1. Introduction

Optical imaging plays a crucial role in advanced biomedical research and disease diagnosis [1]. Compared with conventional medical imaging modalities such as CT, MRI, and ultrasound sonography, optical imaging has various merits such as non-invasive, non-radioactive, low cost, portability, high sensitivity, high spatiotemporal resolutions, and multi-mode expansibility. The most important thing is that optical imaging can fully utilize the optical properties of various endogenous and exogenous chromosomes in the tissue for high-specific molecular imaging. This way, optical imaging provides valuable functional information on the physiological and pathological activities at the cellular and molecular levels [1,2].

However, due to the high optical scattering in the biological tissue, high-resolution pure optical imaging is limited only to the superficial regions within about 1 mm [3,4]. To realize high-resolution optical imaging in deeper tissues, photoacoustic imaging (PAI) emerges as a valuable alternative imaging method, which converts photons absorbed by the tissue to ultrasound pulses through thermal expansion for signal detection [5,6]. As a result, this hybrid imaging method can take advantage of the rich optical contrast in the biological tissue, thus inheriting the powerful molecular imaging ability from optical imaging. In the meantime, this method benefits from the low scattering of ultrasound imaging. It can provide acoustic-resolution images with a penetration depth of about one to two orders higher than the high-resolution pure optical imaging modalities. This unique feature makes it an ideal tool for in-vivo molecular imaging in deep tissues. As one primary embodiment of PAI, photoacoustic endoscopy (PAE) further overcomes the low penetration depth of the laser, enabling PAI in deep organs that are not accessible from the outside of the body [7–9].

Over the past decade, PAE has achieved significant progress in many aspects, such as imaging speed, spatial resolution, and multimodal detection [8,9]. Unfortunately, there is still no mature PAE system for clinical use. In particular, due to the low penetration depth, the limited wavelength selection, and the dilemmatic probe design in the mainstream PAE techniques, there is little study of spectroscopic PAE in deep tissues, which is the prerequisite of universal high-penetration molecular PAE. Today, the most recent interests in PAE have been directed to building optical-focusing-based systems with high repetition lasers of ∼ tens of kHz [10–19]. However, these lasers are mostly single-wavelength-based and with small pulse energy lower than 0.3 mJ [20–24]. They can obtain the delicate blood vessel networks within 1 mm depth in the superficial tissue layers with high resolution, contrast, and frame rate but don’t well embody the high-penetration advantage of PAI. Although single-element acoustic-resolution-based photoacoustic endoscopy (AR-PAE) has the potential for high-penetration imaging, many of these systems still employ high-repetition-rate lasers for high imaging speed and sacrifice the penetration depth [20–25]. Comparatively, array-based PAE systems can use tunable optical parametric oscillator (OPO) lasers with high pulse energy for high-penetration imaging. However, the manufacturing of transducer arrays is complicated and costly, and their spatial resolutions are limited. More importantly, these array-based PAE probes are generally bulk in size (typically more than 2.5 cm in diameter), so they are not beneficial to finding a suitable animal model to carry out preclinical studies. Other PAE systems and potential PAE imaging methods, including intravascular PAE [26–29] and Fabry-Perot-based forward-looking probes [30,8,9], are also generally limited in penetration depth due to technical limitations such as the small pulse energy, the limited transducer sensitivity, or the imaging principles [31–35].

So far, multispectral photoacoustic computed tomography (PACT) has become a mature technology for label-free monitoring the hemodynamic changes, mapping melanoma, tracking circulating cancer cells, or performing other biomedical functionalities [36]. With the rapid development of various exogenous contrast agents such as small organic dyes, nanoparticles, quantum dots, and genetically encoded proteins, molecular imaging with multispectral PACT has experienced great prosperity. With the deepening of life science and preclinical studies, there is a growing demand to develop high-penetration multispectral AR-PAE techniques for the in-vivo endoscopic molecular imaging of animals’ inner organs. However, this kind of research is rare, and we aim to find a technical solution to address this issue.

In this work, we have built a combined multimodal system of acoustic-resolution-based photoacoustic/ultrasound endoscopy (AR-PA/USE) and wide-field optical imaging for the imaging of the rabbit rectum. We did numerical simulations, field test experiments, phantom experiments, and in-vivo experiments to characterize the developed system and preliminarily evaluate the feasibility of an AR-PAE-based multimodal system for high-penetration molecular PAE imaging in in-vivo animal experiments.

2. Methods

2.1 System setup

Figure 1(a) illustrates the schematic of the combined AR-PA/USE and wide-field optical imaging system. The pulsed excitation light was provided by a 532 nm pumped OPO laser (SpitLight OPO 600 mid-band, Innolas, München, Germany). The OPO laser was operated at a repetition rate of 20 Hz, and its tunable ranged from 680 to 1320 nm. This laser also had a second harmonic generation (SHG) modal, generating light from 400 to 660 nm. The excitation laser beam first passed through a set of cylindrical focus lenses and a diaphragm for spatial filtering and shaping. Then, the free-space laser was directed by dielectric mirrors into the water tank, passed through a 45° CaF₂ optical window, and entered the probe. The length of the probe housing was approximately 15 cm, and an aluminum off-axis parabolic mirror (#37-282, Edmund Optics, NJ) was mounted at the distal end for reflection of the excitation laser and collection of the ultrasound. Due to the high optical scattering of the biological tissue, the incident light provided a diffused illumination. The optical density on the tissue surface was kept below 20 mJ/cm², which was within the American National Standards Institute (ANSI) safety limit [37]. The collected ultrasound was reflected by the 45° optical window and finally received by a 10 MHz planar ultrasonic transducer, as shown in Fig. 1(b). The probe housing was made of a transparent polyethylene (PE) tube, whose outer diameter was 8 mm with 0.5 mm wall thickness. The diameter of the parabolic mirror was 6.25 mm, and the distance from the probe end to the parabolic mirror (or rather the dead zone) was approximately 2 cm.

 

Fig. 1. (a) Schematic illustration of the high-penetration AR-PAE-based multimodal system; (b) Illustration of laser incidence, US detection, and wide-field optical endoscopic imaging schemes of the system, and a photograph of the probe’s distal end; (c) The timing diagram of the data acquisition. The PA/US signals were generated at 20 Hz and recorded successively by the DAQ card; (d) Photograph of a rabbit performing in an in-vivo experiment. CL1, positive cylindrical lens; CL2, negative cylindrical lens.

Download Full Size | PDF

For the photoacoustic (PA) and ultrasound (US) signal acquisition, a digital delay/pulse generator (DG645, Stanford Research Systems, CA, USA) triggered the OPO laser, the pulser/receiver (DPR500, Imaginant, NY), and the data acquisition (DAQ) card (PCI-5124, 100 MS/s, 12-Bit, National Instruments Corporation, TX) simultaneously, controlling their relative delays, as seen in Fig. 1(c). The probe was driven by a stepping motor with a belt-gear set for the circular scan. The PA/US signals were amplified by the DPR500, digitalized, and then stored on a personal computer for future processing. A LabVIEW program controlled the DAQ and scanning control. Notably, although the ultrasonic transducer was far from the parabolic mirror (∼25 cm), the attenuation of the US in water was not significant compared to that in biological tissue. The acoustic attenuation coefficient in soft tissue is 0.54 /(cm*MHz), whereas it is 0.002 /(cm*MHz) in water.

For wide-field optical endoscopic imaging, a white light micro-LED of about 1 mm in size was placed near the parabolic mirror for illumination. The reflected light from the tissue passed through the water tank, a reflection mirror, a telescope modular (with 8 × 25 times magnification), and was finally collected by a collimated digital color camera. The reflection mirror was mounted on a translation stage, which can be removed when performing the AR-PA/USE scan to prevent blocking the incident laser. Figure 1(d) shows a system photograph during an in-vivo experiment on a rabbit.

2.2 Image reconstruction algorithm

Performing the 2D circular scan is equivalent to a virtual point-focused transducer scanning around the axis of the probe. The virtual transducer has the same numerical aperture (NA) as the parabolic mirror. In the conventional sectorial B-mode PAE images, the value for an arbitrary pixel at (r,θ) in polar coordinates is usually given by the following equation:

This study restored the elongated lateral resolution in the out-of-focus regions with an improved back-projection (IBP) algorithm. This method is based on a virtual detection surface concept and has been described elsewhere [38,39]. It can be given as

Here, θ₁, θ₂, …, θ_Nd are the angular positions of the transducer in the scan. The virtual surface is assumed to be composed of Np point detectors, with R_i,k donating the k-th point detector’s position when the transducer is at angular position θ_i. In US imaging, the acoustic velocity was set to be v/2 in Eq. (2). The coefficient A is a weighting factor to indicate whether the transducer can receive the pixel’s signal. This empirical equation can be directly applied in most existing AR-PA/USE systems based on a fixed-focus transducer, effectively improving the depth of focus and significantly enhancing the imaging quality. In this work, the image reconstruction was implemented in 2D in MATLAB and accelerated by GPU through a MEXCUDA function.

2.3 System performance tests

We conducted a field test experiment demonstrating our proposed system's dynamic focusing ability. Herein, a 50-µm-thick tungsten wire was scanned over ten different imaging depths as a point target. The metal wire was subjected to constant illumination in PAI by a multimode fiber with a 3 mm core size. The PA/US results obtained at different depths were put into one image for comparison. To explore the ability of the proposed AR-PA/USE system for extensive penetration imaging, two polyvinyl chlorides (PVC) tubes were placed under chicken breast tissue of different thicknesses and imaged under a wavelength of 805 nm. Both PVC tubes had 1 mm inner and 2 mm outer diameter. One was filled with 0.5 mg/mL indocyanine green (ICG) in 22.5% albumin, and the other was filled with the albumin solution only. PA/US images of the two tubes buried at different depths under the chicken breast tissue were acquired. A small background region was selected at about the same depth to determine the signal-to-noise ratio (SNR) of the two tubes at each depth. The standard deviation (STD) in this region was adopted as a measure of the noise level, and the peak signal of each target was adopted as the signal amplitude.

2.4 In-vivo imaging of the rabbit rectum

To explore the capability of the high-penetration spectroscopic AR-PA/USE in molecular imaging, we performed in-vivo imaging of the rabbit rectum. Male white New Zealand rabbits (weighing approximately 2 kg) were kept in the lab for 48 h for acclimatization and fasted for 24 h before the experiment. The rabbits were first administered rectal enema with normal saline, then fastened on a steel shelf, after which the probe was inserted about 6 cm into the rectum. The rabbits were maintained under gas anesthesia, with a dose of 2% isoflurane and a flow rate of about 1 L/min. In this work, we injected a dose of ICG near the rabbit rectum and compared the 2D and 3D PA/US images at different wavelengths. The wide-field optical endoscopy can monitor the rectum inner surface and ensure the acoustic contact of the probe, which greatly facilitated the experiments. The rabbits were also scanned under T1-weighted MRI (eT1W-aTSE, 352 × 252 pixels, 74 × 74 mm, 24 slices, slice interval 2.56 mm), with an 8 mm PE tube of the same material as the probe housing inserted into the rectum, for identifying the structures and comparing with the acquired PA and US images. We employed 10 to 12 rabbits during this study, and all the rabbits survived after the experiments. All animal experiment procedures were approved by the Department of Laboratory Animals of the Central South University (No. 2020KT-39).

2.5 Deep-learning-based PA signal denoising scheme

Due to the low repetition rate of the laser, the data-averaging times must be minimized. However, under the ANSI optical exposure limits, the PA signal at longer near-infrared (NIR) wavelengths is weak due to the low optical absorption, so multiple acquisitions are generally required for higher SNR. As seen in Fig. 2(a), we found that the noise at these longer NIR wavelengths was mainly from the step motor. These noises were time-varying, strip-like, with high amplitudes but a small duty ratio in the time domain, as indicated with the black circles. Compared with the commonly used direct averaging of the acquired A-lines, this kind of noise can be efficiently removed within a few acquisitions by kicking out the most deviated data point at the same time instant.

 

Fig. 2. The PA data denoising scheme for removing the electrical noise from the step motor. (a) One presentative set of A-line data acquired at 800 nm, including the six original A-lines, the direct averaged A-line, the denoised signal with six A-lines, and the A-line generated by the deep-learning model with only one acquisition; (b) The B-mode images reconstructed with one original A-line, the average A-line, the denoised A-line, and A-line generated by the deep-learning model; (c) The structure of the CNN model for denoising the electrical noise.

Download Full Size | PDF

In this work, we recorded six A-lines at each transducer position. After kicking out the most deviated data point by comparing the STDs, the remaining five data points were averaged and taken as the denoised data. Then one original A-line data and the denoised data (as indicated with the red dotted lines in Fig. 2(a)) were sent to train a denoising model based on the Convolutional Neural Networks (CNN) [40], as seen in Fig. 2(c). This CNN model had seven layers. The mean absolute error (MAE) was taken as the loss function, and the peak signal-to-noise ratio (PSNR) of the output A-lines was employed to evaluate the trained model. The Adam optimizer was applied to train the model at a learning rate of 10⁻⁵. The total number of the A-lines in training was 400, the number of batch samples was 32, and the number of iterations was 200. The training process can be finished in 1 minute. For more details regarding the training of this deep-learning model, please see Supplementary section A.

As seen in Fig. 2(b), the B-mode image reconstructed with the original one-time acquired A-line was quite noisy due to the electrical noise. Direct averaging the six A-lines will not significantly reduce the noise. After denoising with the STD-based data-kicking method, we can see that the electrical noise can almost be removed entirely. We take this denoised data as the correct data to train the deep-learning-based denoising model. We can see that the B-mode image given by the deep-learning-based method with only one acquisition looks almost the same as the image reconstructed with the denoised data with the six acquisitions. Hence, before each 3D scan, we first performed a B-mode scan with six A-lines at each transducer position to train the denoising model. Then in the 3D scan, only one acquisition was taken for one A-line. The noise-free data can be directly generated by the deep-learning-based denoising model, which significantly speeds up the scanning process.

3. Results

3.1 Field test

Simulation results have proved that the IBP algorithm could effectively improve the full-width half maximum (FWHM) and SNR in the off-focal regions (see Supplementary section B). As expected, the IBP algorithm yields superior lateral resolutions to the conventional B-mode method with the field test data. Figures 3(a)–(c) show that the lateral resolution gradually increased from 0.39 mm (at ∼6.6 mm) to 1.03 mm (at ∼20.6 mm) for PAI. A similar trend in the lateral resolution changes was determined for US imaging (Figs. 3(d)–(f)). Results show that the IBP method improves the lateral resolution up to 2.7 times (1.05 to 0.39 mm at the 6.6 mm depth) for PAI and 5.7 times (1.05 to 0.185 mm at the 6.6 mm depth) for US imaging in the out-of-focus regions. The black lines in Figs. 3(c) and 3(f) represent simulated results when the virtual detection surface has similar acoustic parameters to the parabolic mirror. Results show that the measured lateral resolutions are close to the theoretical calculations for PAI. For US imaging, the measured values were slightly larger than the thermotical calculations. This may be due to the influence of the PE probe housing, which reduced the effective size of the parabolic mirror.

 

Fig. 3. Field test experimental results of AR-PA/USE system. (a) and (b) Overlaid PA images of the point target by the B-mode and IBP methods, respectively; (c) Changes in FWHMs with image depth for PAE imaging, calculated with the B-mode method (blue), IBP method (red), and simulated results with the IBP method (black), respectively. Herein, the blue and red lines were obtained from (a) and (b); (d)-(f) Corresponding results for US imaging. In this study, the origins of both x and y axes were all to the rotation center unless noted otherwise. The maximum amplitudes of the PA and US images are normalized for facilitated comparison.

Download Full Size | PDF

3.2 Penetration depth test

Figure 4 shows the penetration test results, where two tubes were buried at 9, 14, and 19 mm beneath the surface of the chicken breast tissue. The tube on the right side was infused with ICG. For facilitated comparison, the reconstructed PA and US images at each depth share the same colormap. Notably, in most US images, both tubes are visualized. In contrast, only the tube with ICG solution can be seen in the corresponding PA images, which indicates the high specificity of PAI if wavelengths of the excitation light are appropriately selected. As seen in Figs. 4(c) and 4(d), the tube with ICG is approximately 10 dB higher in SNR than the other tube in the PA images, but the SNRs of the two tubes are comparative in the US images. The background signals increased significantly with the increase in depth, while the relative signal intensities of the tubes decreased. At a depth of 14 mm from the surface of the chicken breast tissue, numerical analysis shows that the SNRs for the two tubes are 16.2 dB (left) and 24.9 dB (right) in the PA image, and the SNRs are 46.1 dB (left) and 38.1 dB (right) in the US image. At the 19 mm depth, although we see some noise close to the two tubes in the US image, the tube with ICG can still be clearly distinguished in the PA image. This is due to the PA image's high specificity and speck-free nature [41]. These results demonstrate the centimeter-scale high-penetration of this PA/US endoscopic system.

 

Fig. 4. PA and US imaging results of two PVC tubes beneath chicken breast tissue of different thicknesses. (a) Photograph of the two tubes in the experiments; (b) The PA (left column) and the US (right column) images calculated with the IBP method when the tubes were buried at 9 mm (first row), 14 mm (second row), and 19 mm (third row) beneath the surface of the chicken breast tissue; Positions of PVC tubes are indicated with white dashed boxes; (c) SNRs calculated with the conventional B-mode method (dotted) and the IBP method (solid) for the left side tubes (blue, without ICG) and the tubes at the right side (red, with ICG); (d) Corresponding US imaging results.

Download Full Size | PDF

3.3 Cross-sectional in-vivo imaging results

Figure 5(a) shows the schematic of the cross-sectional endoscopic scan. Figure 5(b) shows the T1-weighted MRI image orthogonal to the AR-PA/USE probe, boosting the contrast between different anatomical structures. Figure 5(c) shows the overlaid PA and US images reconstructed with the IBP methods at different wavelengths when a dose of ICG was injected. These were taken at about the same position as the MRI image in Fig. 5(b). It’s noted that the ex-vivo experiments have proved that the IBP methods can significantly improve the endoscopic images. However, in this work, although the IBP method showed improvements for some image features (as seen in Supplementary section C), the overall image quality looks close to the B-mode method, which may be due to the motion effects.

 

Fig. 5. Cross-sectional imaging of the rabbit rectum with the combined AR-PA/USE and wide-field optical multimodal endoscopic system. (a) The schematic of the cross-sectional scan; (b) The T1-weighted MRI image; (c)-(h) The overlaid PA and US images of the rabbit rectum at different wavelengths; (i) The wide-field optical image of the rectum inner surface; (j) The wide-field optical image of a positive resolution test target plate, showing the pattern with a line interval of 45 circle/mm; (k) Wide-field optical image of a one circle/mm pattern (top) and its corresponding intensity profile along the red dotted line (bottom); (l) Edge function (black) and derived PSF (red) of the wide-field optical imaging system, respectively.

Download Full Size | PDF

Here the US images provide continuous boundaries of some pelvic bones and tendons, which aids in a better understanding of the acquired PA images. In contrast, the PA image mainly shows the 2 to 3 mm thick rectum wall due to the existence of the dense blood vessels. Almost no other structures could be seen out of the rectum wall in the PA images, which is consistent with the MRI image. Due to the strong optical absorption in the rectum wall at 532 nm and 670 nm, strong PA signal reflections by the pelvic tendons and bones could be seen in the two PA images, as evidenced by the twice signal flight time indicated by the arrows. In contrast, such PA signal reflections at wavelengths longer than 730 nm are not commonly seen, mainly because these reflections are weak and close to the noise level. Since the ICG injection sites were far from the rectum wall, we can easily recognize the ICG signals spectroscopically, which is rendered in green. The PA signal of the injected ICG can be well distinguished in the images with 730 nm, 800 nm, and 840 nm, but it is absent in the 532 nm, 670 nm, and 900 nm images. The depth of the ICG injection site was measured to be about 5 mm from the rectum wall. These results preliminarily demonstrated the molecular imaging ability for deep-seated targets with AR-PAE.

Wide-field optical imaging, narrow-band imaging (NBI), and fluorescence imaging are commonly used in the clinic for the endoscopic diagnosis of the organ’s inner surface. Figure 5(i) shows the wide-field optical endoscopic image of the rabbit rectum inner surface. Within the 7 mm field of view, we can directly monitor the physiological state of the rectum, such as hyperemia and ischemia, and see the morphology of the blood vessels on the surface. The micro-LED in the probe housing wall can also be imaged. A positive resolution test target plate (R2L2S1P1, Thorlabs Inc.) was employed to obtain the spatial resolution, as seen in Figs. 5(j)-(l). Results showed that the system could distinguish the 45 circle/mm pattern. The resolution was measured with a 1.0 circle/mm pattern. The intensity profile along the red dotted line in Fig. 5(k) was taken as the edge function. By differentiating this curve, the system’s point spread function (PSF) is obtained. The FWHM of this PSF curve was taken as the spatial resolution and measured to be about 14.4 µm. This indicates that it’s feasible to build a combined AR-PA/USE and wide-field optical endoscopic system for cross-scale imaging, which provides a profound understanding of the pathological states of the organs. Also, with the wide-field optical imaging mode, we can efficiently find the regions of interest or suspicion, such as the superficial regions with abnormal angiogenesis, to guide the following AR-PA/USE scan to find deep lesions. Because the AR-PA/USE scan is limited by the low repetition rate of the OPO laser, this will significantly reduce the scanning time and improve the diagnostic efficiency.

3.4 Volumetric PA/US imaging results

Figures 6(a)-(i) show volumetric PA and US images of the rabbit rectum, with a dose of ICG injected. The wavelength was 800 nm, the peak of the ICG absorption spectrum. The volume size was 4 × 4 × 3 cm, consisting of 60 cross-sectional slices in the x-y plane. Because the ICG is isolated from the rectum wall, we rendered the ICG in green for better visualization. The US images show the pelvic structures around the rectum, and the PA images give the distribution of the ICG. Figures 6(j)-(o) show the PA and US images of the cross-sections of A1 and A2, as indicated in Fig. 6(i). The US images give continuous but different structures. We see the ICG was present in the A1 plane but absent in the A2 plane. For better visualization of the injected ICG, a small volume containing the ICG was extracted and shown in Fig. 7, which gives a more detailed mapping of the ICG distribution. These results demonstrate the reliability of our system in 3D imaging.

 

Fig. 6. Volumetric PA and US images rendering the structures of the rabbit rectum and the injected ICG distribution. (a)-(c) are the PA, US, and merged 3D images in the view of the x-y plane; (d)-(f) and (g)-(i) are the results of two other different views; (j), (l), and (n) are the 2D images of the A1 plane, as indicated in (i); (k), (m), and (o) are the corresponding results of the A2 plane. Please see Visualization 1.

Download Full Size | PDF

 

Fig. 7. Zoomed-in views of the volumetric PA and US images of the injected ICG and the surrounding structures. (a)-(c) are the PA, US, and merged 3D images in the view of the x-y plane; (d)-(f) and (g)-(i) are the results of two other different views. The bounding box sizes are 12 × 24 × 18 mm in the x, y, and z directions, respectively.

Download Full Size | PDF

We compared the volumetric images between 532 nm and 800 nm, as seen in Fig. 8. The top row shows the overlaid PA and US images at the wavelength of 532 nm in four different views, and the bottom row is for 800 nm. The PA and US images with 800 nm show similar results as in Fig. 6, except no ICG was injected here. In the 532 nm images, strong PA signal reflection can be seen, which is the same case in Fig. 5(c). These results indicate the 3D spectroscopic imaging ability of our system. The results also imply that molecular imaging at a longer wavelength is preferred due to the low background signal intervenes.

 

Fig. 8. Comparison of the PA and US images of the rabbit rectum at 532 nm (top row) and 800 nm (bottom row). Please see Visualization 2.

Download Full Size | PDF

4. Discussion

Optical molecular imaging is vital in basic biomedical and preclinical research. Compared with pure optical imaging modalities, PAI can uniquely break the optical diffraction limit by converting bulletin and scattered photons to the US for detection, thus making it a powerful tool for high-resolution molecular imaging of deep-seated targets. However, while molecular PACT prevails, the development and application of molecular AR-PAE technology have been frustrating. Current PAE technology fails to integrate the merits of high-penetration, multispectral, compactness, and multimodal integration into a single probe, thus failing to meet the growing needs for high-penetration molecular endoscopic imaging of animals’ inner organs. This problematic situation is due to several critical technical difficulties: (1) due to the limitation of the existing nanosecond pulsed-laser technology, the laser source of PAE can only either be with a high repetition rate or with high single pulse energy; (2) although nonlinear fiber technology [42] and dye-laser technology have been employed to tune the commonly used 532 nm high-repetition laser, the wavelength tuning ability of these methods are still limited; (3) the probe design of AR-PAE is more complicated than pure optical and acoustic endoscopic probes due to the consideration of acoustic coupling and integration of acoustic/optical paths. (4) the PAE probe should integrate EUS, wide-field optical endoscopy, or other optical imaging modalities to comprehensively understand the acquired PA images, making the probe’s design even more difficult. Thus, building a compact AR-PAE probe for practical molecular PAE imaging on small animals is challenging.

To address these issues, we propose building a single-element-based AR-PAE system, which employs a tunable OPO laser as an excitation source and a step motor for scanning control. Here are many reasons for this design. First, compared with the array-based design, although the single-element-based AR-PAE probe may sacrifice the imaging speed, it can achieve a smaller probe size and higher spatial resolutions more easily. Compared with the 8 mm diameter AR-PAE probe in this work, the array-based probes (for example, radial-array-based PAE probes are about 2.5 cm in diameter) are not favorable for most current molecular imaging scenarios on small animals [43,44]. Although a 532 nm laser is typical for hemoglobin imaging with PAE, a NIR laser is more preferred for molecular imaging, which can dramatically mitigate the strong background signal from the blood absorption, improve the sensitivity to the target molecules, and extend the imaging depth. As a result, this system employed a step motor other than a DC motor or torque coil-based method to coordinate with the low repetition rate of the OPO laser. A step motor is needed in many cases of molecular imaging, such as A-line averaging and pump-probe-based data acquisition for background-free imaging. In addition, this system naturally incorporates EUS, which offers valuable anatomical information for target localization. As seen from the 2D and 3D figures, it’s hard to interpret the acquired PA images without US imaging. In contrast, while various emerging optical ultrasound detection methods hold high promise for building high-sensitive small-sized PAE probes, these PAE probes are weak in US transmission for EUS.

Furthermore, we improved the proposed method on the following two aspects to facilitate its practical application. Firstly, we proposed an IBP method to enhance the lateral resolution in the off-focus regions. This way, tight acoustic focusing can be kept throughout the imaging depth with improved SNR. Using a 10 MHz transducer, the lateral resolutions of the system at 19 mm depth reached approximately 1 mm for both PA and US imaging modes. Secondly, the low scanning speed has always been a significant concern for the single-element-based PAE probes, primarily when a laser of tens repetition rate is used. In particular, the step motor’s electrical noise dominates the PA signals at a longer wavelength like 800 nm, which generally requires multiple acquisitions to filter out. We developed a deep-learning-based denoising method to avoid signal averaging. In this way, the scanning speed is ultimately limited by the laser repetition rate, establishing a good foundation for improving the scanning speed. Considering the proposed method's scanning speed is still low (20 seconds for a 400-A-line B-mode image in this work), we suggest integrating the wide-field optical imaging mode into the system for scanning guidance, even though the EUS can partly serve for imaging guidance. The wide-field optical imaging remains the primary endoscopic imaging modality in the clinic due to its high spatial resolution, imaging speed, and contrast. With the cross-scale imaging ability of the system, we can efficiently find the regions of interest or suspicion in the superficial tissues where subtle morphological and spectral abnormalities exist and then perform the following slow AR-PA/USE scan to find deep lesions under the guidance.

As a preliminary study of molecular PAE, our work has many limitations. The biggest one is that the AR-PAE system we built was in a rigid form with a water tank on the table. Hence, it was far from clinical use. Therefore, we are now making a flexible AR-PA/USE probe with a miniature step motor of only 4 mm diameter for the distal-end-based scan. This probe has a small high-frequency transducer in the side wall, and the total diameter is controlled to less than 8 mm. Alternatively, we are also considering building a rigid shaft-scanning-based handheld probe by placing the step rotator to the proximal end, which is expected to achieve a smaller probe size of less than 5 mm. Another disadvantage of our built system is that the wide-field optical imaging was by a remote external CMOS camera. There are many small CMOS sensors commercially available (for example, the OV6946, with 160k pixels, only 5 × 2 mm (length × diameter) in size, and 3 mm blind imaging distance), and it’s straightforward to mount them to the probe’s distal end for forward-looking wide-field optical imaging. We are also considering employing an image guide fiber for both PA excitation and wide-field imaging, which enables the side-view co-registered cross-scale imaging and the integration of fluorescence imaging. We will continue to improve the system by optimizing the focused acoustic field design and developing advanced image reconstruction algorithms. Besides, we only demonstrated our method by imaging an injected ICG in this work. Because the ICG only has a little targeting effect and quickly diffuses in the tissue, the PA images of the injected ICG are sometimes unstable. In future work, we will conduct more molecular imaging studies with advanced targeting strategies and more contrast agent types [45]. As the system’s performance improves, we will monitor the drug delivery and try more functional imaging, such as mapping the temperature [46] and the oxygen partial pressure in the tissue [47].

In brief, we proposed to build a compact single-element-based AR-PAE probe to fulfill the growing need for low-cost, high-penetration molecular imaging in basic biomedical and preclinical studies. We suggest employing a tunable OPO laser for PA excitation for its high-penetration and low background signal intervention. Hence, a step motor should be selected to work with the OPO laser for precise and flexible scanning control. The image reconstruction algorithm should consider restoring the lateral resolution in the off-focal regions to achieve acoustic focusing over the whole imaging depth. We suggest using a deep-learning-based algorithm for denoising when the strip-like electrical noise is prominent. In addition, we recommended that the system integrates EUS and other optical imaging modalities for scanning guidance and improved diagnosis with the cross-scale multimodal imaging results. Although this system was based on a water tank in a rigid form, it’s viable to build a hand-held AR-PA/USE probe with a higher central frequency and smaller size in the subsequent studies. We will also perform more advanced molecular imaging studies using contrast agents with active targeting abilities. With the proposed methods in this work and future efforts, the deep-penetration molecular AR-PAE will potentially play a critical role in biomedical studies.

5. Conclusion

In this work, we have set up an AR-PAE-based multimodal system and preliminarily explored the potential of deep-penetration molecular PAE imaging for the first time. We developed an IBP reconstruction algorithm to extend the system’s depth of focus and create a deep-learning-based denoising algorithm to avoid signal averaging and achieve the shortest scanning time. With a 20 Hz OPO laser, this system can finish a B-mode scan consisting of 400 A-lines in 20 seconds. We demonstrated that with an 8 mm diameter probe, a 10 MHz transducer, and a wavelength of around 800 nm, the proposed system could reach a penetration depth of 14 mm with a lateral resolution of about 0.77/0.65 mm and an SNR of about 25/38 dB for PA/US imaging. Results also show that the injected ICG, the rabbit rectum wall, and the surrounding structures can be distinguished spectroscopically. These results indicated that it’s feasible to build a compact single-element AR-PAE-based multimodal probe for the centimeter-scale-deep molecular imaging of the animal’s inner organs, benefiting the basic biomedical or preclinical studies.