2.1.

Acoustic Detection

All three arrays employed in this study were custom-made cylindrically focused curved arrays (Imasonic S.A., France) with varying geometrical properties, yet manufactured from the same piezocomposite material to cover the frequency band up to 7.5 MHz with peak sensitivity at 5 MHz. To facilitate a concise description for the remainder of this text, we introduce a nomenclature based on the number of elements per array: MSOT64 (64 elements), MSOT128 (128 elements), and MSOT256 (256 elements). Figure 1 illustrates apertures and overall arrangements of the three systems (left column MSOT64, right column MSOT128/256) and defines the coordinate system, which we will refer to in the following. The geometric characteristics of all three arrays and their respective single elements are summarized in Table 1.

Fig. 1

The first row shows the transducer geometries for (a) MSOT64 and (b) MSOT128/MSOT256. The second row illustrates the two system setups, including laser, data acquisition (DAQ), and sample holder with transparent membrane. (c) MSOT64 achieves even in-plane ($z=0$) illumination by using a ten arm fiber bundle arranged from one side at an angle of 33 deg relative to the circumference of the sample. (d) MSOT128/256 enables similar illumination by arranging a second ten arm fiber bundle symmetrically from both sides at an angle of 24 deg. Both fiber bundles cover an angle of 270 deg in the $xy$-plane.

Table 1

Geometric specifications (SE=single element).

	Azimuth span $\mathrm{\Phi }$ (deg)	Elevation span $\vartheta$ (deg)	Array radius ${\mathrm{R}}_{\mathrm{A}}$	SE radius ${\mathrm{R}}_{\mathrm{se}}$	SE pitch p (mm)	SE length L (mm)
MSOT64	172	21.6	40	40	1.880	15.08
MSOT128	270	20.2	40	40	1.470	14.10
MSOT256	270	21.6	40	37	0.735	13.95

The MSOT64 geometry is depicted in Fig. 1(a). We define the origin of the coordinate system to coincide with the center of the hypothetical sphere described by the array such that the $xy$-plane at $z=0$ represents the imaging plane offering a cross-sectional view of the object. Translation along the $z$ axis, thereby, enables full-body imaging. At $z=0$ the extended length of the curvature in elevation constitutes cylindrical focusing along the $z$ axis, while the comparably small pitch allows a sufficient field of view (FOV) for tomographic imaging in the $xy$-plane. As the elements of all three arrays in this study are composed of the same piezocomposite material, but with varying geometrical sizes, we chose as the norm a single element of the MSOT64 and gave it the area size and sensitivity of 100%. Both measures were determined by the manufacturer using standardized protocols and are, thus, comparable between arrays.

MSOT128, depicted in Fig. 1(b), enables a more complete tomographic view, which is expected to significantly improve image quality. A smaller pitch increases the FOV of a single element and, thereby, should improve transversal resolution, i.e., perpendicular to the radius. Consequently, single element area size is only 72% of the MSOT64 norm and the sensitivity 55%. Nonetheless, due to twice the number of elements compared to MSOT64, the resulting full array sensitivity within areas of overlapping FOV should be 110%.

MSOT256 shares most of the geometrical characteristics with MSOT128 and is also characterized by Fig. 1(b). However, the system employs twice the number of elements, whose elevation focus is closer (${\mathrm{R}}_{\mathrm{se}}=37\mathrm{mm}$) compared to the in-plane array focus (${\mathrm{R}}_{\mathrm{A}}=40\mathrm{mm}$). This geometry is expected to achieve the largest FOV and the best transversal resolution. Even though the area size of a single element is merely 36% of the MSOT64 norm and corresponding sensitivity 29%, total sensitivity of the MSOT256 is expected to be 114% within overlapping FOV.

Using an ultrasound field simulation program [Field II (Refs. 24 and 25)] and the geometrical characteristics of a single element, we have numerically simulated the expected sensitivity field up to 7.5 MHz in the imaging plane ($z=0$) for the single elements of each array—sensitivity being defined here as the maximum expected amplitude from a given position $(x,y,0)$. By appropriately rotating and summing single element sensitivity fields we can obtain compound sensitivity fields for the three arrays employed (Fig. 2). The short solid lines marked by arrows and an element number on the images indicate the first and last element of the transducer. Black ellipses designate the region of full-width half-maximum (FWHM), i.e., where the amplitude drops to at most 50% of the peak value. For MSOT64, depicted in Fig. 2(a), the expected field is characterized by a diameter at FWHM of 15.5 mm and a small displacement from the center of the array, due to the limited coverage of 172 deg. For MSOT128, shown in Fig. 2(b), calculations show a horizontal extent of 21 mm and a vertical extent of 17.1 mm, due to the asymmetric distribution of detectors. The reduced pitch and toroidal focusing of MSOT256 elements result in an FWHM diameter of 32.6 mm, shown in Fig. 2(c).

Fig. 2

Theoretical calculation of the two-dimensional sensitivity field of detection (0.5 to 7.5 MHz) within the imaging plane of the detector. Black ellipses mark the full-width half-maximum (FWHM) range. The first and last element of each array is noted on the images by a short solid line marked with an arrow and element number. (a) 64 element array covering 172 deg; diameter at FWHM: 15.5 mm. (b) 128 element array covering 270 deg; vertical diameter at FWHM: 17.1 mm; horizontal diameter at FWHM: 21.0 mm. (c) 256 element array covering 270 deg; diameter at FWHM: 32.6 mm. (d) Visibility region (gray) for MSOT64 and (e) MSOT128/256 that allows stable reconstruction of feature boundaries (ROI, region of interest).

A second parameter relating to image quality is the angular coverage of the imaging plane. It has been shown²⁶ that boundaries, whose normals do not pass through at least one detection element, are invisible and after reconstruction lead to blurring or reconstruction artifacts.²⁷ For better understanding, Figs. 2(d) and 2(e) characterize visibility for MOST64 and MSOT128/256, respectively. The detection or visibility region is, thereby, highlighted in gray and marks the region in which reconstruction of all features and boundaries is stable. The complementary white region with the dashed circle, on the other hand, allows only some boundaries to be reconstructed stably. In MSOT64, less than half of a 30-mm-diameter region of interest (ROI) lies within the visibility region, while for MSOT128/256 the ROI is fully covered.

2.2.

Experimental Arrangement

The experimental arrangement utilized was comparable to a system previously described.^18,28 Measurements employed a wavelength tunable pulsed laser (Phocus™, Opotek Inc., USA) with a pulse duration of $<10\mathrm{ns}$, repetition rate of 10 Hz, and peak pulse energy of 90 mJ at 750 nm. The laser beam was coupled into a ten arm fiber bundle. The MSOT64 system employed a bundle with rectangular outputs of size $3.2\times 0.72{\mathrm{mm}}^2$ at the distal end. The arms were arranged over an arc of 270 deg next to the detection array, as shown in Fig. 1(c), and established ring illumination at the imaging plane. The systems housing the MSOT128 and MSOT256 employed a slightly different illumination arrangement, depicted in Fig. 1(d). Ring illumination was implemented by five rectangular outputs of $12.4\times 0.2{\mathrm{mm}}^2$ size covering 270 deg in azimuth on both sides of the arrays. In all setups, the maximum light fluence at the surface of a cylinder of diameter 20 mm, as depicted in Fig. 1, did not exceed $17\mathrm{mJ}/{\mathrm{cm}}^2$ at 750 nm. Each array was connected to a custom-made data acquisition (DAQ) system that digitized up to 512 channels in parallel at 10 Hz repetition rate and $40\text{MSamples}/\mathrm{s}$. All DAQs low-pass filtered the input signals at 15 MHz to avoid aliasing and subsequently amplified them by the same gain. The three detection systems ($\mathrm{DAQ}+\text{array}$) had comparable electrical impulse responses, an instance of which was used to deconvolve all signals prior to reconstruction.

Image reconstruction was performed using a model-based algorithm.²⁹ The algorithm computes a model-matrix that links measured pressure values with a discrete grid of positions (image pixels) in the imaging plane by numerically evaluating the forward solution of the optoacoustic wave equation. Model inversion was achieved using a regularized iterative least-squares algorithm. All imaging results presented in this study were attained using the same set of parameters, i.e., $450\times 450\text{pixels}$ over $30\times 30{\mathrm{mm}}^2$. Due to the large model-matrix and iterative inversion, reconstruction time per image necessitated offline reconstruction after the data acquisition.

2.3.

Numerical Simulation

To facilitate objective observations between the three detectors, we employed numerical simulations. The simulations assumed a rectangular region of $30\times 30{\mathrm{mm}}^2$ centered at the origin in the $xy$-plane ($z=0$) and placed 900 microspheres equally spaced by 1 mm in $x$- and $y$-direction. Figure 3(a) shows the numerical setup, which also represents the reconstruction target. To obtain acoustic data for the three arrays, we first employed an ultrasound field simulation program [Field II (Refs. 24 and 25)] to compute the acoustic field received by each detection element following a Dirac-like excitation at the 900 microsphere positions. Then we convolved these geometric impulse responses, sampled at $40\text{MSamples}/\mathrm{s}$, with the analytical optoacoustic signal emitted by a spherical absorber³⁰ and obtained acoustic data frames of size $2030\times 64$, $2030\times 128$, and $2030\times 256$. Microsphere size was chosen to span six acoustic samples at $40\text{MSamples}/\mathrm{s}$, i.e., 125 ns or 187.5 μm at $1500\mathrm{m}/\mathrm{s}$ speed of sound.

Fig. 3

(a) Numerical setup covering a rectangular region of $30\times 30{\mathrm{mm}}^2$—900 microspheres of 187.5 μm diameter equally spaced in 1 mm steps. (b) Top view photograph of the microsphere-containing phantom showing the sphere distribution and the fluence distribution. The disk diameter was 30 mm; the polyethylene spheres attained diameters of 180 to 212 μm. (c) Schematic of the experimental phantom assembly.

2.4.

Phantoms

Known phantoms were also employed to provide for control measurements that could compare the performance of the three detectors employed. Validation of numerical results was achieved using black polyethylene (PE) microspheres of diameters ranging from 180 to 212 μm (Cospheric LLC, USA). Due to practical difficulties replicating the regular sampling grid used in the numerical simulation, we decided to implement an irregularly sampled disk of 30 mm diameter. For this, we prepared a cylindrical tissue mimicking agar (2% w/v) phantom of 50 mm height and 30 mm diameter including a fatty emulsion imparting a reduced scattering coefficient of ${\mu }_{\mathrm{s}}^{\prime }=10{\mathrm{cm}}^{-1}$. The phantom also included a cylindrical cavity at the bottom of 25 mm length and 8 mm diameter. We then dispersed microspheres on top of the phantom aiming to achieve an as even distribution as possible. This way the microspheres described an absorbing disk, which was illuminated from one side through $\sim 25\mathrm{mm}$ of scattering agar using a cylindrical fiber bundle inserted through the bottom cavity of the phantom. Fluence variations within the disk were captured by a CCD camera at 680 nm and these measurements were later employed for correction of the reconstructed image. Figure 3(b) shows the acquired photograph from the back-illuminated microsphere layer.

2.5.

Mouse Imaging In Vivo

To extend numerical and phantom findings to tissue observations, we imaged an eight-week-old female CD-1^® nude mouse in vivo (Charles River Laboratories, Germany). Animal handling was conducted in compliance with protocols approved by the Bavarian Animal Research Authority. The animal was anesthetized using 1.8% Isoflurane (Forene©, Abbott AG, Switzerland) vaporized in 100% oxygen at $0.8\mathrm{L}/\min$ and subsequently placed within an animal holder, as described in Ref. 28, in a supine position relative to the arrays. The holder, designed to fit into all three systems, ensured identical positioning of the animal in relation to the detector for the three experimental arrangements. The imaging protocol involved a single-wavelength full-body scan at 850 nm with a step size of 0.5 mm. At each step we recorded 50 frames to capture periodically occurring motion induced by heart beat or breathing. This imaging sequence was repeated for all three systems.

3.1.

Numerical Simulation

Results obtained from numerical simulation are shown in Fig. 4. Figure 4(a) corresponds to the reconstruction obtained from MSOT64, Fig. 4(b) from MSOT128, and Fig. 4(c) from MSOT256. All images were normalized to 1 and negative values were set to zero as they have no physical meaning. Two rectangular regions ($4\times 4{\mathrm{mm}}^2$) from the center and the periphery of each array are shown magnified at the bottom of each of the images they were taken from. The images demonstrate two major differences. First, a broader region of the object is visualized for an increasing number of elements employed, as predicted by the sensitivity fields of Fig. 2. The second difference regards transversal resolution, i.e., the resolution perpendicular to radial lines originating in the middle of the image. Close to the array center, all systems reconstruct the simulated microspheres with a similar performance and accuracy in obtaining their actual dimensions as shown at the bottom right magnification insets in Figs. 4(a) to 4(c). However, the further from the image center a microsphere is positioned, the more its reconstruction is transversally elongated. The strength of this effect is directly related to the pitch of the detector elements in each array, i.e., MSOT64 elements have the largest pitch and cause the strongest distortions; MSOT256 elements have the smallest pitch and the least distortions. The bottom-left magnification insets in Figs. 4(a) to 4(c) illustrate this effect.

Fig. 4

Top row shows simulated reconstruction results from (a) MSOT64, (b) MSOT128, and (c) MSOT256. For each image, negative values were removed and the remainder normalized to 1. Details at the periphery and the center, framed ($4\times 4{\mathrm{mm}}^2$) and color-coded by array, were magnified for display at the bottom left and right, respectively. Graphs below depict diagonal cross-sections (top-left to bottom-right) of the magnified detail images for (d) the periphery and (e) the center.

A quantitative assessment of both improvements seen as a function of detector elements employed is given in Figs. 4(d) and 4(e), which show a profile (cross-section) of the six magnified images from top-left to bottom-right. Close to the image center, Fig. 4(e), all geometries correctly resolve a microsphere diameter of 190 μm at FWHM with comparable sensitivity (amplitude). The noise appearing between the four distinct peaks is a consequence of the high resolution and large imaging region chosen ($30\times 30{\mathrm{mm}}^2$), which required a degree of regularization inversely proportional to the number of detectors, i.e., stronger for MSOT64 than for MSOT256. On the other hand, transversal resolution at 10 mm distance from the array center, shown in Fig. 4(d), drops for all systems. MSOT64 resolves elongated microspheres of length $\sim 730\mu \mathrm{m}$; MSOT128 achieves 450 μm and MSOT256 270 μm for an actual microsphere size of 187.5 μm. On the radial axis, all systems resolve microsphere width as $\sim 250\mu \mathrm{m}$. Figures 4(d) and 4(e) also illustrate the drop in amplitude when moving further from the array center. While all systems exhibit reduced sensitivity with increasing distance from the center, the drop is sharpest for MSOT64 and marginally better for MSOT128. By contrast, MSOT256 achieves twice the sensitivity of MSOT64.

3.2.

Phantom Measurements

Experimental results from the PE microsphere phantom are shown in Fig. 5, which follows a figure arrangement similar to Fig. 4. Due to irregular sampling, we selected three central (4 to 6) and three peripheral (1 to 3) microspheres, which were captured sufficiently well by all geometries, and plotted their transversal profiles in Figs. 5(d) and 5(e). We note that variations in amplitude and shape within close spatial proximity, e.g., within the magnification insets, are due to experimental imperfections such as microsphere clustering, see Fig. 3(b), alignment between phantom and array center or elevational displacement of individual microspheres (on the order of hundreds of micrometer along $z$ axis). Nonetheless, selected microspheres confirmed the observations derived from numerical simulation. In particular, microspheres 4 to 6 from the array center show similar shapes, amplitudes, and resolutions for all systems. Figure 5(e) shows profiles obtained along the diameter of the microspheres revealing an FWHM of $\sim 200\mu \mathrm{m}$ for all detectors employed. Conversely, selected microspheres (1 to 3) from the periphery were resolved with different diameters as a function of the number of detector elements employed, as shown in Fig. 5(d). We also observed higher sensitivity for MSOT128 compared to MSOT64 and the highest sensitivity for MSOT256. Similar observations can be made for transversal resolution in 10 mm distance from the array center (microsphere 3): MSOT256 resolves a width of 200 μm, MSOT128 300 μm, and MSOT64 450 μm. In relative terms, i.e., from MSOT64 over MSOT128 to MSOT256, improvements in transversal resolution were expected; compared to simulation, however, absolute resolutions seem too high. The explanation for this apparent outperformance of simulation results lies in the necessary light fluence correction and segmentation, which both are nonlinear operations but consistent for all arrays. Considering the pixel size of $67\mu \mathrm{m}/\text{pixel}$, the transversal resolution discrepancy for MSOT256 is $\sim 1\text{pixel}$, for MSOT128 close to 2 pixels, and for MSOT64 close to 4 pixels. Thus, experimental results from the phantom confirm the superior image quality of MSOT256.

Fig. 5

Top row shows experimental phantom reconstruction results from (a) MSOT64, (b) MSOT128, and (c) MSOT256. For each image, negative values were set to zero and the remainder normalized to 1. Furthermore, images were segmented based on their local maxima. Details at the periphery and the center, framed ($4\times 4{\mathrm{mm}}^2$) and color-coded by array, were magnified for display at the bottom left and right, respectively. Graphs below depict diagonal cross-sections of individual microspheres (circled blue, red, and green) for (d) the periphery (numbered 1 to 3) and (e) the center (numbered 4 to 6). Individual cross-sections were artificially spaced to allow sufficient distance for comparison.

3.3.

Mouse Imaging In Vivo

Figure 6 shows the results obtained from the same nude mouse imaged sequentially and without averaging in all three systems. To qualitatively compare achievable imaging performance, we have selected three content-rich anatomical regions of increasing diameter. As the mouse was identically positioned during all three scans and we took 50 frames per position, we were able to precisely choose the same cross-sectional image at the same respiratory time point for all three systems. Columns of Fig. 6 show results by system, with the last column depicting cryoslices at equivalent positions from a comparable mouse.

Fig. 6

Single-shot in vivo mouse images at 850 nm wavelength: the first column shows results obtained from MSOT64, second column that from MSOT128, and third column that from MSOT256. The last column shows mouse cryoslices at equivalent positions. The mouse head is shown in (a) to (d): 1. sagittal sinus; 2. temporal artery; 3. extra-cranial blood vessel; 4. deep cerebral vessel. The liver region is shown in (e) to (h): 5. spinal cord; 6. aorta; 7. vena cava; 8. vena porta; 9. liver; 10. stomach. The kidney region is shown in (i) to (l): 11. kidney; 12. spleen.

Analysis of the mouse head revealed that all arrays captured large vascular structures on the periphery as well as deep inside the brain. Compared to other body segments, the mouse head was covered reasonably well by all geometries. The temporal artery (marked 2), the sagittal sinus (marked 1), and deep-seated extra-cranial (marked 3) and cerebral blood vessels (marked 4) were visible in all cases. However, closer inspection of Figs. 6(a) to 6(d) reveals improving resolution for increasing number of detector elements. For example, the sagittal sinus is only vaguely captured by MSOT64 with a transversal diameter of 600 μm at its widest point, whereas it is more accurately reconstructed with MSOT128 (550 μm diameter) and MSOT256 (450 μm diameter). Similar inspection of the temporal artery revealed a reconstructed size of 670, 470, and 380 μm from MSOT64, MSOT128, and MSOT256, respectively. Comparison of the MSOT256 image and the cryoslice photograph on Fig. 6(d) shows that MSOT256 achieves the most accurate reconstruction of vessels over MSOT64 and MSOT128, offering an overall appearance that is very close to the one seen on the photograph.

MSOT results from the liver level are depicted in Figs. 6(e) to 6(h). MSOT64, shown in Fig. 6(e), allows localization of larger features, such as the spinal cord (marked 5), the aorta (marked 6), or the vena cava (marked 7). However, several anatomical details in the periphery of the animal are lost, consistent with the findings in Figs. 4 and 5. Conversely, MSOT128, Fig. 6(f), and MSOT256, Fig. 6(g), reveal finer structures and more elaborate views of vasculature showing improved peripheral resolution and fewer artifacts.

Similar observations were obtained from images at the kidney region, where the diameter of the mouse body was $\sim 23\mathrm{mm}$. Figure 6(i), obtained from MSOT64, exhibits resolution drop and visualization of coarse features, especially in the animal periphery as in Fig. 6(e) and affords only limited transversal resolution. Nonetheless, characteristic landmarks (spinal cord, aorta, vena cava) and large organs (both kidneys marked 11, spleen marked 12) can be identified. Figure 6(j) depicts the MSOT128 result and shows better clarity in resolving anatomical features compared to MSOT64. As expected from the liver cross-section, the best image quality is achieved by MSOT256, depicted in Fig. 6(k). The significantly improved transversal resolution and sensitivity field is best demonstrated in the right kidney (marked 11).

Due to its feature-rich content, the liver region enables a detailed comparison of cryosection and MSOT images. Figure 7 compares the performance of MSOT256 against a corresponding cryoslice photograph serving as the gold standard, to offer a more elaborate explanation of the signals observed on the optoacoustic images. Five significant landmarks have been encircled and numbered (1 to 5), whereby higher brightness in the MSOT image, Fig. 7(a), corresponds to darker colors in the cryosection, Fig. 7(b). The first landmark marks the vasculature surrounding the spinal disk; its structure comprises a bright dot at the center top, which is characteristic of a blood vessel perpendicular to the imaging plane, and a surrounding bracket indicating blood vessels in-plane. Landmarks 2 to 4 highlight different regions within the liver. Area 2 is dominated by elongated in-plane vasculature bifurcating from the vena cava and vena porta toward the periphery. On the other hand, areas 3 and 4 contain vessels perpendicular to the imaging plane—bright dots that correspond to dark dots in the cryosection. Additionally, MSOT highlights short branches originating from these dots. Such features can be attributed to vascular branching close and perpendicular to the imaging plane. Finally, we also marked the stomach (area 5) because of its prominent position and sharp boundary.

Fig. 7

Mouse cross-section at the liver level. (a) MSOT256 result in vivo at 900 nm wavelength. (b) Equivalent cryoslice obtained from a similar mouse. Corresponding features have been encircled and numbered: 1. Spinal cord; 2 to 4. vascular structures within the liver; 5. stomach.

To better understand the relative in vivo performance of the three systems employed, Fig. 8 offers a quantitative comparison between the reconstructions obtained by the 64-, 128-, and 256-element systems on the spinal disk. Image profiles taken from the radial and transversal directions (marked r and t) are highlighted in Fig. 8(a) as dashed (MSOT64), dotted (MSOT128), and solid (MSOT256); corresponding traces are shown in Figs. 8(b) and 8(c). A blood vessel perpendicular to the imaging plane was selected to mark the intersection point for orientation purposes. This vessel could be resolved with 128 and 256 elements, but not with 64 due to its close proximity to the top and sides of the surrounding structure. The radial profile illustrates how the transversal resolution of the three systems can affect the result. In MSOT64, the vessel and the adjacent top edge are fused with emphasis given to the edge. In MSOT128, the vessel gives the strongest signal, however, it cannot be resolved from the edge. Finally, only MSOT256 is able to clearly determine the distance between the two as 280 μm (peak to peak). On the other hand, from the transversal profile, the vessel diameter can be resolved as 560 μm using MSOT128 and as 350 μm with MSOT256. MSOT64 again cannot resolve the vessel due to a lack of sensitivity and transversal resolution.

Fig. 8

Profiles of spinal cord at liver level. (a) Magnifications for MSOT64, MSOT128, and MSOT256. Cuts in radial (marked r) and transversal (marked t) directions are line-coded by array. (b) Profile of radial trace from top to bottom. (c) Profile of transversal trace from left to right. Prominent features: top and sides of in-plane vascular structure, blood vessel perpendicular to the imaging plane.