1.Center for Biomedical Engineering, School of Information Science and Technology, Fudan University, Shanghai 200433, China 2.Institutes of Brain Science, Fudan University, Shanghai 200433, China 3.Department of Materials Science, Fudan University, Shanghai 200433, China 4.Shanghai Key Laboratory of Medical Imaging Computing and Computer Assisted Intervention, Shanghai 200433, China
Fund Project:Project supported by the National Natural Science Foundation of China (Grant Nos. 11974081, 11827808, 51961145108), the Natural Science Foundation of Shanghai, China (Grant No. 19ZR1402700), the Shanghai Rising Star Program, China (Grant No. 20QC1400200), and the Shanghai Academic Research Leader Program, China (Grant No. 19XD1400500)
Received Date:09 November 2020
Accepted Date:22 December 2020
Available Online:24 May 2021
Published Online:05 June 2021
Abstract:With the advantages of non-radiation and low-cost, ultrasound imaging has been widely used in clinical diagnosis. However, due to the compromise between temporal and spatial resolution, the traditional ultrasound imaging method which collects images line by line fails to obtain the images at a high frame rate. Ultrafast ultrasound imaging method based on the plane-wave emission can achieve a high frame rate with the capability for instantaneous hemodynamic imaging of neurovascular response. Besides, by the coherent compounding of the echo signals received from emitting a set of tilting plane-waves, the image quality can be improved in terms of resolution, contrast and signal-to-noise ratio (SNR). Considering the fact that received signals are the mixture of echo signals from the low-speed soft tissue and high-speed blood flow, a clutter filtering method is used to remove the signals of soft tissue. In this study, the ultrafast ultrasound sequence of emission and reception is designed to image the spinal cord by using the groups of multiple steering-angle plane-waves. The so-called f-k migration algorithm based on an exploding reflector model (ERM) is used for coherent plane-wave compounding. Eigenvalue decomposition (EVD) is then applied to tissue and blood signal extraction. The static tissue signals correspond to eigenvectors with low Doppler frequency shift and large magnitude eigenvalue while the high-speed blood flow signals correspond to eigenvectors with high Doppler frequency shift and small eigenvalue. Therefore, frequency and amplitude thresholds can be applied to the accurate separation of the blood component and the tissue component. After the signal extraction, the hemodynamic imaging of blood vessels can thus be obtained from the power Doppler results. The experiments are carried out by using a programmable ultrasonic array system and a high-frequency linear array transducer L22-14vX with a central frequency of 15.625 MHz. The sample acquisition frequency is set to be 62.5 MHz. In rat experiments in vivo, 14040 angle-stilting images per second are compounded into 520 images, each of which is compounded from 27 tilting images (the tilting angles ranging from –10° to 10°). The experiments are conducted on the anesthetized rats with part of the vertebral plate removed. The ultrafast B-mode images are acquired from a 14.0 mm ×12.7 mm region-of-interest. The 520 frames acquired in one second are processed by the clutter filter based on eigenvalue decomposition and Doppler shift analysis. The eigenvectors and eigenvalues corresponding to the soft tissue are discarded. The power Doppler images of the spinal cord micro-vessels are obtained. The experimental results indicate that the ultrafast ultrasound Doppler imaging method is effective for monitoring the hemodynamic variation in spinal cord. The micro-hemorrhage can be identified from the power Doppler images. The quantitative results indicate that the SNR increases with the steering angles increasing. Compared with the results obtained by using 9 steering angles, 5 dB SNR enhancement can be obtained by using the 27 steering angles. In conclusion, the ultrafast ultrasound Doppler technology has the potential applications in spinal cord microvascular imaging and hemodynamic evaluation of neurovascular function of spinal cord. Keywords:ultrafast ultrasound Doppler/ clutter filtering/ eigenvalue decomposition/ spinal cord micro-vessels
3.实验设置多通道超声数据发射与采集在可编程的Vantage 256相控超声实验平台(Verasonics Inc, WA, USA)上进行. 探头型号为L22-14vX, 包含128个阵元, 中心频率为15.625 MHz, 阵元间距为0.1 mm. 信号采样频率为62.5 MHz, 数据经由总线传输至计算机. 仿体采用长6 cm、宽5 cm、高5 cm的琼脂块, 在6—8 mm深处制备直径为2 mm的管路以模拟血管. 用注射泵以0.1 mL/s的速度将混有散射子的悬浊液注入中管道中模拟血流. 仿体测量中采用每秒3500次的超快超声脉冲发射序列, 该序列由500个子序列构成, 各子序列由–10°—10°之间等间隔分布的7个倾斜平面波组成, 复合成像帧频为每秒500帧. 成像区域为15.0 mm × 12.7 mm的矩形成像区域. 在体实验对象为Sprague-Dawley大鼠(雄性, 7周龄, 体重约350 g). 通过手术将T13-L2段椎弓板和棘突切除, 克服超声经过椎骨所致能量衰减. 用异氟烷麻醉大鼠(5%用于诱导麻醉, 2.5%—3.0%用于术中维持麻醉). 随后, 制作刺入型脊髓损伤模型. 实验结束时, 大鼠脱位处死. 实验中采用每秒14040次的超快超声平面波发射序列, 该序列由520个子序列构成, 每个子序列发射–10°—10°之间等间隔分布的27个倾斜角度平面波, 因此, 复合成像帧频为每秒520帧. 实际成像区域为14.0 mm × 12.7 mm的矩形成像区域(实验设定的成像深度为5.9 mm, 实际成像深度是设定成像区域的对角线长度). 实验装置示意如图4所示. 图 4 大鼠脊髓血流超声成像实验装置示意 Figure4. Schematic diagram of experimental ultrasonic imaging set-up for blood flow of rat spinal cord.
4.实验结果24.1.仿体实验结果 -->
4.1.仿体实验结果
对1 s内获得的回波数据进行波束合成与多角度平面波相干叠加, 共得到500帧复合B模式图像. 为提取仿体血流信号, 对500帧图像数据进行EVD处理和特征向量的多普勒频移分析. 特征值阈值选取会影响超快超声功率多谱勒成像结果的对比度和SNR, 相关选取标准仍有赖于经验性参数[28]. 通常, 动态血流信号对应多普勒频移高且特征值小的信号成分, 静态组织信号对应多普勒频移低且特征值大的信号成分. 图5(a)为归一化多普勒频移对应的特征向量个数的直方图, 纵坐标为特征向量个数. 由图可得, 低频分量集中分布在归一化多普勒频移小于0.1的区间, 因此将截止频率设为0.1以滤除低频组织信号. 在体实验中也采用了该阈值参数. 图5(b)为数据集的特征向量与特征值的对应关系, 图5(c)为特征向量与归一化多普勒频移的对应关系, 图5(d)为特征值与归一化多普勒频移的对应关系, 图中虚线为归一化多普勒频移为0.1的阈值线, 实线为幅值为–80 dB的阈值线. 将多普勒频移小于0.1且特征值大于–80 dB的特征向量滤除, 重建后的成像结果(第400帧)如图6所示. 图6(a)—(d)分别是滤波前的仿体成像结果、滤波后的仿体血流图、滤波后的仿体软组织图和仿体血流的功率多普勒成像结果. 如图6(d)所示, 通过上述杂波滤除方法可以得到6—8 mm深度处仿体血流的功率多普勒成像结果. 图 5 特征值分解与多普勒频移分析结果 (a) 归一化多普勒频移对应特征向量个数的直方图; (b)特征向量的特征值; (c) 特征向量的归一化多普勒频移; (d) 特征值对应的归一化多普勒频移 Figure5. Eigenvalue decomposition and Doppler shift analysis results: (a) Histogram of the number of eigenvectors corresponding to normalized Doppler shifts; (b) eigenvalues of eigenvectors; (c) normalized Doppler shifts of eigenvectors; (d) eigenvalues versus normalized Doppler shifts.
图 6 仿体血流成像结果(第400帧) (a) 滤波前的成像结果; (b) 滤波后的血流成像结果; (c) 滤波后的软组织成像结果; (d) 功率多普勒成像结果 Figure6. Imaging results of the 400th frame of the phantom blood flow: (a) Original image before clutter filtering; (b) blood flow image after clutter filtering; (c) soft tissue image after clutter filtering; (d) power Doppler imaging result.
24.2.大鼠在体实验结果 -->
4.2.大鼠在体实验结果
对1 s内得到的回波数据进行波束合成与多角度平面波相干叠加, 共得到520帧复合B模式图像. 对520帧图像数据进行EVD处理和特征向量的多普勒频移分析, 结果如图7所示. 图7(a)为归一化多普勒频移对应特征向量个数的直方图, 纵坐标为特征向量的个数, 相较于仿体血流, 实际血流有不同的流速, 因此其多普勒频移的分布区间更宽. 若频率阈值选取过高, 会损失低速血流信号成分, 导致有效信息缺失; 本实验采用与仿体实验一致的经验频率阈值, 即取归一化多普勒频移为0.1. 图7(b)为数据集的特征向量与特征值的对应关系, 图7(c)为特征向量与归一化多普勒频移的对应关系, 图7(d)为特征值与归一化多普勒频移的对应关系, 图中虚线为归一化多普勒频移为0.1的阈值线, 实线为幅值为–130 dB的阈值线. 将多普勒频移小于0.1且特征值幅值大于–130 dB的特征向量滤除. 图8为多角度平面复合和滤波前后的成像结果. 图8(a), (b)分别为发射单角度倾斜平面波得到的波束合成后的结果和发射27个倾斜角度平面波相干复合成像的结果, 图8(c)为滤波后的多帧脊髓血流图像, 从中可见微血流的变化情况. 图 7 特征值分解与多普勒频移分析结果 (a) 归一化多普勒频移对应特征向量个数的直方图; (b)特征向量的特征值; (c) 特征向量的归一化多普勒频移; (d) 特征值对应的归一化多普勒频移 Figure7. Eigenvalue decomposition and Doppler shift analysis results: (a) Histogram of the number of eigenvectors corresponding to normalized Doppler shifts; (b) eigenvalues of eigenvectors; (c) normalized Doppler shifts of eigenvectors; (d) eigenvalues versus normalized Doppler shifts.
图 8 基于多角度平面波复合成像的大鼠脊髓血流成像结果 (a) 单角度平面波发射成像; (b) 多角度平面波复合成像; (c) 杂波滤除结果. 每一次发射超声平面波的时间间隔为71.225 μs, 27次发射倾斜平面波与接收反射回波的总时长为1.923 ms, 对多角度信号经相干复合可获得单帧超声图像, 其所对应的成像帧率为每秒520帧 Figure8. Blood flow imaging results of rat spinal cord based on multi-angle compounding method: (a) Beamforming results after a single emission; (b) multi-angle compounding images; (c) images after clutter filtering. The time interval between each emission is 71.225 μs. Each compounded frame is obtained using 27 steering-angle plane-waves within a period of 1.923 ms. Consequently the frame rate is 520 frames per second.
为讨论复合平面波成像中倾斜平面波角度数在超快超声多普勒血流成像中的作用, 分别设定每个子序列中发射倾斜平面波的个数为N = 3 [–1°—1°], N = 9 [–3°—3°], N = 17 [–7°—7°], N = 27 [–10°—10°](倾斜平面波的角度在设定区间内均匀分布), 复合成像帧频均为520帧/s, 成像结果如图9所示. 图9(a)—9(d)的1—3图分别为上述4种情况滤波之前、滤波之后和功率多普勒成像结果. 当N = 3 [–1°—1°]时, 成像质量差, 无法辨别脊髓内部微血管分布(如图9(a)). 当N = 9 [–3°—3°]时, 多普勒成像方法能够对脊髓内部微血管结构进行成像, 但难以呈现完整的微血管结构(如图9(b)). 随着成像角度数的增大, 当成像角度N = 17 [–7°—7°]和N = 27[–10°—10°]时, 在多普勒成像结果中可以看到较为清晰的微血管结构(如图9(c),(d)), 其中微血管分布主要由靠近背侧和靠近腹腔的两部分血供构成, 该成像结果与组织病理学中脊髓血管的解剖结构形态相似[29]. 随着角度数的增大, 图像的对比度和SNR得到改善. 并能观察到水平方向–2.5 mm, 竖直方向1.5—2.0 mm处由细针刺入脊髓所致的局部微血流缺失现象. 图 9 不同角度复合平面波成像结果对比图(复合帧频均为每秒520帧) (a) 3个角度[–1°—1°]倾斜平面波复合成像结果; (b) 9个角度[–3°—3°]倾斜平面波复合成像结果; (c) 17个角度[–7°—7°]倾斜平面波复合成像结果; (d) 27个角度[–10°—10°]倾斜平面波复合成像结果. 1为单帧原始B模式图像, 2为杂波滤除之后的成像结果, 其中可见微血流变化, 3为1 s内采集数据得到的功率多普勒血流图(1, 2色标单位为dB, 3为归一化数值的多普勒成像结果) Figure9. Comparison of compounded images with different numbers of steering angles (composite frame rate is 520 frames per second): (a) Images compounded of data from emitting 3 [–1°—1°] steering plane-waves; (b) images compounded of data from emitting 9 [–3°—3°] steering plane-waves; (c) images compounded of data from emitting 17 [–7°—7°] steering plane-waves; (d) images compounded of data from emitting 27 [–10°—10°] steering plane-waves. Images labeled 1 are original B-mode images; images labeled 2 are imaging results after clutter filtering in which changes of blood flow can be observed; images labeled 3 are power Doppler images of micro-vessels (data was obtained within 1 s).
5.讨 论在图9的功率多普勒血流图中选定一个感兴趣区域(如图9(b)—(d)编号3的图中矩形虚线框所示), 进而分析结果的对比分辨率. 如图10所示, 分别将角度数N = 9 [–3°—3°], N = 17 [–7°—7°], N = 27 [–10°—10°]的成像结果中该感兴趣区域放大. 图10(b)给出了图10(a)中虚线深度处的幅度曲线, 进而比较微血管成像分辨率. 由图可得, 当N = 27时, 可由箭头所指处的幅度峰值分辨两相邻微血管; 而当N = 9时, 因图像分辨率降低, 故无法分辨此处的两相邻微血管. 图 10 对比分辨率随复合平面波角度数增加的变化情况 (a) 当角度数N = 9, 17, 27时, 图9(b)-(d)编号3的图中矩形虚线框中图块的放大结果; (b) 当角度数N = 9, 17, 27时, 图10(a)中虚线深度处的幅值曲线图 Figure10. Change of contrast resolution with the increase of the number of steering angles: (a) Enlarged image blocks in the rectangular dashed box in No.3 figure in Fig.9 (b)-(d) (angle numbers are 9, 17, 27, respectively); (b) amplitude curve of the dotted line position in Fig.10(a) (angle numbers are 9, 17, 27, respectively).
由图10可得, 随复合平面波角度数增大, 血流信号相较于背景噪声得到增强, 以下对图像的SNR加以量化分析. 当N = 27时最大的角度范围为–10°—10°, N = 9时的角度范围约为–3°—3°. 从图10(a)中选择3个尺寸为0.2 mm × 0.2 mm的感兴趣区域(编号为1, 2, 3), 白色方框为血流信号区域, 最临近的灰色方框为参考背景区域. SNR的计算公式如下[30]: