1. Introduction

Photoacoustic imaging (PAI) has attracted extensive investigation in recent years owing to the unique advantages of rich optical contrast and deep acoustic penetration [1–3]. One major implementation of PAI is optical-resolution photoacoustic microscopy (OR-PAM), which provides micron or even submicron resolution by focusing excitation light on an optical diffraction-limited spot. Combined with pixelwise raster scanning, high-resolution morphological imaging of finer biological structures such as capillaries [4,5], cells [6,7], and cell nuclei [8,9] using OR-PAM is possible. However, restricted by the scanning technique, video-rate (e.g., ≥30 Hz) imaging in OR-PAM remains challenging. For example, to achieve a 64 × 64-pixel image at 30 Hz, the scanning rate must reach 1920 Hz for fast axis scanning (B-scan) and 30 Hz for slow axis scanning. The three types of scanning mechanisms currently used in OR-PAM, that is, mechanical scanning, optical scanning, and hybrid scanning, are all shy to meet this requirement. Mechanical scanning utilizes a stepping motor to carry the imaging head to any site on the imaging object, that is, to achieve an unlimited field of view. However, this method has a low scanning rate. A B-scan rate of 1 Hz over 1 mm is typically achieved using this method [10,11]. Although a faster voice-coil motor was employed by Wang et al. to boost the mechanical scanning speed, the B-scan rate was still limited to 40 Hz over a range of 1 mm, far below the video-rate speed illustrated above [12]. Optical scanning based on two-axis scanners such as MEMS devices or galvanometer mirrors was later implemented in OR-PAM [13,14]. Although optical scanning is widely adopted in conventional optical imaging and may meet the criteria of video-rate imaging, the severely deteriorated field-of-view and image quality of this method owing to ultralow sensitivity at oblique scanning angles makes it unsuitable for PAI. Consequently, water-immersed optical scanning has been proposed to improve imaging sensitivity [15–17]. The scanner was immersed in water to achieve a coaxial design of the optical and acoustic beams. Thus, high sensitivity, including oblique scanning angles, was maintained at all times. However, the speed of the scanner in water is significantly reduced because of the resistance of water is higher than that of air, making it unsuitable for video-rate imaging. A hybrid scanning method combining fast axis optical scanning and slow axis mechanical scanning has also been proposed to achieve high imaging sensitivity and high imaging speed [18–20]. A cylindrically focused ultrasound transducer can be applied to this design to obtain high speed and sensitivity [21,22]. However, owing to the low speed of mechanical scanning achieved via a motorized translation stage or rotation stage along one axis, the speed of hybrid imaging is still far below the video rate. Different from the aforementioned scanning mechanisms, several nonscanning approaches have been explored in recent years. Yang et al. [23] proposed motionless volumetric spatially invariant resolution photoacoustic microscopy (SIR-PAM), which utilizes a digital micromirror device to generate propagation-invariant sinusoidal fringes and applies two-dimensional Fourier-spectrum acquisition to realize scanning-free imaging with an extended depth of field (DOF). Although the DOF is satisfactory, the imaging speed is still limited to 21 s to acquire a volumetric image. Li et al. [24] presented photoacoustic topography through an ergodic relay (PATER) for high-throughput PA imaging at a frame rate of 2 kHz. Encoding the PA waves through ergodicity, PATER uses broad light illumination with only a single-element unfocused ultrasonic transducer to acquire an image without scanning. However, a time-consuming calibration (320 × 240 pixels, 100-time averaging, 60-min) is required before imaging each object, and the overexposure to laser dosage may affect the dynamics of high absorption samples.

To overcome these challenges, an alternative method of improving speed is to reduce the number of measurements using compressed sensing (CS). CS is a sparse sampling strategy that allows the reconstruction of an image from undersampled data. To date, CS has been widely used in many imaging modalities, including PAI, to reduce the number of measurements and improve the speed [25,26]. However, CS leads to significant image deterioration: to maintain decent image quality, the rate of undersampling in CS must be limited. Hence, the improvement in the imaging speed with CS is also limited. Moreover, the CS technique has a large computational overhead owing to the iterative image reconstruction. The number of iterations must be balanced with the computational burden, and the recovered image quality is often far from optimal. Hence, it is still challenging to apply CS to achieve high-resolution video-rate imaging in PAI.

Herein, we propose single-pixel nonscanning photoacoustic microscopy (SPN-PAM) to achieve video-rate high-resolution imaging. SPN-PAM applies frequency-domain single-pixel imaging (SPI), which is a technique that originates from computational ghost imaging [27–29]. A single-pixel detector was used to capture images, and a spatial light modulator is employed to modulate the illuminated light patterns. A single-pixel detector collects signals at each modulation pattern to obtain the encoded spatial information of the object. By decoding the correlation between light modulation and the signal intensity of the detector, an object image can be reconstructed [30,31]. Compared to the conventional raster-scanning imaging method, SPI enables a nonscanning mechanism, effectively eliminating the limitation due to scanning. Moreover, the reconstruction process of SPI is based on inverse transformation rather than an iterative process in the CS technique [32,33]. Hence, the computational overhead is significantly reduced.

The established SPN-PAM enabled sparse sampling in the frequency domain to achieve high-resolution video-rate PAI for the first time. Compared with conventional sparse sampling in the spatial domain, that is, the CS technique mentioned above, sparse sampling in the frequency domain can be performed at a much higher undersampling rate because it selectively sparses only redundant information in the image. Consequently, the speed can be improved more than 20 times while maintaining the image quality at a decent level. In both in vitro cell imaging and in vivo mouse vascular imaging, we experimentally validated the high-resolution photoacoustic microscopic image reconstruction of SPN-PAM at ultralow sampling rates. The practicability of video-rate imaging was verified by performing hemodynamic and blood reperfusion imaging in mice. To the best of our knowledge, this is the first time that high-resolution PAI has been achieved at a real-time video-rate (≥30 Hz). The results provide new insights into boosting imaging speed in high-resolution PAI and further promote PA applications in dynamic biological imaging.

2. Materials and methods

2.1 Experimental set-up and principle of SPN-PAM

A schematic of the custom-built single-pixel nonscanning photoacoustic microscopy (SPN-PAM) imaging system is shown in Fig. 1(a). Briefly, a 532 nm wavelength Nd: YAG nanosecond pulsed laser (GKNQL-532, Beijing Guoke Laser Corp.) was used as the excitation light source. The emitted laser beam was further enlarged using a beam expander (BE, GCO-2501, Daheng Optics). The enlarged laser beam was reflected off a reflection mirror (RM), which was then transferred to a commercial digital micromirror device (DMD, V-7001, Vialux) with an incident angle of 24°. The active mirror array area is 14.0 × 10.5 mm² with 1024 × 768 pixels. In this study, we used a square area (768 × 768 pixels) in the middle of the DMD to modulate the illumination light. The maximum binary pattern switching rate of the DMD is 22.7 kHz, which was triggered in the slave mode to maintain consistency with the laser repetition rate at 20 kHz. The projected structured illumination patterns were relayed using two 4f configurations and spatially filtered by a pinhole (PH). The 4f configurations consisted of convex lenses (CL) (CL1, f₁ = 250 mm, CL2, f₂ = 50 mm, CL3, f₃ = 200 mm), and an objective lens (NA = 0.1, PLN4X, Olympus). We set up our system on a vibration isolation optical table to provide maximum stability of the laser beam. During the experiment, the region of interest was adjusted to the imaging plane of the 4f configuration to excite photoacoustic signals, which were then detected by a focused ultrasonic transducer (UT, 25 MHz central frequency, -6 dB bandwidth of 70%, V324-SM, Olympus). An ultrasonic pulser-receiver (US P/R, 5077PR, Olympus) was used to amplify the PA signals by 59 dB. A data acquisition (DAQ) card (CSE1422, GaGe) was used to digitize the PA signals at a sampling rate of 200 MS/s. All data acquisition and system control programs were developed using LabVIEW software (2016, National Instruments).

 

Fig. 1. (a) Schematic of the single-pixel nonscanning photoacoustic microscopy (SPN-PAM) imaging system. BE, beam expander; RM, reflection mirror; DMD, digital micromirror device; CL, convex lens; PH, pinhole; UT, ultrasonic transducer; US P/R, ultrasonic pulser-receiver. (b) Principle of SPN-PAM. IFT, inverse Fourier transform.

Download Full Size | PDF

The principle of SPN-PAM imaging based on Fourier orthogonal basis patterns is shown in Fig. 1(b). Because Fourier transform can decompose any signal into sinusoids, 2D images can also be decomposed into a series of sinusoidal patterns. Thus, to reconstruct the objective image, the PA signals are first sampled at definite spatial frequencies to form a Fourier spectrum of the imaging object, and a 2D inverse Fourier transform (IFT) is then applied to obtain the image. Here, to make full use of the fast binary switching rate of the DMD, binary amplitude modulation approach was utilized to generate Fourier basis patterns. Briefly, we first generated the grayscale Fourier basis patterns with corresponding spatial frequencies. Then the patterns were resized to match the micromirror array of DMD. Finally, the error diffusion dithering method was used to binarize the patterns. The Fourier coefficient corresponding to each spatial frequency was obtained through the three-step phase-shifting method, and the Fourier spectrum component was obtained by measuring the three-phase values with an equal interval from 0 to 2π (i.e., 0, 2π/3, and 4π/3). The detailed Fourier basis pattern generation and phase-shifting methods are further described in Supplement 1.

2.2 System characterizations

To validate the feasibility of the SPN-PAM system, a positive USAF 1951 resolution test target (GCG-020601, Daheng Optics) was used to characterize the field of view (FOV) and lateral resolution. Photographs of a typical USAF 1951 resolution target are shown in Fig. 2(a). The FOV is determined by the optical illumination spot size at the focal plane. However, owing to the limited beam width of the focused transducer, there is a tradeoff between the FOV and sensitivity of imaging. In this study, the FOV was set to 473 × 473 µm², which is the theoretical value of the optical focus for the 4f configurations. Element 5 in group 2 (line width and spacing between lines are both 78.7 µm, and line length is 393.7 µm) was imaged to validate the imaging FOV, as shown in Fig. 2(b). The effective optical focus was confirmed to be consistent with theoretical estimation. The lateral resolution of the system is determined by the numerical aperture (NA) of the objective and the number of pixels within the imaging FOV. A higher NA of the objective and more pixels within the FOV yielded a better lateral resolution. Therefore, when the NA and pixel count are fixed, there is a tradeoff between the FOV and the lateral resolution. A 0.1 NA objective with a theoretical lateral resolution of ∼2.7 µm was used, and a 128 × 128-pixel FOV was imaged during the resolution test. As shown in Fig. 2(c), the two smallest groups (groups 6 and 7) on the target were well imaged using this system. We plotted the edge intensity profile along the blue dotted line in Fig. 2(c) as the edge spread function (ESF), as shown in Fig. 2(f). The line spread function (LSF) was derived from the first-order derivative of the ESF, and the full width at half maximum (FWHM) of the LSF was measured as the lateral resolution. As shown in Fig. 2(f), the measured lateral resolution was slightly sacrificed to ∼3.73 µm with the expansion of FOV. The cross-sectional profile of element 6 (with a line width of 4.39 µm) in group 6 [Fig. 2(c)] is plotted along the green dotted line. As shown in Fig. 2(e), the three 4.39 µm wide lines were clearly resolved against the background. For comparison, Fig. 2(d) shows the same groups (groups 6 and 7) imaged using a conventional OR-PAM system. As shown in Fig. 2(g), the corresponding lateral resolution was 2.72 µm, perfectly matching the theoretical lateral resolution of a 0.1 NA objective. Compared to conventional OR-PAM, the slight deterioration of lateral resolution in SPN-PAM is mainly caused by the limited number of pixels (128 × 128 pixels) within the enlarged FOV, namely, the pixel size (3.7 µm) is larger than the theoretical lateral resolution (2.7 µm) determined by the NA of objective. However, this resolution is sufficient to evaluate microvasculature.

 

Fig. 2. Validation of the SPN-PAM system using USAF resolution test target. (a) The photograph of a positive USAF 1951 resolution test target. Scale bar, 2.5 mm. (b) FOV measurement by imaging element 5 in group 2. Scale bar, 60 µm. (c) SPN-PAM imaging of groups 6 and 7 on the target. Scale bar, 30 µm. (d) Conventional OR-PAM imaging of the groups corresponding to (c). Scale bar, 30 µm. (e) The cross-sectional profile of element 6 in group 6 of the resolution target, as indicated by the green dotted line in (c). (f) Lateral resolution test result based on the edge intensity profile along the blue dotted line in (c). (g) Lateral resolution test result based on the edge intensity profile along the blue dotted line in (d).

Download Full Size | PDF

Volumetric imaging is feasible for PAI due to its time-resolved acoustic detection. In this manner, the SPN-PAM also enables depth-resolved three-dimensional imaging. To quantify the depth imaging capability of the SPN-PAM system, a 2-mm thickness agarose phantom with carbon fibers (∼7 µm diameter) embedded in it was imaged. Figure 3(a) shows a depth-encoded maximum amplitude projection (MAP) image of carbon fibers acquired by SPN-PAM, and different depth layers are presented in different colors. To show the depth-resolved capability of SPN-PAM explicitly, the profile of a Hilbert-transformed PA signal generated by a single laser pulse is plotted in Fig. 3(b). It can be seen that the imaging depth from the surface layer to the bottom layer reaches about 769 µm. The MAP images of carbon fibers in Layer #1 (from 0 to 244 µm) and Layer #2 (from 525 to 769 µm) are shown in Fig. 3(c) and (d), respectively. Differrent depth of layers can be reconstructed clearly. Nevertheless, it is worth noting that the depth-resolved capability of the current system is still restricted by the depth of field of the optical beam, and meanwhile, the axial resolution can be further enhanced by using a wider bandwidth ultrasound transducer.

 

Fig. 3. Quantification of imaging depth of the SPN-PAM system. (a) Depth-encoded carbon fibers image acquired by SPN-PAM. Scale bar, 60 µm. (b) Depth-resolved PA signal generated by a single laser pulse. (c) MAP image of carbon fibers in Layer 1. Scale bar, 60 µm. (d) MAP image of carbon fibers in Layer 2. Scale bar, 60 µm.

Download Full Size | PDF

2.3 Sample preparation

2.3.1 Cell sample preparation

The murine monocyte–macrophage cell line (RAW 264.7) was obtained from the American Type Culture Collection (ATCC). Cell culture slides (07-2104, Biologix) with removable chambers were used to culture RAW 264.7 cells. The RAW 264.7 cells were cultured in DMEM medium supplemented with 10% (v/v) fetal bovine serum and 1% (v/v) antibiotics (penicillin–streptomycin) and were incubated at 37 °C under a 5% CO2 atmosphere. Diluent Indian ink solution was added to each chamber of the cell culture slide and was incubated for 2 h at 37 °C in a 5% CO2 atmosphere. Because of the phagocytic function of macrophages, India ink particles were phagocytosed by RAW 264.7 cells. The chambers were gently washed with a stock solution composed of phosphate-buffered saline (PBS) to remove any residual ink. Finally, the detachable chamber was removed, and a glass slide with attached cells was reserved for imaging.

2.3.2 Experimental animal preparation

BALB/c nude mice (4–6 weeks old, weighing 18–20 g, Beijing Vital River Laboratory Animal Technology Co. Ltd.) were used to demonstrate the in vivo imaging capability of the SPN-PAM. During the experiments, the mice were anesthetized using a mixture of isoflurane and oxygen (3% for induction, 1%–1.5% for maintenance) at a flow rate of 1 L/min. A heating pad, monitored by a temperature maintenance device (69027, RWD Life Science), was employed to maintain the body temperature of the mice at 37 °C. All animal experiments were performed in accordance with the protocols approved by the Guangdong Provincial Animal Care and Use Committee and followed the guidelines of the Animal Experimentation Ethics Committee of the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences.

3. Results and discussion

3.1 SPN-PAM for undersampled imaging

One of the main advantages of SPN-PAM compared to conventional PAM is that sparse sampling can be utilized in the frequency domain (i.e., Fourier transform domain) by acquiring only the highest-coefficient components, that is, components containing core information of the image. We experimentally explored the possibility of using SPN-PAM to increase imaging speed with reduced measurements. During SPN-PAM imaging, a sequence of Fourier basis patterns was loaded into the DMD to modulate the pulsed laser, and the group 5 area on the resolution target was successively imaged. Because the power of the Fourier spectrum is concentrated at the center frequency, a circular sampling path starting at the center was adopted to acquire the Fourier spectrum from lower to higher frequencies. The full Fourier spectrum of the image is shown in Fig. 4(f). We optimized the undersampling strategy by selectively discarding the nonessential high-frequency components (i.e. periphery frequencies in Fig. 4(f)) of the image in the Fourier transform domain. Undersampling was conducted with sampling rates (SRs) varying from 1% to 100%. Representative images of a resolution target with sampling rates of 100%, 51.85%, 20.83%, 10.83%, and 4.83% were reconstructed and are shown in Fig. 4(a)–(e), respectively, and the corresponding Fourier spectra are shown in Fig. 4(f)–(j). The results show that high-quality images can be obtained using SPN-PAM with undersampling. Similar-quality images can be acquired at sampling rates of 20.83% and 100%. At a sampling rate of 10.83%, the image can be recovered with slight ringing artifacts, as shown in Fig. 4(d). At a sampling rate as low as 4.83%, even though the object is blurry [Fig. 4(e)], it can still be resolved.

 

Fig. 4. Experimental results of SPN-PAM with undersampling. (a)–(e) The reconstructed images with different sampling rates of 100%, 51.85%, 20.83%, 10.83%, and 4.83%, respectively. (f)–(j) Fourier spectrum of different sampling rates corresponding to (a)–(e). Scale bar, 60 µm.

Download Full Size | PDF

Three different metrics were used to quantitatively assess the quality of the reconstructed images: the signal-to-noise ratio (SNR), root mean squared error (RMSE), and structural similarity index (SSIM). The detailed quantitative calculations are described in Supplement 1. The SNR versus the sampling rate of SPN-PAM is shown in Fig. 5(a). With an increase in the sampling rate, the SNR rose rapidly to a maximum value and then dropped slightly as the sampling rate further increased. This drop at a high sampling rate was mainly due to the introduction of periodic high-frequency noise at large sampling rates. In Fig. 5(b), it can be seen that SPN-PAM has a relatively low RMSE overall, which decreases with increasing sampling rate. The SSIM increases markedly to 0.76 when the sampling rate is less than 20%, as shown in Fig. 5(c). Notably, fully sampled images were used as reference images to derive RMSE and SSIM quantitative metrics. These results show that SPN-PAM enables high-quality imaging even at low sampling rates.

 

Fig. 5. Assessing the image quality of SPN-PAM quantitatively with different sampling rates. (a) SNR. (b) RMSE. (c) SSIM.

Download Full Size | PDF

To clarify the advantages of the Fourier basis patterns adopted in SPN-PAM, the results were also compared against other orthonormal basis patterns constructed using a Hadamard matrix (see Supplement 1, Section 2). Table S1 in Supplement 1 provides a more detailed quantitative comparison. When the sampling rate was lower than 10%, the Fourier basis outperformed the Hadamard basis in terms of reconstruction quality. In addition, to fully sample a 64 × 64-pixel image, the number of measurements for the Fourier spectrum via three-step phase-shifting is 3 × 64 × 64 = 12288, which can be further reduced to about 1.5 × 64 × 64 = 6144 considering the conjugate symmetry of the Fourier spectrum. The number of measurements required for the Hadamard spectrum to fully sample the image is 2 × 64 × 64 = 8192. Hence, compared to Hadamard, the proposed Fourier basis reduces measurements by 25%, which is of great significance in boosting imaging speed.

The experimental results presented above provide a solid foundation for optimizing the SPN-PAM undersampling strategy. For a 64 × 64-pixel image, the entire data acquisition process required approximately 0.015 s at a sampling rate of 4.83%. The imaging speed was increased more than 20 fold, which meets the ≥30 Hz video-rate imaging demand. Notably, the sampling rate can be further reduced for object images that contain more low-frequency components. The undersampling of Fourier coefficients would also reduce the bandwidth of the imaging system to some extent. To improve the sampling efficiency under limited measurements, i.e., to minimize the system bandwidth loss as much as possible, our future work may explore more optimized sparse sampling methods such as self-adaptive strategy for better resolution and imaging quality.

The nonscanning scheme of SPN-PAM avoids motion artifacts compared to conventional mechanical raster scanning PAM. The multiplexing illumination of SPN-PAM enables a superior SNR at the same laser fluence because each pixel reconstruction is composed of measurements from multiple sinusoidal structured patterns. Therefore, the SNR is enhanced owing to multiple illumination averaging. With these additional merits, high-speed SPN-PAM under lower laser fluence is feasible. In addition, with the advent of DMD with a higher binary modulation rate, the imaging speed of SPN-PAM can be continuously increased in the future.

3.2 In vitro cell imaging

We further validated the feasibility of SPN-PAM by imaging RAW 264.7 murine macrophages in vitro. Macrophages imaged using a conventional white-light microscope before using SPN-PAM are shown in Fig. 6(a). A full-view image of macrophages using SPN-PAM is shown in Fig. 6(b), and a close-up image of the region indicated by the green box in (b) is shown in Fig. 6(c). Figure 6(c) clearly shows the morphological structure of macrophages and is consistent with the image observed by a white-light microscope, especially the detailed donut-shaped structure formed by the existence of nuclei indicated by the white arrows. Undersampling was also performed in this experiment, and satisfactory imaging results were obtained, as shown in Fig. 6(d)–(f). Thus, SPN-PAM provides new insights into high-throughput measurements of individual cells. When SPN-PAM is integrated with a multiwavelength pulsed laser, such as a laser source based on stimulated Raman scattering [34,35], functional and molecular information at the single-cell level can also be obtained. The boost in imaging speed owing to undersampling would also facilitate the dynamic visualization of single-cell metabolism. With respect to the phototoxicity limitations in biological tissues and cells, the undersampling of SPN-PAM also enables high-quality imaging with less laser exposure, thus lowering toxicity.

 

Fig. 6. Imaging results of cells in vitro. (a) Conventional white light imaging of macrophages. Scale bar, 40 µm. (b) SPN-PAM imaging of macrophages at the full sampling rate. Scale bar, 60 µm. (c) Close-up image of the region indicated by the green box in (b). Scale bar, 20 µm. (d)–(f) SPN-PAM imaging of macrophages with sampling rates of 51.85%, 20.83%, and 10.83%, respectively.

Download Full Size | PDF

3.3 In vivo animal imaging

We demonstrated the in vivo dynamic imaging performance of SPN-PAM on mouse ears. Throughout the experiment, the mouse ears were spread flat on an adjustable transparent glass holder. Ultrasonic gel was evenly applied between the mouse ear and a water tank made of a transparent membrane for ultrasound coupling. Figure 7 shows two distinct vascular network regions in the ear. SNRs under different sampling rates were analyzed to assess the in vivo compressed imaging quality. Evidently, good quality images were obtained at ultralow sampling rates, such as, 4.83%. Because the undersampling scheme is similar to a low-pass filter, when the sampling rate increases, smaller vascular structures containing more high-frequency components are reconstructed more clearly. Meanwhile, periodic high-frequency noises under higher sampling rates would slightly deteriorate the SNR. For in vivo imaging, the optical scattering of biological tissues should not be neglected. The structured illumination fringes were found to be more susceptible to scattering than focused excitation light in conventional OR-PAM [23]. The resolution and imaging quality of SPN-PAM would inevitably be deteriorated in living biological tissues. To alleviate this effect, tissue optical clearing [36] is an alternative solution to improve the penetration depth and the image quality.

 

Fig. 7. SPN-PAM imaging of in vivo mouse ear vascular networks under distinct sampling rates. (a)–(e) The reconstructed vascular networks on mouse ear region one with sampling rates of 100%, 51.85%, 20.83%, 10.83%, and 4.83%, respectively. Scale bar, 60 µm. (f)–(j) The reconstructed vascular networks on mouse ear region two with sampling rates of 100%, 51.85%, 20.83%, 10.83%, and 4.83%, respectively. Scale bar, 60 µm.

Download Full Size | PDF

After validating the in vivo compressed imaging capability, we utilized SPN-PAM to visualize the hemodynamics of the mouse ear. To monitor the rapid physiological changes in vivo, we fixed the sampling rate at 10.83% to achieve 30 Hz video-rate imaging. Visualization 1 in the Supplementary Materials displays the real-time capture of the blood flow. As shown in the video, the flow of red blood cells in smaller branch vessels was prominent. In addition, to observe the blood perfusion, we temporarily blocked the blood flow in the mouse ear and conducted imaging immediately after the blockage was released. The dynamic changes in blood flow and visualization of blood reperfusion can be clearly observed from Visualization 2 in the Supplementary Materials. Eight representative images obtained during reperfusion are shown in Fig. 8. The laser fluence on the tissue surface was ∼1.2 mJ/cm², which is a far lower ANSI safety limit of 20 mJ/cm². With high temporal and spatial resolutions, SPN-PAM can be a powerful tool for investigating hemodynamic coupling by detecting transient changes in blood volume or blood flow at the microvasculature scale. The technique can also shed light on other physiological studies, such as neurovascular coupling, given that neural activities can be inferred via estimation of blood supply.

 

Fig. 8. Representative mouse ear vascular images of blood perfusion at a frame rate of 30 Hz. Scale bar, 50 µm.

Download Full Size | PDF

4. Conclusion

Herein, we propose a video-rate high-resolution single-pixel nonscanning photoacoustic microscopy (SPN-PAM) imaging system that utilizes Fourier basis patterns for spatial light modulation. The FOV of our system was 473 × 473 µm², and the lateral resolution achieved was 3.73 µm. At a sampling rate of 10.83%, the in vivo imaging speed reached 30 Hz, which is the highest speed reported thus far among all OR-PAM systems. SPN-PAM takes advantage of the sparse property of the image in the frequency domain, that is, the Fourier transform domain, thereby enabling decent reconstruction image quality while achieving high imaging speed.

Compared to the point-by-point raster scanning scheme of conventional PAM, SPN-PAM provides a new solution to overcome the speed limitations of mechanical and optical scanning. This technique reduces motion artifacts and nonlinear distortions induced by scanners. In addition, the multiplexing illumination of the technique enables an enhanced SNR owing to the multiplexed averaging principle. Compared to the nonscanning SIR-PAM approach, the focus of our work is to achieve video-rate imaging with sparse sampling, which provides a new solution to overcome the speed limitations in conventional OR-PAM. Moreover, we utilized binary amplitude modulation approach to generate patterns, which is more convenient for operation because the elaborate alignment of the spatial filter position and aperture size is not needed. Hence different orthonormal basis patterns can be implemented in one system. In our work, two orthonormal basis patterns constructed using Fourier and Hadamard matrix have been demonstrated in the same system to investigate the compressed sampling capability. Compared to the ergodic relay based nonscanning approach, the SPN-PAM method is calibration-free. Further, the lateral resolution of SPN-PAM is better. The lateral resolution of PATER is 110 µm, approximately equal to 1/2 of the acoustic wavelength in fused silica at the central frequency of the ultrasonic transducer under high-speed wide-field mode. Moreover, in terms of applications, owing to sparse sampling of SPN-PAM, the laser dosage can be reduced significantly, thus lowering phototoxicity for high absorption biological samples. However, limited by the field of view and transmission-mode experimental set-up, the SPN-PAM is not suitable for imaging of wide-field thick targets. With a wide-field transparent ultrasound transducer, this problem can be resolved.

With the above-mentioned merits, high-quality video-rate imaging under low laser fluence is conceivable. In this study, we successfully realized high-quality image reconstruction at ultralow sampling rates for both in vitro cell and in vivo mouse imaging. The video capture of hemodynamics in mouse ears showed that SPN-PAM is a powerful tool for obtaining deeper insight into high-speed photoacoustic microscopy imaging. In future studies, we plan to further improve our method using multispectral imaging to expand its applications to high-throughput measurements of individual cell metabolism, rapid biological functional observations, and molecular imaging. Moreover, by integrating a transparent ultrasound transducer, reflection-mode imaging can be implemented in our system, thus facilitating thick targets imaging, e.g., whole mouse brain imaging. Meanwhile, with wide-field highly sensitive ultrasound transducer, the FOV of our current system can be further enlarged. Nevertheless, the tradeoff between the FOV and the lateral resolution needs to be noticed. With these improvements, the video-rate single-pixel nonscanning PAM would be an alternative powerful tool for biomedical imaging as well as clinical translation research in the future.