1.College of Electronics and Information Engineering, Tongji University, Shanghai 201804, China 2.Faculty of Health, University of Witten/Herdecke, Witten D-58448, Germany
Fund Project:Project supported by the National Natural Science Foundation of China (Grant No. 60771030), the National High Technology Research and Development Program of China (Grant No. 2008AA02Z308), the Shanghai Foundation for Development of Science and Technology, China (Grant No. 08JC1421800), the Open Project of State Key Laboratory of Function Materials for Information, Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, China (Grant No. SKL2013010), and the Open Project of Key Laboratory of Medical Image Computing and Computer Assisted Intervention of Shanghai, Shanghai Medical College of Fudan University, China (Grant No. 13DZ2272200-2)
Received Date:02 January 2019
Accepted Date:25 April 2019
Available Online:01 July 2019
Published Online:05 July 2019
Abstract: The current source reconstruction and magnetic imaging is a new technique to non-invasively obtain spatial information regarding cardiac electrical activity using magnetocardiogram (MCG) signals measured by the superconducting quantum interference device (SQUID) on the human thorax surface. Using MCG signals to reconstruct distributed current sources needs to solve the inverse problem of magnetic field. The beamforming is a type of spatial filter method that has been used for distributed source reconstruction and source imaging in electroencephalogram (EEG) and magnetoencephalogram (MEG). In this paper, the dipole moment of distributed current source is estimated with corresponding each spatial filter based on the cardiac source field model. The purpose is to enhance the intensity contrast of the dipole moment of distributed current sources in distributed source spatial spectrum estimation with beamforming, so that the reconstructed-pseudo sources beyond the heart can be removed for imaging cardiac electric activity well. A new beamforming method of improving intensity contrast (IIC) of distributed source spatial spectrum estimation is developed for imaging cardiac electric activity in P-wave, due to cardiac magnetic signals in P-wave lower than that of the peak value of R-wave, which has a relatively low signal-to-noise ratio (SNR). For enhancing the accuracy of current source reconstruction in P-wave, the IIC divided into two steps: firstly, to introduce the lead-field matrix, which represents the measurement sensor-array sensitivity to magnetic field current sources, into a weight matrix of the spatial filter for making the output estimation of the filter more sensitive to the current sourcedistribution, so as to improve the intensity contrast of the reconstructed distributed sources.Secondly, by setting a threshold of source intensity from experience, to extract the reconstructed source with locally-maximal dipole strength at each time for eliminating the relatively weak pseudo sources in other locations, so as to enhance the accuracy of current source reconstruction during P-wave. In this paper, the IIC and three other methods, including minimum variance beamforming (MVB), suppressing spatial filter output noise-power gain (SONG) and trust region reflective (TRR), are compared by using the theoretical analysis and simulation experiments of MCG current source reconstruction during P-wave. The results show that the IIC has higher intensity contrast of the single source spatial spectrum estimation, and possesses better accuracy of the current source reconstruction. The 61-channel MCG signals of two healthy subjects and their imaging of cardiac electrical activity during P-wave also are analyzed. The result shows that the IIC is better than the other three methods. It is indicated that two healthy subjects have stronger electrical activity in the atrium than that in the ventricle at Ppeak time, also that the electrical activity has the direction feature when the right-atrium is depolarized during P-wave. In summary, the IIC is useful for imaging the cardiac electrical activity. However, it is needed to carry out a further research on patients with local myocardial ischemia and left or right coronary artery stenosis, and to establish the evaluation index for imaging of cardiac electrical activity in such patients. Keywords:cardiac electrical activity/ magnetic imaging/ inverse problem/ spatial filter/ current source reconstruction
全文HTML
--> --> --> -->
2.1.磁场分布源的模型及其空间谱估计
假设SQUID系统在人体胸腔表面测量到的心脏磁场由n个呈网格分布的电流源产生. 测量面上第k通道的心磁测量信号用标量${b_k}\left( t \right)$表示, 则t时刻c通道阵列信号的列向量可用 ${{b}}\left( t \right) = $$ {\left[ {{b_1}\left( t \right),{b_2}\left( t \right),...,{b_c}\left( t \right)} \right]^{\rm{T}}} $表示. 磁场与分布源的关系模型可用如下线性方程表示[13,16]:
${{b}}\left( t \right) = \sum\limits_{j = 1}^n {\left[ {{{{L}}_j}\left( t \right){{q}}\left( {t,{{{r}}_j}} \right)} \right]} + {{v}}\left( t \right),$
本文在改进空间谱估计对比度的基础上, 采用了设置源强度阈值, 提取每个时刻重建分布源中局部强度极大电流源的方法. 这是因为在重建磁场分布源时, 需要消除环境噪声和计算误差引起的大量分布的重建弱源与重建的伪源(可能超出心脏范围). 本文假设源空间中每个分布源的最小邻域中包括26个其他分布源, 图1中它们的位置分别用红色和蓝色表示. 并用每个时刻空间谱估计的最大强度${\widehat \varsigma _{\max }} =$$ \mathop {\max }\limits_{j = 1,2,...,n} \sqrt {{\mathop{\rm tr}\nolimits} \left[ {{\mathop{\rm E}\nolimits} \left( {{{\widehat {{q}}}_j}{{\widehat {{q}}}_j}^{\rm{T}}} \right)} \right]} $作为提取局部极大源的阈值$\gamma = {\widehat \varsigma _{\max }} \times \beta $的基数, 其中比例系数$\beta $可根据经验确定. 这样, 改进源强度对比度后, 可以通过设定阈值, 消除部分重建的伪源与大量较弱的重建分布源. 最后, 利用那些局部强度极大的电流源来研究心脏的电活动. 图 1 分布源位置及指定的26个邻域位置, 分别用红色和蓝色表示 Figure1. The position of the source and 26 specified adjacent positions are shown in red and blue colors, respectively
-->
3.1.仿真数据
假设包含心脏的人体躯干是沿坐标系Z轴水平分层的导体(horizontally layered conductor, HLC), 心脏–躯干模型如图2(a)所示. 层间电导率之差与该处静电势的乘积被称作二次电流密度, 其方向与沿Z方向测量的心脏磁场平行[17]. 根据毕奥萨法尔定律, 导体中二次电流密度产生的磁场可以忽略[10]. 对图2(a)中包含心脏在内的导体 图 2 (a) 水平分层导体. 其中黑色直线表示分层躯干的边界, 黄色椭圆体表示心脏; (b), (c) 测量通道1和61的磁场导联$l_{j,k}^\zeta , \left( {\zeta = X, Y, Z} \right)$曲线. 其中红、绿和蓝色分别表示X, Y和Z方向导联的曲线 Figure2. (a) The horizontally layered conductor (HLC), where the black line indicates the boundary of HLC, and the yellow spheroid the heart; (b),(c) The X, Y and Z lead-field based plots of channels 1 and 61 are expressed in red, green and blue, respectively
$\begin{split}{G^0} = & \left\{\left( {x, y, z} \right)\left| x \in \left[ { - {\rm{10}}{\rm{.5,15}}{\rm{.5}}} \right]{\rm{ cm, }}\right.\right.\\ & y \in \left[ { - {\rm{15}}{\rm{.5,10}}{\rm{.5}}} \right]{\rm{ cm}}, z \in \left[ {{\rm{2}}{\rm{.5,20}}{\rm{.5}}} \right]{\rm cm}\}\end{split}$
假设参考路径1是一条起止点坐标为Ps = (7, 5, 6)cm和Pe = (–3, –5, 16)cm的空间直线. 如图3(a)—图3(e) 中黑色圆圈所示, 该路径上相邻分布源的间距为$\sqrt 3 \;{\rm{ cm}}$. 单电流源通过参考路径上相邻位置的时间$\Delta t = 0.01\;{\rm{ s}}$, 速度${\rm{MV}} = 0.01\sqrt 3 /$$0.01 = \sqrt 3 \approx 1.73\;{\rm{ m/s}}$. 图3(b)—图3(e) 给出了SNR = 20 dB时, 用上述4种方法得到的重建电流源和电活动成像结果. 图中可见, 用IIC每个时刻重建的电流源沿着参考路径1上的黑圈移动. 黑圈中的绿色由浅到深表示时间变化. 在该参考路径以外, IIC出现的伪源最少, SONG与TRR的结果次之. 图 3 (a) 黑色圆圈○和箭头表示沿直线方向的参考路径1; (b)?(e) 是SNR = 20 dB时, IIC, SONG, MVB和TRR方法的源重建结果. 黑圈中的浅绿色到深绿色表示重建电流源位置随时间的变化 Figure3. (a) The black circles ○ and arrow indicate the reference path 1 along the straight line; (b)?(e) The results of the current source reconstruction using IIC, SONG, MVB and TRR methods when SNR is 20 dB. The green points at different color levels denote the reconstructed source locations at different times
表1参考路径1对应的单源估计误差 Table1.The error of single source estimation correspond to the reference path 1
23.3.仿真结果2 -->
3.3.仿真结果2
假设参考路径2包括两条起止点坐标分别为[Ps1,Pe1] = [(7, 5, 6), (7, –5, 6)]cm和[Ps2,Pe2] = [(–3, 5, 16), (–3, –5, 16)]cm的空间直线. 如图4(a)—图4(e)中黑圈所示, 这两条路径上相邻分布源的间距均为1 cm. 仿真中, 令两个电流源分别通过各自参考路径上相邻位置的时间$\Delta t = 0.01\;{\rm{ s}}$, 速度${\rm{MV}} = {{0.01} / {0.01}} = 1\;{\rm{ m/s}}$. 其中一个电流源比另一个电流源提前0.005 s开始移动. 图4给出了SNR = 20 dB时, 4种方法的仿真结果和电活动成像. 图中可见, 用IIC每个时刻重建的电流源沿参考路径2中的黑圈移动. 黑圈中的绿色由浅到深表示时间变化. 在该参考路径以外, IIC出现的伪源最少, 明显优于其他3种方法. 图 4 (a) 黑色圆圈○和箭头表示沿直线方向的参考路径2; (b)?(e) SNR = 20 dB时, IIC,SONG,MVB和TRR方法的源重建结果. 黑圈中的浅绿色到深绿色表示重建电流源位置随时间的变化 Figure4. (a) The black circles ○ and arrow indicate the reference path 2 along the straight line; (b)?(e) The results of the current source reconstruction using IIC, SONG, MVB and TRR methods when SNR is 20 dB. The green points at different color levels denote the reconstructed source locations at different times