取血流信号,因为血液的反射系数大约只有组织的1/500~1/10[1-4]。在超声回波中,组织和血流信号混在一起,只有提取并得到信噪比高的血流信号,才能计算出高灵敏度和高精度的血流速度。这两种信号在理，

本文编号：1478038