Non-Invasive 3D Photoacoustic Tomography of Angiographic Anatomy and Hemodynamics of Fatty Livers in Rats

Abstract

Non-alcoholic fatty liver disease is the most common liver disorder worldwide, which strongly correlates to obesity, diabetes, and metabolic syndromes. Complementary to mainstream liver diagnostic modalities, photoacoustic tomography (PAT) can provide high-speed images with functional optical contrast. However, PAT has not been demonstrated to study fatty liver anatomy with clear volumetric vasculatures. The livers of multiple rats are non-invasively imaged in vivo using the recently developed 3D PAT platform. The system provides isotropically high spatial resolution in 3D space, presenting clear angiographic structures of rat livers without injecting contrast agents. Furthermore, to quantitatively analyze the difference between the livers of lean and obese rats, the authors measured several PAT features and statistical differences between the two groups are observed. In addition to the anatomy, a time-gated strategy is applied to correct respiration-induced motion artifacts and extracted the hemodynamics of major blood vessels during the breathing cycles. This study demonstrates the capabilities of 3D-PAT to reveal both angiographic anatomy and function in rat livers, providing hematogenous information for fatty liver diagnosis. 3D-PAT, as a new tool for preclinical research, warrants further improvements to be transferred to human pediatric liver imaging.

Keywords: hepatic steatosis; liver imaging; non-alcoholic fatty liver; photoacoustic computed tomography.

Figure 1

Schematics and performance of the 3D‐PAT system. a) Schematic of the 3D‐PAT imaging system. The 1064 nm laser beam is expanded by a diffuser fixed on the bottom of the hemisphere. b) Close‐up view of the imaging aperture for rat experiments. The rat is placed on a heating pad with its liver at the center of the expanded laser beam. Two of the four arc‐shaped ultrasonic transducer arrays are marked. c) Orthogonal projection of a rat liver imaged by 3D‐PAT. The close‐up shows a slice within the volume with a blood vessel selected for resolution evaluation. Scale bar, 5 mm. d) PA amplitude profile along the yellow dotted line in (c), representing a full width half maximum (FWHM) of 380 µm.

Figure 2

3D‐PAT of lean and obese rat livers in vivo. Maximum amplitude projections (MAPs) of a lean rat liver in a) axial, b) sagittal, and c) coronal views. d) Coronal perspective projection of the lean rat liver angiogram with depth encoding. MAP images of an obese rat liver in e) axial, f) sagittal, and g) coronal views. ML, median lobe. LLL, left liver lobe. IRLL, inferior right lateral lobe. CP, caudate process. SEV, superior epigastric vessels. HPV, hepatic portal veins. h) Coronal perspective projection of the obese rat liver angiogram with depth encoding. Scale bars, 5 mm.

Figure 3

Quantitative comparison between lean and obese rat livers. a) Coronal perspective projection of the binary liver mask and vessel segmentation mask. BG, background; LM, liver mask; VM, vessel mask. b) Coronal perspective projection of the binary liver mask and vessel skeleton. VS, vessel skeleton. c) Perspective projection of AI with the binary liver mask. Statistical comparison of d) liver mask volume, e) VVO, f) VND, g) AI, and h) estimated SoS between lean and obese rat livers. Data are presented as box plots; p‐values are calculated using one‐tailed Welch's (unequal variances) t‐tests. **p < 0.01. Scale bars, 5 mm.

Figure 4

Schematics of calculating the angiographic irregularity (AI) and speed of sound (SoS) estimation. a) Schematics of vessel distribution diversity (VDD) calculation. A sliding window is scanned across the 3D image. At each position, the image histogram is acquired, and the entropy is calculated based on the normalized histogram counts. The value is then assigned to the center of the scanning window to form a 3D map M _VDD. b) Schematics of morphological irregularity (MI) calculation. At each x–y slice of the 3D image, a sliding window is scanned across the slice. The window is rotated from 0° to 180°, and the dominant normalized singular value is recorded at each angle. The MI at the window center is assigned according to the biggest difference in the dominant normalized singular value of all angles. Similar processes are repeated over all y‐z and x‐z slices to form the 3D map M _MI. The AI is calculated as the dot product of the two maps. c) Schematics of the speed of sound (SoS) estimation. The raw PA signal is universally back‐projected (UBP) with different SoSes in tissue to form the image, and 3‐D fast Fourier transform (3D‐FFT) is performed to acquire the spectrum. The image sharpness is calculated as the mean of the spectrum amplitude. The value that maximizes the sharpness is chosen as the estimated SoS.

Figure 5

Respiration‐based time gating schematics and results. a) Schematic of the respiration‐based time gating for dynamic images. Image shows a part of raw PA signal from one transducer, where periodic oscillations appear as the result of breathing. b) Close‐up raw PA signal for one respiration cycle. Each cycle is divided into multiple phases, and each phase is picked up through all cycles before summing over round‐trip signals for reconstruction. c) Motion‐contrast‐encoded orthogonal projection of a rat liver with a slice showing cross‐sections of SEV and HPV. d) Relative changes of the PA signals from two voxels in c). Data are normalized according to the first phase and plotted as means ± standard errors of the mean. e) Relative changes of the cross‐sectional vessel areas in ©. Data are normalized according to the first phase and are plotted as means ± standard errors of the mean.