Super-resolution ultrasound imaging method for microvasculature in vivo with a high temporal accuracy

Abstract

Traditional ultrasound imaging techniques are limited in spatial resolution to visualize angiogenic vasa vasorum that is considered as an important marker for atherosclerotic plaque progression and vulnerability. The recently introduced super-resolution imaging technique based on microbubble center localization has shown potential to achieve unprecedented high spatial resolution beyond the acoustic diffraction limit. However, a major drawback of the current super-resolution imaging approach is low temporal resolution because it requires a large number of imaging frames. In this study, a new imaging sequence and signal processing approach for super-resolution ultrasound imaging are presented to improve temporal resolution by employing deconvolution and spatio-temporal-interframe-correlation based data acquisition. In vivo feasibility of the developed technology is demonstrated and evaluated in imaging vasa vasorum in the rabbit atherosclerosis model. The proposed method not only identifies a tiny vessel with a diameter of 41 μm, 5 times higher spatial resolution than the acoustic diffraction limit at 7.7 MHz, but also significantly improves temporal resolution that allows for imaging vessels over cardiac motion.

Conflict of interest statement

The authors declare no competing interests.

Figures

Figure 1

Comparison of several imaging modalities. (a) CadenceTR contrast-enhanced imaging with microbubbles acquired by commercial ultrasound scanner (Sequoia 512, Siemens), (b) conventional B-mode imaging, (b) temporal MIP vascular imaging, (d) proposed super-resolution imaging of ROI. Same raw data is used to reconstruct images (b–d). The white solid rectangle represents balloon-injured area that the plaque is expected to be developed. The white arrow indicates the same vessel branch that shows a correlation of images acquired by two different ultrasound scanners. White dashed rectangle represents selected ROI used in Figs 2 and 3.

Figure 2

The spatial resolution of the proposed imaging method. (a) The detectable smallest vessel is chosen (white solid line) in the ROI of Fig. 1(b–d) indicated by the white dashed rectangle, (b) Spatial profile of the selected vessel. FWHM is estimated by 41 μm (<λ/5).

Figure 3

Repeatability of the proposed method. Vessel images were chosen in the ROI of Fig. 1(b–d) indicated by the white dashed rectangle (a) temporal MIP vascular network imaging using eigen-decomposition method. (b–d) Super-resolution images using sequentially acquired three datasets (2,000 frames per each image) from the same region of interest.

Figure 4

B-mode image (a,d) and corresponding super-resolution perfusion image overlaid on the B-mode image at diastole (b,e) and systole state (c,f). Top panel images are acquired from the injured side and bottom panel images are acquired from the non-injured side. Significant plaques are shown in the B-mode image on the injured side. The yellow dotted line represents adventitia region and white arrows indicate vasa vasorum in the adventitia. (Supplementary Movies 1 and 2 are available).

Figure 5

Haematoxylin and eosin stained vessel on the injured side (a) and non-injured side (b). Thirty images acquired at ×40 magnification are stitched to reconstruct an overall image of the vessel for (a) and (b). Significant plaque development is found in the injured side (a). Vasa vasorum on adventitia in the selected region was stained by anti-von Willebrand factor. A large number of vasa vasorums are found in adventitia on the injured side (c), but a few vasa vasorums are found in adventitia on the non-injured side (d).

Figure 6

Block diagram for signal processing of super-resolution ultrasound imaging. BF: Delay-and-sum beamformer; QD: Quadrature demodulator; CF: Eigen-based spatio-temporal clutter filter; ED: Envelope detector; DV: Deconvolution with the system PSF; ∑: Integrator with STIC data alignment based on estimated cardiac pulsation.

Figure 7

Conceptual demonstration of sub-wavelength localization using deconvolution on synthetic data. (a) PSF of the imaging system. FWHM is assumed as 150 µm. (b) The locations of the two neighboring targets (Ground truth). Two targets are positioned 70 µm apart. (c) The synthetic signal received from two targets is shown in (b) using the imaging system with PSF shown in (a). This signal is modeled as a received signal in the imaging system whose has PSF shown in (a). Two targets cannot be separated in the image due to their distance is shorter than the spatial resolution of the imaging system. (d) Deconvolution results of received signal shown in (c) using the system PSF shown in (a). Two targets are distinctly identified. (e) The synthetic signal received from two targets when noise is added is shown in (b) using the same imaging system. (f) Deconvolution results of (e), where two targets are clearly identified with minimal interference due to noise.

Figure 8

Estimated cardiac pulsation by counting the numbers of flowing microbubbles (blue solid line). After applying low-pass-filtering, the frames with minimum value are chosen as reference frames for synchronization (red solid line).

Figure 9

Graphical diagram of STIC data acquisition. Sequentially acquired multiple datasets are synchronized to form a single cardiac cycle event based on the cardiac pulsation estimated by the numbers of bubbles.

Figure 10

Experiment setup of rabbit imaging. Microbubbles were injected via ear vein access. A hockey stick linear array ultrasound transducer was used for imaging. Transducer holder is used to removing operator-dependent motion in this study.