Superharmonic Ultrasound for Motion-Independent Localization Microscopy: Applications to Microvascular Imaging From Low to High Flow Rates

Abstract

Recent advances in high frame rate biomedical ultrasound have led to the development of ultrasound localization microscopy (ULM), a method of imaging microbubble (MB) contrast agents beyond the diffraction limit of conventional coherent imaging techniques. By localizing and tracking the positions of thousands of individual MBs, ultrahigh resolution vascular maps are generated which can be further analyzed to study disease. Isolating bubble echoes from tissue signal is a key requirement for super-resolution imaging which relies on the spatiotemporal separability and localization of the bubble signals. To date, this has been accomplished either during acquisition using contrast imaging sequences or post-beamforming by applying a spatiotemporal filter to the B-mode images. Superharmonic imaging (SHI) is another contrast imaging method that separates bubbles from tissue based on their strongly nonlinear acoustic properties. This approach is highly sensitive, and, unlike spatiotemporal filters, it does not require decorrelation of contrast agent signals. Since this superharmonic method does not rely on bubble velocity, it can detect completely stationary and moving bubbles alike. In this work, we apply SHI to ULM and demonstrate an average improvement in SNR of 10.3-dB in vitro when compared with the standard singular value decomposition filter approach and an increase in SNR at low flow ( [Formula: see text]/frame) from 5 to 16.5 dB. Additionally, we apply this method to imaging a rodent kidney in vivo and measure vessels as small as [Formula: see text] in diameter after motion correction.

Conflict of interest statement

CONFLICT OF INTEREST

P.A.D. is a co-inventor on a patent describing dual-frequency imaging and is a co-founder of SonoVol, Inc., a company which has licensed this patent.

Figures

Fig. 1.

(A) Schematic of the elevation cross-section of the dual-frequency assembly with LF transducers in red and HF array in green. W = 2.9 mm, α = 27°, (yc, zc) = (8.45 mm, 0.73 mm). (B) Photograph of the dual-frequency probe used in experiments, illustrating the LF transducers running parallel to the 21-MHz array front face. (C) Hydrophone measurement of LF beam pattern in the elevational-axial plane. The −6 dB contour of the beam is marked with a dashed line. The axial dimension is measured relative to the face of the HF array. Panels (A) and (B) reproduced from Cherin et al with permission.

Fig. 2.

An overview of data collection and processing for super harmonic ULM. (A) The imaging sequence used for this study. Chunks of 100 dual-frequency frames collected at a frame rate of 500 Hz are separated by B-mode frames for motion tracking. Radio-frequency data is saved after 1,000 DF frames. (B) Speckle tracking is performed between a manually selected reference frame and each B-mode frame to estimate the non-rigid deformation of the kidney during imaging. (C) Dual-frequency images are processed using a threshold and peak detector to localize microbubbles. These positions are then corrected according to the displacements estimated from speckle tracking or thrown out if the parent B-mode patch is not well-correlated with the reference patch.

Fig. 3.

A comparison of SHI-ULM and AA using a pair of 46 μm tubes in a water bath. Yellow scale bars are 500 μm. (A) Super-resolution image generated from 25,000 frames. (B) Maximum intensity projection of the SHI frames used to generate the image in panel A. (C) Average profiles within the regions of interest from panels A and B. The full-width half-maximum values of the AA and ULM profiles are 113 and 44 μm, respectively.

Fig. 4.

Velocity maps of crossed 46 μm tubes in a water bath. White scale bars are 500 μm. (A) Direction map, flow direction indicated by color wheel. (B) Map of the average speed for each pixel.

Fig. 5.

Maximum intensity projections for SVD-filtered and super harmonic images of a 200 μm tube in different flow regimes. All images are displayed on a 25 dB dynamic range for comparison. MIPs of super harmonic images collected at (A) 0.25 μL/min and (B) 15.0 μL/min. MIPs of SVD-filtered images collected at (C) 0.25 μL/min and (D) 15.0 μL/min.

Fig. 6.

SNR vs. flow rate for dual-frequency and SVD-filtered images (blue and red lines, respectively). DF imaging results in an average improvement of 10.3 dB over SVD imaging.

Fig. 7.

Example of super harmonic ultrasound localization microscopy applied to a rodent kidney with motion correction. (A) B-mode scan of the kidney used as a reference for motion correction. (B) Maximum intensity projection of super harmonic images used to generate the ULM image (frames with motion discarded). (C) ULM image generated from 25,000 frames with motion correction applied.

Fig. 8.

Selected vessels from rodent kidney 3D dataset. (A–C) Example vessels cropped from ULM images. (D) Average profiles of the vessels in panels A–C with full-width at half-maximum values of 20.9, 17.2, and 29.1 microns, respectively.

Fig. 9.

A comparison of ULM with and without motion correction based on sparsely interleaved B-mode frames. (A) Rodent kidney vessels are smeared out by respiratory and cardiac motion artifacts. (B) Fine detail of the vessel structure is recovered by a combination of removal of decorrelated frames and using speckle tracking to estimate nonrigid displacements.

Fig. 10.

Tracking bubbles in vivo allows for the mapping of blood velocity in a rodent kidney. (A) The average direction of microbubbles for the ULM image in Fig. 7. (B) The magnitude of the velocity for the same dataset.

Fig. 11.

An example of 3D ultrasound localization microscopy with super harmonic imaging by mechanically scanning the transducer in the elevational dimension. This image was generated with 17 slices spaced at 500 microns and contains vessels on the order of 20 microns.