2.1.

Dual-Modality Preclinical Imaging System

The TriTom is a dual-modality imaging platform that includes both a PA detector array and an FL imaging scientific complementary metal-oxide-semiconductor (sCMOS) camera. A top-down diagram [Fig. 1(a)] of the imaging chamber and data acquisition process details the system’s key components. The cylindrical imaging chamber (1) has an inner diameter of 190 mm and wall thickness of 5 mm made of glass quartz. The rotational stepper motor (2) (K10CR1, Thorlabs, Newton, New Jersey, United States) has a hollow shaft placed through it that acts as the mount point for samples. The motor’s maximum speed of rotation is $10\mathrm{deg}/\sec$ and determines the time of a 360-deg TriTom scan as 36 s. The orthogonal excitation termini slots (3), orthogonal to the scanning plane of the PA array, are positioned on the outer circumference of the cylindrical imaging chamber. The epi-excitation termini slots (4) are positioned 60 deg from the orthogonal excitation termini slots towards the PA detector array. All excitation termini are vertically aligned with the centermost element of the PA array. Randomized fiber bundle outputs, of lightbar surface dimensions $1\times 30{\mathrm{mm}}^2$, are slotted into the excitation termini slots (3, 4). The fiber bundle inputs are routed to the laser’s (PhotoSonus, Ekspla, Vilnius, Lithuania)³⁴ optical parametric oscillator (OPO) output. The laser’s OPO output had a wavelength range from 670 to 1064 nm and was pulsed at 10 Hz with a 5-ns pulse duration.

Fig. 1

Diagram of the (a) TriTom imaging chamber layout and (b) assembled TriTom imaging chamber. Quartz glass water tank (1), rotational stepper motor and sample/animal mounting point (2), orthogonal excitation termini slots (3), epi-excitation termini slots (4), PA detector array (5), FL imaging camera, adjustable objective, and optical filter wheel (6), data acquisition board (7), and Windows 10 PC (8). Laser excitation (${\mathrm{E}}_{\mathrm{ex}}$) light is shown as a wavy red arrow, FL emission (${\mathrm{E}}_{\mathrm{em}}$) light is shown as a wavy yellow arrow, a PA wave is shown as a curved blue line, and the direction of data acquisition flow and triggers are shown as blue arrows.

The acoustic waves induced by laser excitation are detected by a curve-linear 96-element PA transducer detector array (5) (Imasonic SAS, Voray sur l’Ognon, France). The piezoelectric transducer elements ($1.3\times 1.3{\mathrm{mm}}^2$) were positioned on the centerline of the curved array with inter-element spacing of 0.1 mm. The center frequency of the array was $6\mathrm{MHz}\pm 10\%$ (at $-6\mathrm{dB}$) with bandwidth $\ge 55\%$. The array was positioned vertically with the centermost element aligned to the rotational axis, which was spaced 65 mm from the array’s center element. For FL imaging, an sCMOS camera ($2048-2040\text{pixels}$) (6) (Dhyana 400D, Tucsen, Fuzhou, China) was positioned outside of the imaging chamber with two optical components between it and the chamber’s circumference. The camera had a quantum efficiency (QE) of 45% at 800 nm, but had an operational ($>20\%$ QE) wavelength range of 400 to 900 nm. The camera’s readout noise was 2.5e- for its high dynamic range mode (16-bit). A combination objective lens and aperture (Contrastech, Hangzhou, China) with a focal length of 50 mm and f/2.0 to 22 was placed directly after the camera. A filter wheel (Starlight Xpress, Berkshire, United Kingdom) was mounted inside the imaging module that allows the camera’s acquisition of FL images, with an appropriate optical filter selected. The described configuration resulted in an FL imaging optical detection channel with a $40\times 40{\mathrm{mm}}^2$ field-of-view (FoV) plane focused at the rotational axis.

The signals from the PA detector are captured by a 128-parallel channel data acquisition board (DAQ) (7) (DAQ128, PhotoSound Technologies, Houston, Texas, United States) capable of acquiring 4096 samples per channel at a sampling rate of 40 MHz. The DAQ had a gain range of 46 to 91 dB (40 dB preamplifier, 6 to 51 dB programmable gain) and was optically triggered by a single fiber routed from the fiber bundle input. Captured signals were filtered through a 10 MHz low-pass filter and transferred to a Windows 10 PC (8) via a USB 3.0 connection, along with data from the stepper motor and camera, for processing. This combination of collected data was then used to reconstruct PA and FL volumes.

2.2.

Characterization of the TriTom Imaging Platform

Three phantoms were prepared to characterize the spatial resolution, sensitivity, and multispectral imaging capabilities of the PA imaging modality. Two phantoms were prepared to evaluate the optical spatial resolution and sensitivity of the FL imaging modality. The laser’s OPO output energy was set to $80\mathrm{mJ}/\text{pulse}$ for all wavelengths used in phantom experiments, to minimize variations in fluence for each scan. The TriTom imaging chamber was filled with degassed deionized (DI) water at a temperature of $25.0°\mathrm{C}\pm 0.5°\mathrm{C}$ for all phantom scans.

2.2.1.

Photoacoustic spatial resolution phantom

The PA spatial resolution phantom [Figs. 2(a), 2(b), and 2(d)] was assembled by threading black polyamide 6,10, $Ø50\mu \mathrm{m}$, monofilament strings through a resolution phantom matrix. A total of fourteen strings were used to construct the PA spatial resolution phantom with seven threaded vertically (spaced 4 mm apart) and seven threaded horizontally (spaced 10 mm apart).

Fig. 2

Diagrams of the PA spatial resolution phantom’s (a) top view and (b) side view, as well as the PA sensitivity phantom’s top view (c); the multispectral phantom had a similar configuration to the PA sensitivity phantom. Photos of (d) the PA spatial resolution phantom, (e) PA sensitivity phantom, and (f) a phantom installed in the system’s imaging chamber. Key components: $Ø50\mu \mathrm{m}$ monofilament matrix (1) and sample microcuvettes (2).

The assembled phantom was scanned using the system’s PA modality at an excitation wavelength of 800 nm. The PA spatial resolution of the system was determined by extracting line profiles of vertical and horizontal monofilaments from the reconstructed volume with a voxel size of 0.02 mm (Fig. 3). Gaussian distributions were fitted over the line profile data and the full width at half max (FWHM) was calculated as the spatial resolution metric. To determine the transverse (axial) resolution, radial and tangential line profiles were measured from each vertical monofilament between a $\pm 10\mathrm{mm}$ vertical window with 1-mm steps [Figs. 3(a), 3(c), and 3(e)]. Since the transverse PA spatial resolution would be the average of the radial and tangential FWHM, an elliptical eccentricity chart was calculated using these two FWHM values for each vertical monofilament line profile measurement [Fig. 3(f)]. Similarly, the vertical resolution was measured by taking the vertical line profile of the two visible horizontal monofilaments between a $\pm 10\mathrm{mm}$ horizontal window with 1-mm steps [Figs. 3(b) and 3(d)].

Fig. 3

Red dashed line profile on a transverse (a) and longitudinal (b) slice of the spatial resolution phantom monofilaments. (c)–(d) Sample Gaussian fits over measured line profiles for the $x$-axis (c), $y$-axis (e), and $z$-axis (d). (f) Heatmap of eccentricity values from the ellipse formed by the $x$-axis and $y$-axis FWHM values.

2.2.4.

Optical spatial resolution phantom

The optical spatial resolution phantom used was a 1951 USAF resolution chart (Edmund Optics, Barrington, New Jersey, United States) [Fig. 4(a)], a standard in optical spatial resolution measurement.⁴⁰ The phantom was mounted onto the TriTom’s hollow shaft perpendicular to the imaging axis, and the objective was manually adjusted to focus at the phantom’s plane with a fully open aperture and filter wheel set to open (no optical filters). The camera’s vertical and horizontal optical spatial resolution was measured both in air [Fig. 4(b)] and with the TriTom imaging chamber filled with DI water [Fig. 4(c)].

Fig. 4

Images of the optical resolution phantom: (a) a 1951 USAF resolution test chart outside of the imaging chamber, (b) inside the imaging chamber with air, and (c) inside the imaging chamber filled with DI water.

Vertical and horizontal line profiles were taken from the elements of the chart. If the three distinct peaks of each element were detected in its line profile, then the element is considered to be resolvable (Fig. S1 in the Supplementary Material). The optical resolution was determined to be the smallest resolvable element of the chart. The line pairs per mm values associated with each element were converted to microns, to be consistent with the reported PA spatial resolution.

2.3.

Animal Studies

To demonstrate the in vivo imaging capabilities of the TriTom, three mice were imaged using TriTom instruments at two research institutions. The first and second animals’ experimental protocol was approved by the Institutional Animal Care and Use Committee at the Georgia Institute of Technology (GIT). The third animal’s study was approved by the Ethics Committee of the Saratov State University (SSU). All mice were anesthetized with isoflurane gas and imaged in a TriTom with water temperature at $37.0°\mathrm{C}\pm 0.5°\mathrm{C}$ (Fig. S2 in the Supplementary Material).

The animals imaged at GIT were 6-week-old BALB/c nude mice (Charles River Laboratories, Wilmington, Massachusetts, United States) that had been injected with a $50\text{-}\mu \mathrm{l}$ dose of $50\mu \mathrm{g}/\mathrm{ml}$ glycol-chitosan-coated gold nanospheres (GC-AuNPs) mixed with $50\mu /\mathrm{ml}$ ICG (Sigma-Aldrich, St. Louis, Missouri, United States) prior to imaging. The first mouse was imaged 6-hr post injection while the second mouse was imaged 24-hr post injection. The GC-AuNP and ICG mixture was administered as a single bolus injection subcutaneously to the right 4’th mammary fat pad for both animals. This injection site was chosen because it is not close to highly optically absorbent blood-rich organs. The purpose of the study was to track the regional clearance of the injected contrast agent in vivo. Additionally, a PTFE microcuvette filled with copper sulfate ($\mathrm{OD}=1.2{\mathrm{cm}}^{-1}$ at 800 nm) was taped to the back of the mice as a PA fiducial. During imaging, the animals were anesthetized with 2% isoflurane mixed with oxygen, secured in the mouse restrainer, and mounted in the TriTom imaging chamber. PAI scans were then acquired at 532 and 710 nm.

The first mouse was excited with a wavelength of 532 nm to highlight superficial structures of the mouse as well as being close to an absorption peak of the GC-AuNPs (Fig. S3 in the Supplementary Material). Similarly, 710 nm was close to an absorption peak of the GC-AuNPs and would also highlight deoxygenated hemoglobin (Hb) filled organs, due to Hb’s relatively high optical absorption with respect to oxygenated hemoglobin (${\mathrm{HbO}}_2$).

The second mouse was excited with a wavelength of 770 nm to induce FL emission from the ICG component of the dye. A band-pass optical filter 814 to 851 nm was used to filter out the laser excitation light and pass the ICG FL emission to the optical detector. The IVIS Lumina II In Vivo Imaging System (Caliper Life Sciences, Hopkinton, Massachusetts, United States) was used in combination with the IVIS Acquisition Software for an FL imaging comparison. FL images were acquired targeting ICG with a 745-nm excitation filter and an 840-nm emission filter with a 6.5-cm field of view. During data collection, the mouse was anesthetized with the Visualsonics Inc. Anesthesia System (2% isoflurane mixed with oxygen) integrated into the IVIS imaging system via a nose cone.

For the third in vivo imaging study, a female BALB/c mouse was imaged in the TriTom system at SSU. The mouse was anesthetized using 3% isoflurane mixed with air and the hair covering the imaging area was trimmed and removed with depilatory cream. The animal was then mounted in the mouse restrainer and transferred to the TriTom for imaging. A single PAI scan was acquired at 1064 nm with an average laser energy of $380\mathrm{mJ}/\text{pulse}$ to generate deep anatomy images of ${\mathrm{HbO}}_2$-rich structures.

2.4.

Reconstruction and Visualization

3.1.

Photoacoustic Spatial Resolution

PA reconstructions of the spatial resolution phantom are shown in Fig. 5. All seven vertical monofilaments were detected in the resolution phantom’s SR transverse reconstruction slice [Fig. 5(a)]; similarly, two of the horizontal monofilaments were detected in the SR longitudinal reconstruction slice [Fig. 5(c)], albeit with lower sensitivity. The HR reconstruction slices [Fig. 5(b) and 5(d)] show a $10\times$ magnified view of a single monofilament. The entire volume was rendered as a maximum intensity projection (MIP) in 3D with a greyscale colormap [Fig. 5(e)].

Fig. 5

(a) The central transverse slice ($30\times 30{\mathrm{mm}}^2$) at voxel size 0.10 mm of the phantom’s vertical monofilaments, (b) a zoomed slice ($3\times 3{\mathrm{mm}}^2$) at voxel size 0.02 mm of a vertical monofilament (1). (c) A longitudinal slice ($30\times 30{\mathrm{mm}}^2$) at voxel size 0.10 mm of the phantom’s horizontal monofilaments, (d) a zoomed slice ($3\times 3{\mathrm{mm}}^2$) at voxel size 0.02 mm of a horizontal monofilament (2). (e) A 3D MIP rendering of the phantom’s 0.10-mm voxel reconstruction with the vertical monofilaments (1), horizontal monofilaments (2), and phantom support rods (3) visible. (f) A heatmap of vertical monofilaments’ average transverse FWHM between $\pm 10\mathrm{mm}$ along the $z$-axis (height). (g) A heatmap of the horizontal monofilaments’, the two visible in a height range of 30 mm, longitudinal FWHM between $\pm 10\mathrm{mm}$ along the radial direction.

Comparing the HR slices of the vertical [Fig. 5(b)] and horizontal [Fig. 5(d)] monofilaments, the transverse slice’s monofilament has an expected circular shape while the longitudinal slice’s monofilament is an oblong, dual-peak ellipse. This oblong shape can be attributed to the mechanical tolerance of the PA detector which was not accounted for in the reconstruction. As a result, the TriTom has a superior PA transverse resolution compared to its PA longitudinal resolution.

The FWHM values determined from the monofilaments’ transverse and longitudinal measurements were plotted as heatmaps [Figs. 5(f) and 5(g)] with dimensions of (radius, height). The transverse spatial resolution of the system ranges between 142 and $215\mu \mathrm{m}$ dependent on the radial distance from the PA detector’s focal point, with the best resolution values towards the image’s origin. While minor, the minimum spatial resolution at (4, 0) is offset from the expected (0, 0). This offset is likely due to uneven light distribution that can be corrected by optimizing the light delivery in the TriTom imaging chamber. The longitudinal spatial resolution has a larger range than the transverse resolution, between 487 and $846\mu \mathrm{m}$. Continuing from the point that the horizontal monofilament reconstructions have two amplitude peaks, the line profile and FWHM measurement method for spatial resolution determination does not account for non-single peak Gaussian fits, and, therefore, will produce high estimates for spatial resolution. The reported transverse and longitudinal spatial resolution of the system is the average of the heatmap data: $173\pm 17\mu \mathrm{m}$ for the transverse plane and $640\pm 120\mu \mathrm{m}$ for the longitudinal dimension.

3.2.

Photoacoustic Sensitivity

The reconstruction of the PA sensitivity phantom, constructed with samples of copper sulfate at varying concentrations, is shown in Fig. 6. All sample concentration microcuvettes are detected above the noise floor in both the non-scattering [Fig. 6(a)] and scattering [Fig. 6(b)] environments.

Fig. 6

Transverse slices of the (a) non-scattering and (b) scattering PA sensitivity phantoms. (c) 3D MIP render of the non-scattering PA sensitivity phantom’s PAT volume. Linear regression fits of the PA sensitivity as a function of sample concentration for the (d) non-scattering and (e) scattering PA sensitivity phantoms.

The linear correlation analysis of PA response and copper sulfate concentration is shown in [Figs. 6(d) and 6(e)] for the non-scattering and scattering phantoms, respectively. The resulting r-squared values for the non-scattering (0.9954) and scattering (0.9916) cases indicate a good linear correlation between PA sensitivity and concentration. This linear relationship demonstrates the ability to reliably quantify sample concentrations from a PA amplitude analysis with the TriTom.

The CNR analysis of the sample microcuvettes is shown in Table 1 for both the non-scattering and scattering phantoms. The two lowest concentrations of ${\mathrm{CuSO}}_4·5{\mathrm{H}}_2\mathrm{O}$ (1.9 and 4.9 mM) have $\mathrm{CNR}<2$ for both the non-scattering and scattering environments. This indicates that the combination of the TriTom’s PA mode hardware and the reconstruction algorithm led to a LLOD for absorption coefficients (${\mu }_a$) between 0.122 and $0.258{\mathrm{cm}}^1$.

Table 1

CNR of CuSO4·5H2O samples’ PA reconstructions with and without a scattering environment.

3.3.

Multispectral Imaging

A representative transverse slice of the multispectral phantom excited at 780 nm is shown in Fig. 7(a). In the sample slice, all three PAI contrast agents are visible: ${\mathrm{CuSO}}_4$ (1), ${\mathrm{NiSO}}_4$ (2), and ICG (3). The sulfates are visualized as more uniform circles than the ICG sample, which has a ring shape because of ICG’s affinity to adsorb to the PTFE wall of the microcuvette.⁴⁵

Fig. 7

(a) A transverse slice of the multispectral phantom PAT volume excited at 780 nm. The three samples are visible: ${\mathrm{CuSO}}_4$ (1), ${\mathrm{NiSO}}_4$ (2), and ICG (3). (b) The PA spectrum of the samples, normalized to the ${\mathrm{CuSO}}_4$ optical spectrum.

The measured PA spectra of the contrast agents normalized to the optical absorption spectrum of the copper sulfate sample are shown in Fig. 7(b). The ${\mathrm{NiSO}}_4$ sample’s PA spectrum matches its optical spectrum in profile and closely in relative amplitude, which is an expected result from a PA stable sulfate.⁴⁶ The ICG sample has a significantly different PA spectrum compared to its optical spectrum, in both profile and amplitude. The difference in absorption peaks of 720 and 780 nm in its optical spectrum is diminished in the PA spectrum, where the two peaks have a smaller difference relative to the ICG spectrum’s highest peak. The PA absorption spectrum is also redshifted by 20 nm with respect to the optical absorption spectrum. Both of these departures of the PA spectrum from the optical spectrum of ICG are known results from others’ previous work on characterizing its aggregation state.⁴⁷ In this study, however, the differences in ICG’s PA and optical spectra can be attributed to the aggregation of the dye’s particles on the walls of the PTFE microcuvettes.⁴⁵

3.4.

Optical Spatial Resolution

The optical spatial resolution phantom is shown in Fig. 4, with the cases of the resolution chart outside of the imaging chamber [Fig. 4(a)], inside the imaging chamber without water [Fig. 4(b)], and inside the imaging chamber filled with water [Fig. 4(c)]. The final measured optical spatial resolution results are in Table 2. In air, the measured spatial resolution of the camera is consistent with the theoretical spatial resolution limit of $39\mu \mathrm{m}$ for the camera’s hardware (Sec. S6 in the Supplementary Material). In water, the spatial resolution of the camera is lower than air due to increased scattering and refraction, with the horizontal spatial resolution in water being significantly lower than the vertical resolution because of the imaging chamber’s cylindrical shape affecting the refraction angle of light in the horizontal directions.

Table 2

Optical spatial resolution limit of the TriTom in air and DI water.

3.5.

Fluorescence Sensitivity

Figure 8 shows images of the FL sensitivity phantom imaged using TriTom’s PA and FL modalities, as well as IVIS Spectrum FLI. All samples of IR-800 are detected in both the TriTom and IVIS Spectrum FL images [Figs. 8(a) and 8(b)]. The absence of the highest sample concentration in the IVIS Spectrum FLI was to capture a better dynamic range of the whole image. A representative transverse slices of the TriTom’s FMT volume, reconstructed from the 360 FLI images, and the TriTom PAT volume are shown in Figs. 8(c) and 8(d), respectively. While all five samples are visible in both images, the two lowest concentration samples are barely detectable above the noise background in the PAT image. The CNR comparison plot [Fig. 8(f)] confirms that the FMT sensitivity of the TriTom is superior to the PAT sensitivity, with all samples of the FMT volume having a higher CNR than its PAT counterpart. This result, combined with the fact that IR-800 (quantum yield = 9%)⁴⁸ favors PA compared to other fluorescent agents, that displays the strength of FLI as a companion modality to PAI through its enhanced molecular sensitivity.

Fig. 8

FL images of the FL sensitivity phantom imaged with the (a) TriTom and (b) IVIS Spectrum. A transverse plane of the FL sensitivity phantom’s (c) FMT volume and (d) PAT volume. (e) 3D MIP render of the FMT volume (green) and PAT volume (blue) superimposed. (f) CNR plot and linear fit of the IR-800 samples imaged with TriTom FLI, IVIS Spectrum FLI, TriTom FMT, and TriTom PAT. Identified are the samples of IR-800: $28.0\mu \mathrm{M}$ (1), $13.7\mu \mathrm{M}$ (2), $6.0\mu \mathrm{M}$ (3), $2.6\mu \mathrm{M}$ (4), $0.9\mu \mathrm{M}$ (5), and control water (6).

Compared to the IVIS Spectrum FLI image, the TriTom’s FL CNR is significantly lower for the imaged IR-800 samples. A circular high background noise pattern is detected in the TriTom’s FL images, which is thought to be caused by light leakage of the excitation pulse through the optical filter from light at large incident angles. This high background noise, which propagates to the FMT volume as well, drastically decreases the reported CNR of the TriTom’s FL modality due to an increased background standard deviation.

3.6.

Imaging in Vivo

To evaluate the in vivo PAT imaging capabilities of the TriTom, PAI scans with three excitation wavelengths were acquired in two mice. Figures 9(a) and 9(b) show 3D volume renders of the 532 and 710 nm scans of the first mouse following administration of GC-AuNPs mixed with ICG. The 532-nm PAT volume shows high-resolution images of superficial vasculature in the abdominal region of the mouse, because Hb and ${\mathrm{HbO}}_2$ are highly absorbent at 532 nm. Additionally, high intensity targets at the liver and injection site can be observed indicating accumulation of the GC-AuNPs which have an optical absorption peak near 532 nm. The 710-nm excitation volume provides deeper tissue information, due to improved light penetration, of Hb in internal anatomical structures such as the iliac blood vessels. A single PAI scan at 1064-nm excitation was acquired in the thoracic region of the third mouse [Fig. 9(c)]. At this wavelength, deep ${\mathrm{HbO}}_2$-rich structures are visible including the heart, aorta, vena cava, and kidney vasculature. Figure 9(d)–9(i) show select 2D slices from the 710 and 1064 nm volumes with labeled anatomy and targets that are not apparent in the 3D renderings. These images demonstrate that the TriTom provides high-resolution PAT images of both superficial and deep tissue structures with and without an exogenous contrast agent.

Fig. 9

PAT volumes and slices of in vivo mouse scans. The top PAT row shows 3D renders of the PAT volumes scanned at wavelengths (a) 532 nm for superficial structures, (b) 710 nm for deep structures with deoxygenated hemoglobin, and (c) 1064 nm for deep structures with oxygenated hemoglobin in the coronal (ventral) view. The center PAT row has three slices of interest from the 710 nm volume: (d) axial, (e) coronal, and (f) sagittal. The bottom PAT row also has three slices from the 1064-nm volume, with the same view orientations of (g) axial, (h) coronal, and (i) sagittal. The FLI row has FL $\mathrm{CNR}>2$ maps overlaid on photographs using a TriTom (j) and IVIS Lumina II (k), the color bar is normalized for both instruments. Key anatomical structures and a fiducial are labeled: liver/liver vessels (1), injection site (2), iliac vessels (3), copper sulfate fiducial (4), spleen (5), intestines (6), heart (7), aorta (8), vena cava (9), kidney/kidney vessels (10), vertebra (11), pancreas (12), lungs (13), and hair follicle (14). Scalebars on the 3D (a, b, c), and 2D (d, e, f, g, h, i, j, and k) images are 10 mm. Photoacoustic image color bars are normalized from 0 to 1 from arbitrary units; the FL map color bar represents the CNR of the fluorescent image.

Comparative FLI scans were captured for a mouse using the TriTom’s FL mode and an IVIS Lumina II. 24-hr after the GC-AuNP mixed with ICG injection, the mouse was imaged in a IVIS Lumina II first, then using a TriTom. CNR maps were calculated to directly compare the images acquired with the two systems. The fluorescent image captured using the TriTom [Fig. 9(j)] detects a high intensity target at the injection site (2), consistent with the reference IVIS Lumina II [Fig. 9(k)] image. The TriTom image depicts the contrast agent diffusing from the injection site towards just above the mouse’s right upper hind limb, whereas the IVIS Lumina II image shows the contrast agent localized to the injection site.