2.1.

System Setup

As shown in Fig. 1(a), the multiparametric PAM system utilizes two nanosecond-pulsed lasers (BX40-2-G, edgewave; wavelength: 532 nm). For each laser, the output beam is attenuated by a neutral-density filter (NDC-50C-2M, Thorlabs) and then reduced to an appropriate diameter by an iris (SM1D12D, Thorlabs). After attenuation and reshaping, one of the beams is coupled into a polarization-maintaining single-mode fiber (PM-SMF; F-SPA, Newport) for Raman scattering-based wavelength conversion,^10,11 which partially red-shifts the 532-nm light. The PM-SMF output is collimated by a collimator (CFC-11X-A, Thorlabs) and purified by a bandpass filter (FB560-10, Thorlabs) to isolate the 558-nm Stokes light. Then, the 558-nm Raman beam and the 532-nm beam from the other optical paths are combined via a dichroic mirror (FF538-FDi01, Semrock) and coupled into the imaging head through a two-meter-long regular single-mode fiber (SMF; P1-460B-FC-2, Thorlabs). To compensate for the fluctuation in the laser intensity, $\sim 5\%$ of the laser energy is picked off by a beam sampler (BSF10-A, Thorlabs) and monitored by a high-speed photodiode (FDS100, Thorlabs).

Fig. 1

(a) Schematic of the multiparametric PAM and (b) contour-scan scheme.

In the imaging head, the dual-wavelength beam coming out of the SMF is collimated by an achromatic doublet (AD; AC127-025-A, Thorlabs), reshaped by an iris (SM05D5, Thorlabs), and then focused by a second AD through a correction lens (CL; LA1207-A, Thorlabs) and the central opening of a customized ring-shaped ultrasonic transducer (RT; inner diameter: 2.2 mm; outer diameter: 4.0 mm; focal length: 6.0 mm; center frequency: 35 MHz; 6-dB bandwidth: 70%) onto the object to be imaged. The imaging head is mounted on a three-axis scanner, which consists of two transverse stages (PLS-85, PI miCos GmbH) for 2-D raster scan and one vertical stage (KR15, THK; motor: 28BYG201, Circuit Specialists) for dynamic adjustment of the focal plane. A homemade water tank is used to immerse the RT and CL. A thin layer of ultrasound gel (Aquasonic CLEAR, Parker Laboratories) is applied between the object to be imaged and the transparent polyethylene membrane at the bottom of the water tank for acoustic coupling.

The two lasers, the three-axis scanner, and a data acquisition card (ATS9350, AlazarTech) are synchronized by a field-programmable gate array (PCIe-7841R, National Instruments). The B-scan rate is set to $1\mathrm{mm}/\mathrm{s}$, during which the two lasers are alternately triggered to generate dual-wavelength A-line pairs. Statistical, spectroscopic, and correlation analysis of the successive A-line pairs allows simultaneous quantification of $C_{\mathrm{Hb}}$, ${\mathrm{sO}}_2$, and CBF at the same spatial scale.¹²

Given the small interval ($10\mu \mathrm{m}$) of adjacent B-scans, their surface contours are expected to be similar. Therefore, the surface contour of current B-scan can be extracted in real time and used to guide the contour scanning of the next B-scan. The schematic diagram of the real-time contour scanning is shown in Fig. 1(b). The acquisition starts with a conventional B-scan (B-scan #1) along the $x$-axis without moving the vertical stage. Thus, the trace of the $x-z$ stage appears as a horizontal line, which is denoted as trace 0. Immediately after B-scan #1, the acquired data (data #1) are processed to extract its surface contour, which is denoted as trace 1. Then, the $y$ stage makes a $10\text{-}\mu \mathrm{m}$ step forward and starts B-scan #2 toward the negative $x$-axis, during which the $z$ stage moves along trace 1 in a reverse fashion. After the completion of B-scan #2, its surface contour (trace 2) is extracted by combining trace 1 and data #2. Similarly, B-scan #3 is acquired by following trace 2 and the surface contour (trace 3) is obtained based on trace 2 and data #3 for the next B-scan. The same procedure is repeated until the completion of the image acquisition.

2.2.

Real-Time Contour Scanning

As shown in Fig. 2, the $z$-coordinate of the system’s focal plane, which determines the desired position of the imaging head during the contour scan, is input into the software in advance. Once the scan starts, the first B-scan is performed without moving the $z$ stage. Immediately after this B-scan, the coordinate of the curved surface is extracted, based on which the $z$-coordinate to guide the contour scanning of the imaging head in the second B-scan is calculated. Note that a three-step procedure is used to extract the $z$-coordinate. First, the maximum signal in each A-line is detected. Then, a curve fitting of these maximum signal points is performed to obtain the surface coordinate of the B-scan. Finally, the $z$-coordinate is calculated by the subtraction of the focal plane coordinate from the surface coordinate.

Fig. 2

Flowchart of the algorithm for the real-time contour map extraction. FC, focal plane coordinate; SC, surface coordinate; ZC, $z$-coordinate; FD, feedback data.

Contour scanning starts at the second B-scan. Before each contour B-scan, the $y$ stage moves one step ($10\mu \mathrm{m}$) forward. Then, the imaging head moves along the $x$ and $z$ directions simultaneously to acquire a high-resolution B-scan by following the surface contour extracted from the previous B-scan. After each contour B-scan, the surface coordinate of the B-scan is extracted based on the method mentioned above, which provides feedback on the performance of the contour scanning: the closer the surface coordinate to the focal plane coordinate, the better the performance. If there is a difference between the two coordinates, the $z$-coordinate for next contour B-scan is updated accordingly and then smoothed.

Since this contour scanning strategy is realized without prior knowledge of the whole surface contour map, it is necessary to perform a safety check before starting each contour B-scan. If the upper position, lower position, or the step size exceeds the preset safety range of the $z$ stage, the software terminates the scan and returns an error message. Otherwise, the system starts the next B-scan until the entire image acquisition is completed.

2.3.

Principles of Multiparametric Measurements

Simultaneous measurement of $C_{\mathrm{Hb}}$, ${\mathrm{sO}}_2$, and CBF at the same spatiotemporal scale is realized by statistical, spectroscopic, and correlation analysis of the same PAM dataset. Statistical analysis of the PAM signal allows quantification of the average number of red blood cells (RBCs) within the detection volume as follows:

Spectroscopic analysis of the dual-wavelength (i.e., 532 and 558 nm) A-line pair allows differentiation of oxy- and deoxyhemoglobin (${\mathrm{HbO}}_2$ and HbR, respectively), from which ${\mathrm{sO}}_2$ can be quantified as follows:¹³

Correlation analysis of successive A-lines allows quantification of CBF. Theoretically, the time course of the correlation coefficient follows a Gaussian decay, where the decay constant is linearly proportional to the blood flow speed.¹⁴ Thus, CBF can be derived by fitting the experimentally measured decorrelation curve with the theoretical model.

2.4.

Animal Preparation

Male CD-1 mice (8 weeks old, Charles River Laboratories) were used for the in vivo experiments. Under general anesthesia, the hair in the mouse head was removed and a surgical incision was made in the scalp to expose the skull. Before the PAM experiment, the skull was thinned by a surgical hand drill following removal of the fascia. The skull thinning was performed by following an established protocol to alleviate potential inflammation and avoid possible damage to the brain.^15,16 Once the desired skull thickness ($\sim 100\mu \mathrm{m}$) was achieved, the thinned-skull window was cleaned and dried. The ultrasound gel was then applied between the skull window and the ultrasonically and optically transparent polyethylene membrane at the bottom of the water tank. The temperatures of the water tank and the animal body were both maintained at 37°C. All experimental procedures were carried out in conformity with the animal protocol approved by the Animal Care and Use Committee at the University of Virginia.

3.1.

Phantom Experiments

An optically absorbing phantom with uneven surface was imaged by the PAM system to demonstrate the feasibility of the real-time contour-scan method. A piece of electrical tape (black) attached to the surface of a plastic ball, the phantom was imaged using the 2-D raster scan and the 3-D contour scan, respectively. Then, the surface contours of the tape in both images were extracted (Fig. 3), in which the $x$-axis represents the B-scan direction and “0” in the $z$-axis represents the position of the focal plane (negative and positive values denote that the tape is above and below the focal plane, respectively). As shown in Fig. 3(a), the surface contour of the tape in the raster-scan image is uneven, with most of the region being out-of-focus. In contrast, the surface contour captured by the contour scan is flat, with only a $5\text{-}\mu \mathrm{m}$ standard deviation in depth [Fig. 3(b)].

Fig. 3

Surface contours of the phantom extracted from the PAM images acquired with (a) raster scan and (b) contour scan.

3.2.

In Vivo Experiments

The side-by-side comparison of raster scan and contour scan was further performed in vivo. Specifically, the entire mouse cortex was first imaged using raster scan to show the influence of the out-of-focus issue on the measurements of microvascular structure and function (i.e., $C_{\mathrm{Hb}}$, ${\mathrm{sO}}_2$, and CBF). Then, the same ROI was reimaged using real-time contour scan to demonstrate the improvement. As shown in Fig. 4(a), the 3-D surface contour of the mouse cortex ($6\mathrm{mm}\times 5\mathrm{mm}$) extracted from the experimental data acquired by raster-scan PAM shows a dome shape with a depth range of $\sim 900\mu \mathrm{m}$. In contrast, the surface contour of the same cortical ROI imaged by contour-scan PAM is relatively flat, with a $26\text{-}\mu \mathrm{m}$ standard deviation in depth. Based on the surface contour map acquired using raster scan, the distance of each vessel away from the focal plane was quantified [Fig. 4(b)], which shows that both the central and peripheral regions of the mouse cortex were out of focus.

Fig. 4

Side-by-side comparison of raster scan and contour scan in vivo. (a) Surface contours of the entire mouse cortex extracted from the PAM images acquired using raster scan and contour scan, respectively. (b) Pseudocolor map showing the distance of individual vessels from the focal plane in the raster-scan PAM image. (c) Multiparametric PAM images acquired using raster scan. (d) Multiparametric PAM images acquired using contour scan. Scale bar in (b) is $500\mu \mathrm{m}$ and also applies to (c) and (d). Arrows in (c) and (d) indicate differences between the quantitative measurements by the two types of scanning approaches.

Side-by-side comparison of the PAM images acquired using raster scan and contour scan [Figs. 4(c) and 4(d), respectively] clearly shows the adverse influence of the out-of-focus issue on the multiparametric hemodynamic measurements. As indicated by the white arrows, out-of-focus resulted in increased values in the $C_{\mathrm{Hb}}$ and ${\mathrm{sO}}_2$ measurements. The situation in the flow measurement is more complicated, where out-of-focus could lead to either increased or decreased measurement values depending on the distance of the vessel from the focal plane (indicated by the yellow and pink arrows, respectively). Quantitative analysis is provided in Sec. 4 to rationalize these observations.

4.1.

Influence of the Out-of-Focus Issue on Diameter Measurement

To quantitatively analyze the influence of the out-of-focus issue on the measurement of vessel diameter, side-by-side comparison of the maximum amplitude projection images acquired using raster scan [Fig. 5(a)] and contour scan [Fig. 5(b)] at 532 nm (an isosbestic point of ${\mathrm{HbO}}_2$ and HbR) was performed. Zooming into the periphery of the mouse cortex shows much blurred microvessels in the raster-scan image compared to their counterparts in the contour-scan image [insets in Figs. 5(a) and 5(b)]. This is further evidenced by comparing the cross-sectional profiles of individual vessels in a pair of representative B-scans acquired by the two scanning approaches [Fig. 5(c)]. Clearly, multiple microvascular cross-sections [arrows in Fig. 5(c)] are not resolved due to the compromised spatial resolution in the out-of-focus region. With the aid of segmentation-based single-vessel analysis,³ vessel diameter maps of the two structural images were also generated. As highlighted by the arrow pairs, the microvessels in the central and peripheral out-of-focus regions of the raster-scan image [Fig. 5(d)] show larger diameters than their counterparts do in the contour-scan image [Fig. 5(e)]. Furthermore, distribution analysis shows that there are fewer microvessels with less than $30\text{-}\mu \mathrm{m}$ diameters in the raster-scan image than in the contour-scan image due to the out-of-focus issue [Fig. 5(f)]. In conclusion, the structural analysis shows that the real-time contour-scanning PAM enables cortex-wide microvascular imaging with high spatial resolution.

Fig. 5

Influence of the out-of-focus issue on the diameter measurement. (a), (b) Maximum amplitude projection images of the mouse cortex acquired using raster scan and contour scan, respectively. Insets show the blurred microvasculature due to the out-of-focus issue. Scale bar in (a) is $500\mu \mathrm{m}$ and also applies to (b), (d), and (e). (c) Cross-sectional profiles of the microvessels in the B-scans highlighted by the dashed lines in the insets of (a) and (b). The red arrowheads highlight the vessels that are not resolved in the raster-scan image, and the purple arrowheads highlight vessels that are blurred in the raster-scan image. (d), (e) Vessel diameter maps extracted from the raster-scan and contour-scan images, respectively. (f) Vessel diameter distribution in the raster-scan and contour-scan images. The $x$-axis represents the diameter value and the $y$-axis denotes the ratio of the number of vessel cross-sections with a particular diameter to the total number of vessel cross-sections in the image.

4.2.

Influence of the Out-of-Focus Issue on Hemodynamic Measurements

As shown in Figs. 4(c) and 4(d), the out-of-focus issue leads to increased values in the $C_{\mathrm{Hb}}$ and ${\mathrm{sO}}_2$ measurements. To reveal the underlying mechanism, the $C_{\mathrm{Hb}}$ and ${\mathrm{sO}}_2$ values measured by raster-scan and contour-scan PAM were compared at the single-vessel level. To this end, a four-step procedure was performed: (1) coregister the multiparametric PAM images acquired using raster [Fig. 4(c)] and contour scan [Fig. 4(d)], (2) apply an amplitude threshold to remove the nonvascular background, (3) obtain ratiometric images by taking the ratio of the raster-scan measurements to the contour-scan measurements, and (4) correlate the ratiometric images and the out-of-focus map [Fig. 4(a)] to find out how the $C_{\mathrm{Hb}}$ and ${\mathrm{sO}}_2$ measurements are affected by the degradation in spatial resolution.

According to Eq. (2), $C_{\mathrm{Hb}}$ is codetermined by $E^2(A_{\mathrm{PAM}})$, $\mathrm{Var}(N_{\mathrm{PAM}})$, and $\mathrm{Var}(A_{\mathrm{PAM}})$. $\mathrm{Var}(N_{\mathrm{PAM}})$ is the electronic thermal noise of the PAM system, which remains the same under both scanning conditions and can be ignored in this analysis. When out of focus, the relaxed point spread function of the optical–acoustic dual foci reduce local optical fluence and compromise acoustic impulse response. Moreover, the averaging effect of the broadened point spread function leads to reduced fluctuation of the PAM signal. Indeed, as shown experimentally in Fig. 6(a), the increased imaging voxel size due to out-of-focus results in reduced $E^2(A_{\mathrm{PAM}})$ and $\mathrm{Var}(A_{\mathrm{PAM}})$. The farther it is away from the focal plane, the larger the decrease. Moreover, $\mathrm{Var}(A_{\mathrm{PAM}})$ is more sensitive to the out-of-focus issue than $E^2(A_{\mathrm{PAM}})$, resulting in increased $C_{\mathrm{Hb}}$. Therefore, the $C_{\mathrm{Hb}}$ value in the raster-scan image is larger than that in the contour-scan image when the vessel is out-of-focus [Fig. 6(b)]. According to Eq. (3), ${\mathrm{sO}}_2$ is codetermined by $[{\mathrm{HbO}}_2]$ and [HbR]. When out of focus, the photoacoustic amplitudes at both excitation wavelengths (denoted as PA532 and PA558) are reduced. Our analysis shows that PA558 is more sensitive to the out-of-focus issue than PA532 [Fig. 6(c)], resulting in increased ${\mathrm{sO}}_2$. Therefore, the ${\mathrm{sO}}_2$ value in the raster-scan image is higher than that in the contour-scan image when the vessel is out of focus [Fig. 6(d)].

Fig. 6

Influence of the out-of-focus issue on (a) $E^2(A_{\mathrm{PAM}})$ and $\mathrm{Var}(A_{\mathrm{PAM}})$, (b) $C_{\mathrm{Hb}}$, (c) the photoacoustic amplitude (PA532 and PA558, respectively, denote the amplitudes of the photoacoustic signals acquired at 532 and 558 nm), and (d) ${\mathrm{sO}}_2$. The ratios are defined as the measurements by raster-scan PAM over those by contour-scan PAM.

The side-by-side comparison in Figs. 4(c) and 4(d) shows that out-of-focus also affects the CBF measurement. Difference from the cases of $C_{\mathrm{Hb}}$ and ${\mathrm{sO}}_2$, however, the measured CBF in the out-of-focus region can be either higher or lower than the true value depending on the distance of the vessel from the focal plane. To unravel the underlying mechanism, we carried out an experiment in a blood vessel-mimicking phantom. As shown in Fig. 7(a), defibrinated bovine blood (Quad Five, 910) was driven by a syringe pump (Pump System Inc., NE-300) to flow through a plastic tube (United States Plastic Corporation, 56514) at $5\mathrm{mm}/\mathrm{s}$. The tube was tilted from the focal plane of the PAM system at an angle of $\sim 9\mathrm{deg}$.

Fig. 7

Phantom experiment showing the influence of the out-of-focus issue on the blood flow measurement. (a) Experimental configuration, (b) measurements of blood flow by raster-scan and contour-scan PAM, respectively. (c), (d) Out-of-focus-induced SNR degradation and flow measurement error in the raster-scan experiment, which are in contrast to the consistent SNR and flow speed measured by the contour scan in (e) and (f).

The blood flow within the tube was imaged by both raster-scan and contour-scan PAM for side-by-side comparison [Fig. 7(b)]. In the raster-scan image, the measured flow speed varies with the depth location of the tube along the $y$-axis. In contrast, the contour-scan image shows a relatively constant flow speed of $\sim 5\mathrm{mm}/\mathrm{s}$ [Fig. 7(f)], which is consistent with the preset speed. Besides the difference in the measured flow speed, the thickness of the tube in the raster-scan image does not appear uniform, with the two ends thinner than the center, due to the out-of-focus-induced SNR decrease [Fig. 7(c)], which is in contrast to the consistent SNR in the contour-scan image [Fig. 7(e)]. Furthermore, quantitative analysis [Fig. 7(d)] shows that the measured flow speed decreases when the imaged region is slightly out of focus and then starts to increase when it is further out of focus.

This phantom experiment confirms the out-of-focus-induced biphasic changes in the CBF measurement observed in vivo (Fig. 4) and sheds light on the underlying mechanism. The CBF measurement in multiparametric PAM is based on correlation analysis of the sequentially acquired A-lines, in which the decay constant of the correlation reveals RBC’s dwell time in the detection volume.^12,17 When the imaging area is slightly out of focus, the SNR remains relatively high but the detection volume enlarges, which results in a longer dwell time and thus a lower flow speed. In contrast, when the imaging area is way out of focus, degradation in the SNR becomes non-negligible. As a result, the rapidly fluctuating noise starts to play a confounding role in the correlation analysis, leading to faster decorrelation of the A-lines and higher flow speed.