All-optical in vivo photoacoustic tomography by adaptive multilayer acoustic backpropagation

Abstract

Photoacoustic tomography (PAT) combines high optical contrast with deep acoustic penetration, making it valuable for biomedical imaging. However, all-optical systems often face challenges in measuring the acoustic wave-induced displacements on rough and dynamic tissues surfaces. We present an all-optical PAT system enabling fast and high-resolution volumetric imaging in vivo. By integrating holographic microscopy with a soft cover layer and coherent averaging, the system detects ultrasound-induced surface displacements over a 10 × 10 mm² area with 0.5 nm sensitivity in 1 s. A novel backpropagation algorithm reconstructs a depth-selective pressure image from two consecutive displacement maps. The coherent summation of these depth-selective pressure images enables the reconstruction of a 3D acoustic pressure image. Using adaptive multilayer backpropagation, we achieve imaging depths of up to 5 mm, with lateral and axial resolutions of 158 µm and 92 µm, respectively, demonstrated through in vivo imaging of mouse vasculature and chicken embryo vessels.

Keywords: Backpropagation; Holography; In vivo imaging; Photoacoustic tomography.

Fig. 1

Schematic of full-field optical interferometric imaging of photoacoustic waves at the specimen surface. (a) When an excitation pulse with a beam size of several millimeters is irradiated to the specimen, each absorber inside the specimen absorbs light energy and generates a pulse of acoustic wave. (b) The photoacoustic wave propagates toward the surface and induces displacement at the top surface of the cover layer, at $z=-d$. (c) By adjusting the time delay $t$ between the excitation and the probe pulses, the 2D displacement maps are captured in a time sequence. Color bars: phase in radians. (d) By taking time derivative with the acquired displacement maps, the pressure maps are calculated. These maps visualize the expansion or propagation of pressure waves with time. Color bar: scaled pressure. Scale bars in (c)-(d): 2 mm.

Fig. 2

The single-sheet backpropagation (SSB) method for 3D reconstruction from a single surface pressure map. (a) Each row illustrates that the wave having a same lateral spatial frequency $\mathbf{q}$ can be decomposed with multiple plane waves having different propagation constant $\beta$’s (different columns). (b) Temporal frequency spectra of surface pressure maps obtained with a line-shaped PET fiber embedded in PDMS. (c) The spectrum (blue solid line) averaged along the direction ($y$-direction) of the PET fiber and fitted with a parabola (red dotted curve). (d) Maximum intensity projection images of a knot-shaped phantom reconstructed using 1, 3, and 20 surface pressure maps with the proposed SSB method, and (e) a 3D view reconstructed using 20 maps. (f) For objects located crosswise with a depth difference of 1 mm, a depth-selective thin-volume image was reconstructed by using a single pressure map (first, second) taken at different times, and the full-volume image by using 40 pressure maps (third). (g) The 3D view reconstructed using 40 pressure maps. Scale bars in (d)-(g): 1 mm.

Fig. 3

Experimental demonstration of coherent averaging and adaptive backpropagation. (a) An ‘AOL’ shape structure was made with a black PET fiber of 200 μm diameter (top), which was embedded within an optically opaque PDMS made by mixing with 0.3 % weight TiO₂ particles (bottom). (b) The 2D displacement maps captured at the PDMS surface for different time delays. (c) The map coherently averaged with 20 displacement maps, after compensating for the overall phase drift. (d) The 3D image reconstructed by using the acoustic velocity of 1076 m/s, a literature value for the 10:1 ratio PDMS. (e) The image reconstructed by applying adaptive backpropagation algorithm for finding the proper acoustic velocity. At the velocity of 1045 m/s, we could reconstruct the image with optimal contrast and resolution. Scale bars in (b)-(e): 2 mm.

Fig. 4

Experimental demonstration of the multilayer acoustic backpropagation algorithm. (a) Two 300 μm diameter pencil leads were placed orthogonal to each other in a petri dish at a depth difference of about 1 mm. The dish was filled with DI water and covered with a 3.44-mm-thick PDMS block. (b) The hologram taken at the top surface of the block (left) and the extracted surface displacement map (right). (c) Same as (b), but taken at the bottom surface of the PDMS block. The white and black scale bars in (b)-(c): 200 μm and 2 mm, respectively. (d) Sharpness metric calculated with various acoustic velocities $v_c$ and $v_m$. (e) The 3D image reconstructed without accounting for the propagation angle-dependent attenuation due to acoustic impedance mismatch. (f) The same image but reconstructed after compensating the propagation angle-dependent attenuation. Scale bars in (e)-(f): 2 mm.

Fig. 5

In vivo imaging of vasculatures in mouse hindlimb leg and chicken embryo’s CAM. (a) A 5-week-old nude mouse placed in the sample arm of the all-optical PAT system for in vivo imaging through the intact skin. (b) A photograph of the blood vessels in the hindlimb, taken after imaging session, which was used as ground truth. (c) A photograph taken after further removing the fat tissue revealing vascular structures underneath (area 1). Two vessels of artery and vein are distinctly visible in area 2. (d) The 3D image of the mouse hindlimb reconstructed with the multilayer backpropagation method. Scale bars in (b)-(d): 2 mm. (e) A photograph and its enlarged image of the blood vessels formed in the CAM of chicken embryo incubated for 11 days. Scale bar: 1 mm. (f) Reconstructed volumetric vasculature image of the CAM. Scale bar: 2 mm.