Hybrid photoacoustic and fast super-resolution ultrasound imaging

Abstract

The combination of photoacoustic (PA) imaging and ultrasound localization microscopy (ULM) with microbubbles has great potential in various fields such as oncology, neuroscience, nephrology, and immunology. Here we developed an interleaved PA/fast ULM imaging technique that enables super-resolution vascular and physiological imaging in less than 2 seconds per frame in vivo. By using sparsity-constrained (SC) optimization, we accelerated the frame rate of ULM up to 37 times with synthetic data and 28 times with in vivo data. This allows for the development of a 3D dual imaging sequence with a commonly used linear array imaging system, without the need for complicated motion correction. Using the dual imaging scheme, we demonstrated two in vivo scenarios challenging to image with either technique alone: the visualization of a dye-labeled mouse lymph node showing nearby microvasculature, and a mouse kidney microangiography with tissue oxygenation. This technique offers a powerful tool for mapping tissue physiological conditions and tracking the contrast agent biodistribution non-invasively.

Fig. 1. Experimental concept of hybrid PA/fast ULM imaging.

The dual in vivo images are generated and processed with the following protocols: a Animals are injected with microbubble solution before (or during) imaging. b During the DAQ, multi-wavelength photoacoustic (PA) imaging and plane-wave fast ultrasound imaging are recorded alternately at each position. The minimum DAQ time for fast ultrasound localization microscopy (ULM) imaging is determined by the characterization time of an averaged pixel saturation curve. c Ultrasound (US) images and PA images of multiple wavelengths are stored and processed separately with procedures shown in (d) and (e). d Sparsity-constrained (SC)-ULM imaging procedure includes three main steps: clutter filtering, SC recovery over frames, and location accumulation. e Blood oxygen saturation (sO₂) PA images are generated through linear spectral unmixing from multi-wavelength PA images.

Fig. 2. Microbubble localization efficiency of SC-ULM and PSF-CC ULM with synthetic imaging data.

a Left to right: a simulated ultrasound image of 100 microbubbles randomly located in an area of 6.25 mm² (an average density of 16 microbubbles mm⁻²) with their ground-truth locations (marked as red dots), the recovered microbubble locations (white dots) using the PSF-CC method, and the recovered microbubble locations (white dots) using the SC method. The SC method recovers 92% of microbubbles, while PSF-CC recovers 35%. b Left to right columns represent the zoom-in images of three selected areas with different microbubble densities. The first row is the simulated microbubble ultrasound images. The second and the third row compare the recovered microbubble locations using the two methods. The white dots represent recovered microbubble locations, and the red dots represent the ground-truth locations. c The average microbubble recovery percentage as a function of microbubble densities using the SC method (blue) and the PSF-CC method (black). The error bar is the standard deviation (N = 20). d The average localization error (left y-axis) and the standard deviation (N = 20, right y-axis) as a function of microbubble densities using the two methods.

Fig. 3. The SC method enables in vivo fast super-resolution vascular imaging.

Super-resolution ultrasound vascular imaging of a mouse hindlimb using a SC-ULM, b PSF-CC ULM, and c power-Doppler methods. All the images are recorded from 1.5 sec of data acquisition (DAQ). d The image saturation as a function of DAQ times in SC-ULM (blue) and PSF-CC ULM images (black), respectively. The averaged saturation is calculated from 96125 pixels from three randomly selected regions of interest (ROI) in the vascular structures. The error bars are standard deviations (N = 3). The solid curves are fitted through an asymptotic regression model with an asymptote at 1. The characteristic times are 1.5 and 32.5 sec with SC-ULM and PSF-CC ULM, respectively. The latter is calculated through extrapolation because PSF-CC is slow and cannot reach the same saturation in 12 sec. e The zoom-in view of vascular structures using SC-ULM (top row) and PSF-CC ULM (bottom row) at various DAQ times from 0.5 to 7 sec, respectively. The area shown here is marked by the white rectangles in a and b. The result shows SC-ULM reveals the structure of major vessels at 0.5 sec and nearly complete structures in 1.5 sec, while the PSF-CC image remains sparse after 7 sec. f, g resolution measurements using the Fourier ring correlation (FRC) method of SC-ULM and PSF-CC ULM at 1.5 sec of DAQ in (e), respectively. h The resolution of both methods as a function of DAQ times. The averaged resolution is calculated from four random ROIs with the FRC method. The center line in each box is the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers range of each box is within 1.5 interquartile. Scatter plots of the data used for the boxplot are overlaid on each boxplot. *p = 0.0015, ns denotes not significant.

Fig. 4. In vivo dual PA/SC-ULM image of a mouse popliteal lymph node.

a Near-infrared fluorescence lymphatic imaging (IVIS) of indocyanine green (ICG) shows the locations of mouse popliteal lymph nodes (PLNs) where ICG accumulates. The ICG solution (20 µL, 0.2 mg ml⁻¹) was administered to both hind footpads and drained to PLNs within 15 min. The imaging area is 48 × 87 mm. b 3D ultrasound image of a mouse hind limp close to the popliteal fossa area. c 3D SC-ULM image shows the blood vessels within the hindlimb tissue near the PLN. d 3D PA imaging of ICG confirms the ICG accumulation in the PLN. Three laser wavelengths (750, 790, and 850 nm) are used for multi-wavelength PA imaging, and linear spectral unmixing is used for identifying ICG signal spectroscopically. e 3D dual PA/SC-ULM image. The PLN is labeled with ICG; the SC-ULM shows blood vessels surrounding/within the PLN. The 3D imaging volume in (b–e) is 22.9 mm (x) × 8.0 mm (y) × 1.6 mm (z).

Fig. 5. Motion correction of in vivo mouse renal vascular imaging.

a The images show raw ULM images of a kidney arterial and venous renal tree acquired through 1 sec (first column) and 4 sec (second column) of DAQ time. The images are processed with SC (first row) and PSF-CC method (second row), respectively. SC-ULM image reconstructs renal vasculature within a respiratory cycle (~1 sec), and PSF-CC requires at least (3 respiratory, ~4 sec, to reconstruct a visible renal structure. The long DAQ time greatly degrades the image quality due to the motion. b Motion-induced frame-to-frame correlation changes during 4 sec of ultrasound image recording. The identified motions include respiratory and cardiac motions. c A representative in-plan motion within a 2D ultrasound image. The detectable in-plan motions contain lateral (solid black), axial (dotted black), and rotational (solid orange) components. d The comparison before and after applying out-of-plane (step 1) and in-plane (step 2) motion-correction algorithms to the PSF-CC ULM image. e The comparison of motion-corrected SC-ULM kidney images acquired with one breathing cycle of DAQ time and three breathing cycles of DAQ time. The insets show the zoom-in view of the renal vasculature trees. f The image similarity comparing SC-ULM images before and after the motion correction for one breathing cycle (~1 sec) and three breathing cycles (~4 sec). The center line in each box is the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whisker range is 5th to 95th percentiles. Scatter plots of the data used for the boxplot are overlaid on each boxplot. *p < 0.0001.

Fig. 6. In vivo renal oxygenation and vascular image of a mouse kidney in an oxygen-challenging test.

a Directional blood flow map and b blood speed map of a mouse kidney when the mouse inhales 100% oxygen. c An ultrafast plane-wave ultrasound image of a mouse kidney. d–f The dual-contrast images show co-registered PA renal oxygenation image and ULM blood speed image, recorded in d 100%, e 3%, and f 20% of oxygen inhalations. g The zoom-in ULM blood flow speed maps represent the highlighted regions in (b) under various levels of oxygen inhalations. h Quantitative analysis of the blood speed change as a function of time over 150 sec of recording. i Quantitative analysis of the hemoglobin oxygenation changes as a function of time over 150 sec of recording. The red (and blue) curve represents the percent change of oxygenated (and deoxygenated) hemoglobin. The ultrasound image is recorded with a frame rate of 500 Hz and a central frequency of 15 MHz.