Listening to tissues with new light: recent technological advances in photoacoustic imaging

Abstract

Photoacoustic tomography (PAT), or optoacoustic tomography, has achieved remarkable progress in the past decade, benefiting from the joint developments in optics, acoustics, chemistry, computing and mathematics. Unlike pure optical or ultrasound imaging, PAT can provide unique optical absorption contrast as well as widely scalable spatial resolution, penetration depth and imaging speed. Moreover, PAT has inherent sensitivity to tissue’s functional, molecular, and metabolic state. With these merits, PAT has been applied in a wide range of life science disciplines, and has enabled biomedical research unattainable by other imaging methods. This Review article aims at introducing state-of-the-art PAT technologies and their representative applications. The focus is on recent technological breakthroughs in structural, functional, molecular PAT, including super-resolution imaging, real-time small-animal whole-body imaging, and high-sensitivity functional/molecular imaging. We also discuss the remaining challenges in PAT and envisioned opportunities.

1. Introduction

Photoacoustic (PA) imaging (or optoacoustic imaging) exploits a physical phenomenon that uses light (i.e. “photo-”) to generate sound (i.e. “-acoustic”) through thermal expansion induced by absorption of the photon energy. In fact, the same effect can be induced by any electromagnetic waves, including visible and near-infrared light, radio-frequency waves, x-ray photons, and even γ rays. Even though the original discovery of the photophonic phenomenon by Alexander Graham Bell dates back to 1880 (4, 5), it was not until the early 1990s that the effect was initially applied in biomedical imaging, later known as photoacoustic tomography (PAT, or optoacoustic tomography), in which visible and near-infrared laser light is most commonly used as the excitation source. Physically combining the optical excitation and ultrasonic detection, PAT offers rich optical contrast at depths far beyond those achieved by conventional high-resolution optical microscopy. Enabled by the new developments in laser, ultrasound and computer science technologies in the early 2000s, PAT has advanced quickly from proof-of-concept lab technologies to popular biomedical research tools.

Two major categories of PAT technologies exist: photoacoustic microscopy (PAM) and photoacoustic computed tomography (PACT). Generally, PAM is often used for high-resolution imaging within a depth of several millimeters, while PACT is mostly utilized for volumetric imaging with a larger penetration depth beyond one centimeter at the expense of inferior spatial resolution. One of the most recent advances in PAM is super-resolution imaging, enabled by non-linear mechanisms, such as two-photon absorption (8–10), absorption saturation (11–15), thermal relaxation (12, 16, 17), photothermal bleaching (18, 19) and reversible photoswitching (6, 20, 21). Meanwhile, PACT has made significant progress in attaining three-dimensional (3D) (22–24), real-time (25, 26), and whole-body small-animal imaging (27, 28) performance. The rapid technical evolution of PAT has enabled a wide range of applications in both pre-clinical research and clinical translation. For example, high-resolution PAM is exploited in single cell imaging, including flow oximetry (29), flow cytometry (30, 31) and tissue engineering (32, 33). PACT with superior penetration depths is extensively applied in functional small-animal studies such as mapping neural-activity mapping (34), measuring blood oxygenation (35, 36) and monitoring cardiac functions (37). Moreover, PAT has been increasingly used in oncology, particularly for cancer screening (38–41).

Recent Review articles have focused on different aspects of PAT. For example, Li et al. have presented a report on PAT of blood oxygenation (13). Yao et al. have provided an overview of the recent process in PA molecular imaging (42). Other examples include Reviews on single cell PAM (43), PA blood flow imaging (44), PAT of genetically encoded reporters (45), molecular contrast agents for PAT (46, 47), PA spectroscopy (48, 49), and ultrasonic transducers in PAT (50). In this concise Review, we focus on recent technical advances in PAT and their corresponding applications. We discuss the outstanding technical challenges in PAT and envision future technological and application-related breakthroughs, such as single molecule detection and deep-human-brain functional imaging. Last but not least, we also provide perspectives on promising clinical translations in PAT, such as early detection of cancer and cardiovascular disease and surgical guidance.

2. Overview of PAT

In this section, we introduce the technical basics of PAT, including the imaging principles as well as the multi-scale and multi-contrast imaging capabilities. Readers are referred to more comprehensive Review articles on each of the sub-sections (14, 47, 51).

2.1. Basic Principles

The imaging process of PAT includes two main steps: optical excitation and ultrasonic detection. In the first step, short laser pulses are delivered to illuminate biological tissues. As photons propagate through tissue, biomolecules absorb some of the optical energy that is partially or fully converted into thermal energy, resulting in thermal expansion and local pressure rise. The pressure rise propagates in the tissue as ultrasonic waves. This process is depicted in Fig. 1. The optical wavelength, pulse energy, pulse width, and pulse repetition rate are optimized to maximize the PA signal generation. In the second step, the laser-induced ultrasonic waves are detected at the tissue surface by a single ultrasonic transducer or an array. The received ultrasonic signals are then used to reconstruct the original optical energy deposition inside the tissue using different inversion schemes. Therefore, the rich optical absorption contrast of biological tissues is responsible for the excellent functional, molecular, and metabolic imaging capacities of PAT. On the other hand, the generated ultrasonic waves can propagate in the tissue with much less attenuation and scattering than the light photons, thus PAT can provide acoustically-determined spatial resolution at depths far beyond that achieved by optical microscopy (52). The hybrid combination of light and ultrasound enables PAT to achieve highly scalable imaging performance with consistent and diverse contrast mechanism.

Figure 1.

Basic imaging scheme of PAT. Firstly, laser light is delivered to excite the target (optical absorber) in the tissue. The absorbed optical energy is partially or completely converted into heat, inducing a transient thermal expansion and local pressure rise. The pressure alterations propagate in tissues in the form of ultrasound waves, which are detected by ultrasonic detector(s) placed outside the tissue. After amplification and digitization, the detected signals are used to reconstruct the final PAT image reflecting optical absorption contrast inside tissues.

PAT can be grouped into two major implementations based on the image acquisition strategy. PAM typically relies on a direct image acquisition by raster-scanning focused ultrasound transducers in order to form images (14). PAM can be further classified into optical-resolution PAM (OR-PAM) and acoustic-resolution PAM (AR-PAM). While OR-PAM uses tight optical focusing to provide the optically-determined lateral resolution within the optical diffusion limit (~1 mm in soft tissue), AR-PAM uses weakly focused light and provides acoustically-determined spatial resolution at depths up to several millimeters (14). PACT typically employs wide-field light illumination and multi-channel unfocused ultrasound detection, and depends on inverse image reconstruction methods, such as delay-and-sum, back-projection, time-reversal or model-based algorithms, to form the final images (14, 53). PACT typically operates at lower ultrasound frequencies (1–10 MHz) than PAM (20–75 MHz), providing greater imaging depths (> 1 cm) but lower spatial resolutions.

2.2. Multi-Scale PAT

PAT is capable of high-resolution imaging at different scales, with the trade-off between the spatial resolution and penetration depth. Across the scales, PAT maintains a high depth-to-resolution-ratio (DRR) of more than 200, qualifying it as a high-resolution imaging modality. For example, with sub-micrometer resolution and a penetration depth of < 1 mm, OR-PAM can image single cells and sub-cellular structures such as red blood cells (31, 54), osteoblast (55), mitochondria (2), and neutrophils (56). With sub-millimeter resolution and a penetration depth of >3 mm, AR-PAM can visualize subcutaneous blood vessels in the human palm (57) and deep-brain vasculature in small animals (58). Further, PACT can achieve a penetration of several centimeters for human and whole-body small animal imaging (34, 39, 59, 60). Overall, PAT’s high scalability can enable biomedical studies of the same biological process on cells, small animals, and even humans with consistent contrast mechanisms (61).

2.3. Multi-Contrast PAT

PAT’s multi-contrast imaging capability originates from the physical process making the amplitude of the PA signal to be proportional to the product of optical absorption coefficient of the imaged target and the local optical fluence. In other words, PAT has a 100% sensitivity to variations in the optical absorption contrast. Different molecules typically have unique absorption spectra. Therefore, various endogenous and exogenous labels can be used in PAT to target specific functional and molecular processes. Endogenous labels have several major advantages: (1) no toxicity, (2) unperturbed biology, (3) abundance, and (4) avoidance of regulatory approval. Endogenous molecules such as hemoglobin, melanin, lipid, water, and collagen are imaged in PAT for estimation of blood oxygenation (62–64), imaging-guided photothermal ablation of melanoma (65), cancer screening (41), and intravascular imaging of coronary atherosclerosis (66, 67). On the other hand, similar to fluorescence imaging, the utilization of rich exogenous contrast agents has tremendously broadened PAT’s applications that require high molecular sensitivity and specificity, particularly in cancer diagnosis and treatment. Exogenous contrast agents have two advantages over endogenous ones (68, 69): (1) optimization for greater detection sensitivity and (2) conjugation with targeting molecules (e.g., antibodies) to selectively bind to receptors for molecular imaging. For example, nanoparticles have been extensively used for monitoring photothermal therapy and guiding lymph node biopsy (70–78). Dyed microbubbles have been used for dual-modality ultrasound imaging and PAT in nanoparticle delivery and tumor imaging (79–81). All in all, PAT’s multi-contrast imaging capability can provide optical-absorption-based functional and molecular information, which is, in fact, the major advantage of PAT over pure ultrasound imaging.

3. Recent Advances in PAT technologies

Here, we summarize the most exciting advances in PAT technologies that have overcome the limits in structural, functional, molecular imaging, and other system attributes, including super-resolution PAT, real-time small-animal whole-body PAT, remote-sensing PAT, high-sensitivity functional/molecular PAT, and machine-learning enhanced PAT.

3.1. Structural PAT

In this section, we discuss the new techniques that aim at improving the capability of PAT to image structures with high resolution, high speed and large penetration depth. These techniques include recent breakthroughs in super-resolution and real-time imaging of biological structures at both microscopic and macroscopic levels.

3.1.1. Super-Resolution PAT (SR-PAT)

In conventional OR-PAM, the spatial resolutions are generally limited by the optical diffraction in the lateral direction and the acoustic detection bandwidth in the axial direction. One approach to overcome the optical diffraction in OR-PAM is to explore nonlinear signal generation mechanisms, such as the two-photon absorption, which has long been used in two-photon microscopy deeper penetration and reduced photo-bleaching (82, 83). In contrast to two-photon microscopy, the major technical challenge in two-photon PAM is the separation of nonlinear two-photon signals from their linear single-photon counterparts. One solution by Yamaoka et al. uses time-gated detection and frequency filtering (10, 84–87), while another study by Lee et al. employs a loss modulation technique. Two-photon PAM has provided a lateral and axial resolution of 0.51 μm and 2.41 μm, respectively (Fig. 2(a)) (1). The major advantage of two-photon PAM over single-photon PAM is the much improved axial resolution due to the optical sectioning. However, one clear drawback of two-photon absorption is the significantly reduced PA signal strength due to the low two-photon absorption coefficient, resulting in diminished imaging depth.

Figure 2.

Photoacoustic microscopy with sub-diffraction resolutions. (a) Melanin distribution imaged by two-photon PAM (2PA) (top) and one-photon PAM (1PA) (bottom) (1). (b) Mitochondria imaged by PA nanoscopy based on optical absorption saturation (2). (c) Red blood cells imaged by Grüneisen-relaxation-based PAM (GR-PAM) (top) at three different depths (3). (d) Reversibly-switchable PAM (RS-PAM) (right) has superior resolution on bacteria (left) (6). (e) Nanoparticles imaged by photobleaching-based PAM (PI-PAM) (right) (7).

Other nonlinear mechanisms have also been explored to achieve sub-diffraction resolutions in PAM, including transient optical absorption (88, 89), optical absorption saturation (2, 11), and thermal relaxation (7, 90). All these methods share the same goal–nonlinear PA signal generation, and can be applied to any samples. The transient optical absorption based method modulates the number of ground-state electrons by using a strong excitation laser beam. Similarly, the optical absorption saturation-based method exploits the dependence of ground-state electrons on the excitation optical fluence (Fig. 2(b)) (2, 11). The thermal relaxation-based method utilizes the dependence of Grüneisen parameter on the optical fluence (7). The Grüneisen parameter reflects the conversion efficiency from the optical to acoustic energy in the PA signal generation. The thermal-relaxation-based PAM has improved the lateral resolution from 0.65 μm to 0.41 μm and the axial resolution from 45 μm to 2.3 μm (Fig. 2(c)) (7, 90). Photothermal bleaching and reversible photoswitching are two nonlinear mechanisms that permanently or temporally modulate the sample’s optical absorption coefficient. Both methods rely on the dependence of the absorption coefficient on the local optical fluence. The photothermal bleaching based method measures the difference in PA signals before and after bleaching to improve the spatial resolution (Fig. 2(d)) (7, 91, 92). In the reversible photoswitching based method, light-absorbing proteins such as BphP1, can be switched between two absorbing states (6), (92). The differential PA signals between the two states have nonlinear dependence on the optical fluence, which may lead to a sub-diffraction spatial resolution in the 140 nm range (Fig. 2(d))(6, 93). The photobleaching method, however, may cause damage to the imaged sample whereas photoswitching employs switchable contrast agents, thus restricting broad applicability.

To achieve super-resolution in AR-PAM and PACT, in which the spatial resolution is largely determined by the acoustic diffraction limit, researchers have so far exploited temporal fluctuations, wavefront shaping, and sparse absorber localization technique. For example, recent studies have reported resolution enhancement of 1.4 fold by using the speckle-illumination-induced signal fluctuations, and resolution enhancement of $\sqrt{n}$ fold (with n-th order of cumulant images) by using flow-induced signal fluctuations (94, 95). Using wavefront shaping, a six-fold resolution improvement has been shown in a scattering medium (96). By extracting and superimposing central positions of flowing targets over multiple image frames, localization-based PACT is able to achieve sub-diffraction spatial resolution and reduce the limited-view artifacts (97). All of the above super-resolution methods have great potential to enable deeper photoacoustic imaging with superior resolution over conventional PAT, and allow functional and molecular imaging at depths of several centimeters, far beyond the optical diffusion limit. However, superior spatial resolution usually comes at a cost of diminished temporal resolution, where multiple frames have been combined in order to form a single super-resolution image. As a consequence, these methods may not be capable of capturing rapid biological dynamics.

3.1.2. Dynamic PACT

PAT is well suited for deep-tissue dynamic imaging on small animal models, taking advantage of the near-infrared optical excitation, low-frequency ultrasound detection, and parallel data acquisition. The recent advances in PACT technologies include 3D real-time multi-spectral imaging, whole-body small-animal imaging. It was initially challenging for PACT to achieve real-time image rendering, due to a large number of simultaneously recorded time-resolved transducer channels and the heavy computational workload. Benefiting from the recent progress in graphics processing unit (GPU) based computation, the image reconstruction in PACT has been significantly accelerated to attain high-speed volumetric imaging in humans (98, 99) and mice (100). Moreover, compressed-sensing (CS) based image reconstruction has been applied in real-time PACT with spatial under-sampling (101–106). Ozbek et al. have developed a high-speed PACT system that combines CS with GPU-enhanced image reconstruction, attaining a 3D frame rate of 1.6 kHz (107), more than 10 times faster than the previously reported PACT imaging speed. A principal component analysis based method has also been developed to accelerate PACT’s image reconstruction (108, 109). The downside of the CS-based methods is the reduced performance when imaging densely distributed targets.

Important progress has also been recently achieved in whole-body small-animal imaging with PACT. Traditionally, PACT systems using linear arrays and other acquisition geometries with limited tomographic coverage and/or multiplexed data acquisition have suffered from limited-view artifacts, spatial under-sampling, and breathing motion artifacts, which collectively result in low image quality (27, 110–113). A recent work by Li et al. has reported single-impulse panoramic PACT (SIP-PACT), as shown in Fig. 3(a), which uses a 512-channel ring-shaped ultrasound transducer array coupled with a 512-channel data acquisition system. There is no data multiplexing needed, so each laser pulse can provide a complete 2D image of the mouse’s cross section. SIP-PACT has largely addressed the aforementioned issues and achieved a 2D frame rate of 50 Hz, in-plane spatial resolution of 125 μm, and much-reduced limited-view and breathing-motion artifacts (114, 115). With superior imaging performance, SIP-PACT has been used for tracking circulating tumor cells and quantifying whole-body oxygenation (116). Most recently, Merčep et al. have reported on a new implementation of a full-ring array geometry for high-quality hybrid transmission-reflection optoacoustic ultrasound (TROPUS) computed tomography of small animals (117). Yet, SIP-PACT needs translational scanning along the elevational direction for volumetric imaging. Moreover, it suffers from poor elevational resolution, resulting in volumetric images exhibiting highly anisotropic resolution. Another important recent work in small-animal whole-body PACT is the spiral volumetric optoacoustic tomography (SVOT), developed by the Razansky group (118, 119). The system schematics is illustrated in Fig. 3(b). Using a semi-spherical ultrasound transducer array, SVOT is capable of real-time multi-spectral imaging within a volume of 1 cm³ (119). At each wavelength, SVOT can provide a maximum 3D frame rate of 100 Hz, with an isotropic in-plane resolution of 250 μm (119). SVOT’s multi-spectral imaging capability was demonstrated by real-time mapping of exogenous contrast agent distribution (119). Both SIP-PACT and SVOT have shown great promise for biomedical applications at the whole-body scale, such as monitoring cancer progress, tracking drug delivery, and evaluating treatment responses.

Figure 3.

Small-animal whole-body PACT. (a) Setup of the SIP-PACT system. A dual-wavelength illumination is employed for blood oxygenation estimation. BC: beam combiner; CL: conical lens; MBS: magnetic base scanner; OC: optical condenser; USTA: ultrasonic transducer array and WT: water tank (115). Each wavelength is illuminated at a 10-Hz pulse repetition rate. Adapted with permission from (115). (b) Setup of the SVOT system. The semi-spherical transducer array is rotated and translated along a spiral trajectory around the mouse to obtain consecutive volumetric images (left). Merging the individual volumes (middle) provides the final 3D whole-body image (right). Adapted with permissions from (119).

Besides the conventional acoustic detection using piezoelectric transducers, novel all-optical PAT systems employing optical detection of acoustic waves have become an important research topic (120). All-optical acoustic detectors can provide a better detection sensitivity per unit area and a wider detection bandwidth than piezoelectric transducers, both of which are highly desired in PAT to improve the imaging depth and spatial resolution. For example, Fabry-Perot sensors have been developed in Paul Beard’s group, in which the PA pressure waves are detected by optically reading the modulation of the sensor’s optical thickness (121). The Fabry-Perot-based PACT shows superior image quality over traditional PACT with piezoelectric transducers, mainly due to the broader detection bandwidth extending into the low-frequency range and smaller effective detector size as defined by the size of the probing beam (121). More recent advances include using multiple probing beams to reduce image acquisition time (122–124) and plano-concave sensor to improve the detection sensitivity (125). We will discuss the all-optical PAT with remote sensing in Section 3.4.1.

In addition to the imaging speed, recent advances in PACT also include compact laser systems (23), improved SNR and elevational resolution (126, 127), low-cost light sources (128–132), and high-speed multispectral excitation (26, 112, 133, 134). Together, these advances have greatly enhanced PACT’s imaging performance, especially in pre-clinical applications, and enabled a variety of biomedical studies by simultaneously providing large field of view, extended imaging depth, high imaging speed, functional and molecular sensitivity.

3.2. Functional PAT

Fundamentally, PAT’s functional imaging capability originates from its unique sensitivity to optical absorption contrast. PAT has been previously used for measuring blood oxygenation, blood flow, temperature, and oxygen metabolism, as well as monitoring the tumor microenvironment (e.g., pH), tracking targeted molecular agents, and enumerating the circulating tumor cells. Nevertheless, one major challenge in functional PAT is the quantification accuracy of the physiological parameters in deep tissues beyond the optical diffusion limit, mainly due to the unknown tissue optical properties and the wavelength-dependent optical attenuation (i.e., spectral coloring). To address these issues, a number of methods have been developed to improve the absolute quantification and detection sensitivity. A number of representative examples are described below to highlight this highly active research topic.

Firstly, for blood flow measurement in deep tissue, a family of thermally-encoded methods have been developed. In general, the thermally-encoded flow methods take advantage of PAT’s sensitivity to the blood temperature. By optically or acoustically increasing the local blood volume’s temperature, PAT can track the motions of the heated blood volume and quantify the flow speed. In particular, ultrasound-encoded photoacoustic flowgraphy (UE-PAF) has demonstrated superior performance over Doppler-based PA flow measurement (17). UE-PAF, as shown in Fig. 4(a), uses a single-element focused ultrasonic transducer to periodically elevate the blood temperature within the acoustic focus, while the PACT system using a ring-shaped transducer array acquires the images of the heated blood and tracks its motion (17, 135, 136). Since focused ultrasound waves can penetrate several centimeters into the tissue and induce rapid temperature rise, UE-PAF has successfully measured blood flow in deep mouse brain. Using hemoglobin as the endogenous contrast, the thermally-encoded methods are capable of measuring slow blood flow on the level of 0.1 mm/s (17, 137). More recent works using lasers for optical heating and PA imaging have simplified the system configuration and improved the spatial resolution, at the cost of imaging depth (137, 138).

Figure 4.

Novel PAT methods for quantifying blood flow and oxygenation in deep tissues. (a) Imaging principle of UE-PAF. The ring-shaped ultrasound array tracks the displacement of HIFU-heated blood volume. Adapted with permission from (17). (b) Comparison of sO₂ quantification by using the traditional linear unmixing method (left) and eMSOT (right). The yellow arrows indicate a 0% sO₂ tube inserted inside the mouse tissue. Scale bar: 1 cm. Adapted with permission from (144).

Secondly, for oxygenation measurement, researchers have been investigating PA lifetime imaging (PALI), which relies on the oxygen-dependent excited-state lifetime of the sensor and is not prone to the tissue’s wavelength-dependent optical attenuation. PALI can potentially be applied for imaging guidance or treatment evaluation in photodynamic therapy and radiotherapy of cancers (139). Recent works have demonstrated PALI using methylene blue (127, 140, 141), and achieved oxygenation mapping at 15 mm depth with sub-millimeter spatial resolution (127, 139, 142). Another solution to the issue of unknown optical fluence at depths was recently demonstrated in eigenspectral multispectral optoacoustic tomography (eMSOT) (143, 144). eMSOT assumes that only a small number of fundamental spectra (four base spectra and three fluence eigenspectra) exist in biological tissues, and any fluence spectrum can be predicted by combining these spectra and thus the blood oxygenation can be computed (144). eMSOT has demonstrated superior performance in deep tissues in vivo, illustrated in Fig. 4(b), and reduced mean sO₂ measurement error from 35% by the traditional linear unmixing method to 4% (144).

Thirdly, great efforts have been spent on noninvasive measurement of blood glucose level for diabetic patients. Although measuring blood glucose level has been long sought after as a highly promising PA application, the low detection specificity has prevented its adoption in clinically relevant settings. Namita et al. have recently reported that the glucose concentration in either water or blood can be quantified from the PA signals acquired at 1120 nm (145). Sim et al. have demonstrated that it is possible to measure the glucose level in skin’s secretion products, using PA spectroscopy with a mid-infrared wavelength range of 8–10.5 μm (146). With relatively simple control of the skin conditions, the PA spectroscopy method has great potential for clinical translation (146). Other studies on PA glucose monitoring include linear calibration (147, 148), Helmholtz resonance cell (149) and multi-modal broadband dielectric spectroscopy (150). All these endeavors have contributed to the growing knowledge of optimizing PAT’s configurations and data analysis for accurate glucose measurement.

Last but not least, PAT can measure temperature, taking advantage of the Grüneisen parameter’s temperature dependence. Early studies employed this dependence to measure relative temperatures in single cells (151) and deep tissues (152, 153). Latest works have developed portable blood temperature monitoring (154) and dual-wavelength thermometry to reduce random errors (155). Absolute temperatures in deep tissues have been measured by exploring the dual temperature dependence of the Grüneisen parameter and the speed of sound (156), or by incorporating a multi-illumination scheme (157). Real-time volumetric PA monitoring of temperature distribution during photothermal therapy has further been demonstrated (5). Complementary to ultrasound and MRI thermometry (153), PA thermometry has a great potential to be incorporated with cancer thermal treatments to precisely control the temperature for maximized treatment effect and minimized thermal damage (158).

3.3. Molecular PAT

In addition to measuring various functional parameters, significant advances have been made in PA molecular imaging, especially for its quantification accuracy and detection sensitivity in deep tissues. Readers are referred to the recent Review article on the state-of-the-art molecular PAT technologies (42). One major challenge in molecular PAT is the relatively low detection sensitivity of genetically encoded probes, which are typically expressed at a low concentration level and can be easily shadowed by the strong background signals from blood. Recent methods based on compensating for the optical fluence, enhancing non-radiative relaxation, or extracting the signal’s temporal changes have been developed to improve the quantification accuracy in molecular PAT. These novel engineering strategies have been individually or concurrently implemented with novel PAT-specific molecular probes, aiming to improve the detection sensitivity. In addition to improving PAT’s detection sensitivity, another important research topic is integrating PAT with other imaging modalities with superior localizing capability or complementary contrast mechanism. In this section, we will discuss in details about new molecular PAT technologies, representative molecular probes, as well as their applications.

3.3.1. Enhanced molecular PAT based on modulation of optical properties

Activatable or photoswitchable molecular probes with NIR absorption, such as bacterial phytochrome and semiconducting polymer nanoparticles, have been optimized or engineered as PA contrast agents. By optically or chemically modulating the optical absorption of these molecular probes, one can suppress the background signals through ratiometric or differential measurements. The enhanced detection sensitivity has enabled the studies of reactive oxygen species, pH–vital chemical mediator, cancer metastasis, and cardiovascular diseases (159–161). Other activatable probes, including nitric oxide, hyaluronic acid nanoparticles and CuS nanodots, have also been explored for molecular PAT in mouse tumor models (160, 162–167).

Besides using activatable or photoswitchable probes, researchers have recently demonstrated the use of intraparticle engineering and self-quenched probes to enhance photoacoustic signals (168, 169). In intraparticle engineering, a secondary optical dopant is incorporated into the semiconducting polymer nanoparticle to quench its fluorescence and enhance nonradiative relaxation (168, 169). However, this method suffers from the potential leakage of the dopant from the nanoparticle (168). The self-quenched approach incorporates an electron-deficient structure unit into the backbone of the semiconducting polymers to enhance the heat generation and thus PA signals (168).

3.3.2. Enhanced molecular PAT based on temporal modulations of the PA signals

Alternatively, magnetomotive PAT has demonstrated improved detection sensitivity. While aforementioned techniques in Section 3.3.1 modulate the optical properties of the molecular probes, magnetomotive PAT relies on the physical displacement of the probes driven by an oscillating magnetic field. By “poking” the magnetic nanoparticles using magnetic force, PAT can suppress the background signals from non-magnetic biomolecules, and thus improve the detection sensitivity (170). Magnetomotive PAT has been demonstrated on gold-shelled or polymer-shelled magnetic nanoparticles (171, 172). Moreover, calcium-sensitive indicators have also been used by PAT to study neuron activities. These calcium indicators include arsenazo III (173), GCaMP (174), CaSPA550 (175), NIR metallochromic compounds (176), and acoustogenic probes for copper(II) (177). However, unlike two-photon microscopy, PAT cannot yet visualize individual neurons in the mouse brain, limited by the spatial resolution in deep tissues (173).

3.3.3. Molecular PAT with multi-modal contrast

PAT has been integrated with other imaging techniques, such as magnetic resonance imaging (MRI), Raman imaging and positron emission tomography (PET). MRI, while providing accurate localization of lesions, has difficulty to provide sufficient molecular sensitivity. This limitation of MRI can be complemented by PAT. Dual-modal contrast agents have been developed for PAT and MRI, such as lipid-micelles incorporated with semiconducting polymer dots, a photosensitizer, and PEGylated polypyrrole nanoparticle (178, 179), as well as Gadolinium(III)-gold nanorods for macrophage detection in atherosclerotic inflammation (180). Moreover, PAT has been incorporated with Raman imaging and PET for enhanced molecular performance. Kircher et al. integrated surface-enhanced Raman scattering (SERS) with PAT/MRI for tumor margin detection (181), in which Gold silica–based SERS nanoparticles coated with Gd3+ were used to visualize the brain tumor with high spatial resolution (181). PET was also integrated with PAT/MRI for imaging reporter-gene products with high sensitivity (182).

3.4. Other Technical Advances in PAT

Here, we also introduce other PAT advances that focus on addressing the technical obstacles from various perspectives, such as the need for coupling medium, reconstruction artifacts, and the integration with other imaging modalities.

3.4.1. Non-contact (remote-sensing) PAT

In PAT, the need for a coupling medium (e.g., water, ultrasound gel) or direct contact with the targets has been considered a technical disadvantage in comparison with other pure optical imaging modalities. The need for coupling substantially increases the complexity of the system configuration and experimental procedures. Therefore, new methods have been reported on non-contact PAT that does not need any coupling during the imaging process. These non-contact PAT methods mostly utilize optical sensing of the acoustic signals, for example, by detecting the displacement of the sample surface induced by pressure waves. The pressure-induced surface displacement can be captured by optical interferometers. The recorded optical signals are then used for the image reconstruction. A number of optical interferometer techniques have been explored in non-contact PAT, including Mach-Zehner interferometer (183), Fabry-Perot interferometer (184, 185), Michelson interferometer (186), photoreactive-crystal two-wave mixing (187, 188), and speckle-tracking-based interferometry (189, 190). Moreover, recent studies have focused on compact detecting systems using fiber-based interferometers, which have further reduced the system footprint and may enable endoscopic applications (191–193). This fiber-based approach has been expanded to a line-detector setup with multiple fibers in photoacoustic projection imaging, which is similar to the X-ray computed tomography and capable of all-optical volumetric imaging (194, 195). Overall, the advances in non-contact PAT have the potential to overcome the traditional barriers for clinical applications such as thyroid cancer detection, burn evaluation, and brain surgery guidance (191). Recently, a non-interferometry-based optical sensing method has been developed for detecting the PA pressures through the changes in the tissue’s refractive index (196). This method is promising for integrating with other optical imaging modalities, such as OCT for ophthalmological applications (196).

3.4.2. Machine Learning in PAT

The PAT community is also closely following the current computational revolution in machine learning. In the past several years, different machine learning methods have been explored to reduce the reconstruction artifacts in PACT images, without the need for increasing the acoustic detection aperture or spatial sampling frequency. For example, U-Net, an encoder-decoder-based convolutional neural network originally developed for image segmentation, has been adapted in PACT for direct image reconstruction or correction of the delay-and-sum method (22, 197, 198). The machine learning methods have shown promising performance in eliminating reconstruction artefacts with simulation data, with an l₂–reconstruction error of 1% (198). Reverberation artifacts can also be reduced by training the network to identify the signal reflection patterns (199, 200). A recent study by Antholzer et al. has been successful in correcting the frequency-limited and angle-limited ultrasound detection for real-time projection imaging (201–203). It integrates U-Net with dynamic aperture-length correction (applying learnable weights to reconstruction grid points) (202, 203). Another application of machine learning in PAT is to improve the quantification accuracy without knowing the prior information of the tissue’s optical properties. For example, Kirchner et al. have implemented a context-encoded fluence distribution map with random forest regression to estimate blood oxygenation on simulation data (204). Another study, which also utilizes U-Net with extra skip connections, takes raw radiofrequency data at different wavelengths to reconstruct blood oxygenation and molecular probe concentrations (205). All in all, machine learning, although still a relatively new concept in PAT, has been providing possible solutions for many long-standing technical challenges, mostly for artifact elimination, without complicating the system setup or increasing the cost. Machine learning in PAT may also help automatic diagnosis and staging of diseases, such as breast cancer. However, a typical challenge that machine learning methods face is the lack of training data with ground truth acquired in experimental or clinical settings. Simulation is still the primary source of data generation, which does not provide necessary variation to effectively train the learning models. Data augmentation on simulation data can be used to increase the robustness of network training.

3.4.3. Hybrid photoacoustic-ultrasound (PAUS) imaging

The high complementarity of pulse-echo ultrasonography (US) and PAT has also prompted the development of combined approaches that utilize the same transducer array for visualizing both contrasts. Yet, an efficient integration is not straightforward as it implies accounting for the fundamentally different excitation and image formation strategies of those modalities. Initially, hybrid imaging was achieved by adding fiber-guided light irradiation next to the conventional US linear array probe (206). Although this acquisition geometry attains good performance in sensing back-scattered ultrasonic waves, it is proved to be inefficient in recording of PA signals and has generally resulted in reduced image quality and lack of quantification abilities, mainly because of the limited tomographic view of the linear-array geometry (53). Recent approaches for efficient hybridization consist ofadapting concave array geometry typically used in small-animal MSOT systems (133, 207) for concurrent photoacoustic-ultrasound imaging (208, 209). Due to their larger angular tomographic coverage, hand-held MSOT scanners based on concave arrays outperform those based on linear arrays in terms of signal quantification abilities and overall image quality (210). Most recently, a multi-segment detector array approach incorporating array segments of linear and concave geometry to optimally support both ultrasound and optoacoustic image acquisition has been introduced (211). The new approach enabled the acquisition of high-quality anatomical data by both modalities complemented by functional information on blood oxygenation status provided by MSOT (212).

4. Representative Applications of PAT in Life Sciences and Clinical Studies

The advances in PAT have enabled numerous applications in life science and clinical studies that are not attainable by current technologies. In this section, we highlight several representative applications to demonstrate PAT’s potential impact, while many other exciting works cannot be covered due to the page limit. In Table 1, we have summarized representative PAT applications over a wide range of scales. Readers are strongly referred to other comprehensive reviews on PAT’s latest applications. Overall, from single cell analysis to intraoperative imaging guidance, PAT has been applied in a broad range of fundamental studies, and the translation of PAT from lab technology to clinical impact is accelerating.

Table 1.

Representative Applications of PAT

Applications	Imaging contrast/principle	Measured information	PAT system	Selected ref
Cellular Imaging	Hemoglobin in red blood cells	RBC morphology, oxygen saturation, cell aggregation, stimulation-response	Ultra-high-frequency PAM	(54, 213, 214)
	DNA/RNA in cell nuclei	Cell nuclei morphology, nuclear size and packing density in tumor margin assessment	Ultraviolet (UV)-PAM	(214–216)
	Cytochrome in mitochondria	Mitochondria morphology and interconnectivity	OR-PAM/PA nanoscopy	(2, 151)
	Nanoparticle labeled stem cells	Track stem cell (embryonic, mesenchymal, etc.) migration	Dual-mode ultrasound-PACT	(217–219)
	Stained seed cells in engineered scaffolds	Track cell seeding within and quantify porosity and degradation of bio-scaffold	OR-/AR-PAM	(33, 167, 220)
Functional imaging	Oxy- and deoxy-hemoglobin	Map blood oxygenation of whole mouse cortex	MEMS OR-PAM	(15)
	Photoacoustic thermal tagging	Measure slow blood flow in deep mouse brain	PACT	(17, 136)
	Grüneisen parameter	Measure the tissue temperature during thermal therapy	PACT	(152, 153, 156)
	Hemodynamic fluctuation	Map brain’s functional connectivity during resting state	PACT	(221–224)
Molecular imaging	Nanoparticle, organic dyes, genetically encoded chromophores	Use exogenous contrast agents as the probes for tumor imaging and cancer staging	PACT	(42, 225, 226)
Cancer imaging	Circulating tumor cells	Count the circulating tumor cells in melanoma as an independent indication of metastasis	PA flow cytometry	(227–231)
	Breast cancer	Estimate the breast tumor size based on the angiogenesis or labeled by nanoparticles	PA mammography	(232–239)
	Melanoma	Evaluate the volume, depth and thickness of melanoma and provide staging	PACT	(40, 41)
	Prostate cancer	Visualize prostate brachytherapy seeds, detect prostate cancer cells labelled by nanoparticles	PACT	(240–244)
Surgery or treatment guidance	Sentinel lymph node (SLN)	Track the needle tip and map SLN location during biopsy	PACT	(226, 245, 246)
	PA-guided surgery	Real-time guidance for endonasal surgery and minimally invasive fetal surgery	PACT integrated with surgical tools	(247–252)
	High-intensity focused ultrasound	Monitor temperature and blood oxygenation during HIFU ablation	PACT integrated with HIFU	(136, 253, 254)
	Photothermal therapy	Monitor the nanoparticle delivery during photothermal therapy	PACT	(255–257)

4.1. PAT Applications in Life Sciences

For fundamental research, PAT can visualize tissue functions with molecular specificity, on the single vessel, single cell, or even subcellular levels (43). Quantitative analysis of the photoacoustic signals acquired from single vessels/cells can provide information about their anatomical, mechanical, functional, and molecular properties, which can be used to enhance the understanding of biological processes at the microscopic level. It is important to note that the label-free imaging capability of PAT is often highly desirable in biomedical studies in order to improve the throughput and mitigate the labeling artifacts. For example, Wang et al. have used high-resolution high-speed PAM to track the oxygen release from single red blood cells flowing in capillaries in a mouse brain (258). Moreover, a recent work has demonstrated label-free histology-like assessment of tumor margins in human lumpectomy specimen, taking advantage of PAM’s intrinsic sensitivity to DNA and RNA within the ultraviolet wavelength range (259). This work has also been expanded to image the whole organs ex vivo by microtomy-assisted PAM (260).

4.2. PAT on the Move to Clinical Studies

For human studies, PAT has been increasingly utilized for cancer detection (199, 200), surgery guidance (201–204), and treatment monitoring (205–207). Early-stage breast cancer detection is one of the major applications of PAT moving quickly towards clinical translation, taking advantage of PAT’s high sensitivity to tumor hypoxia and angiogenesis. To this end, several groups have reported promising clinical studies on breast cancer patients. For example, the Twente PA mammography, illustrated in Fig. 5(a), has recently demonstrated the detection of 31 malignant breast tumors, with a sensitivity of 96.8% (208). The LOUISA-3D system can provide volumetric human breast vasculature, clearly showing the tumor structure (209). MSOT has also been applied in human breast imaging, providing tumor visualization with recurring features such as high-peripheral low-intratumoral vascularity, and increased total blood volume heterogeneity (261). Another direction of clinical study using PAT is dermatology. For example, Chuah et al. have developed volumetric MSOT (vMSOT) to image skin cancers with good correlation with histology results, demonstrating universally high accuracy regardless of the tumor types and skin conditions (262). In another study of dermatology, dual-mode PAT/OCT was able to identify skins with different types of dermatitis, such as chronic hyperkeratotic hand eczema, chronic hand eczema, and atopic dermatitis inflammatory skin disease (Fig. 5(c)) (263). PAT has also been applied in oral health and otorhinolaryngology. One example is the portable OR-PAM of early-stage oral cancer, which depicts the vascular networks in the human lips up to 0.5 mm in depth (Fig. 5(b)) (264). In another study by Zhao et al. (265), PAT was used for detecting thyroid cancer together with color Doppler ultrasound (Fig. 5(d)), in which PAT provided detailed information of the tumor vasculature at a depth of 2 cm on patients with suspicious malignant thyroid nodules (265). More recently, Ivankovic et al. (266) demonstrated a hand-held vMSOT system for real-time volumetric imaging of the human carotid artery with a 2-cm³ field-of-view (Fig. 5(f)). This system opens up a wide range of applications in cardiovascular diseases, especially in early assessment of stroke risk. MSOT has also been utilized in assessing Crohn’s diseases by evaluating hemoglobin signals in the intestinal walls (Fig. 5(e)) (267), and metastasis status of sentinel lymph nodes (SLN) in melanoma by using indocyanine green (ICG) as the contrast agent (Fig. 5(g)) (268). Besides the above human studies, several other PAT technologies are on the horizon of clinical translations (269, 270). For instance, transrectal PAT is under development for prostate cancer detection and biopsy guidance (59, 244). PAT-enhanced robotic surgical system is promising for guiding endonasal and fetal surgery with high precision (247, 248, 250, 271). Temperature mapping is another focus of PAT’s treatment guidance, especially for HIFU ablation and interstitial laser phototherapy (136, 253, 254, 272).

Figure 5.

Representative PAT applications on humans. (a) Twente photoacoustic mammoscopy for breast cancer imaging; left: overview of the system, right: MAP image of the breast showing the lesion clearly on the right side (233). (b) In vivo imaging of oral vasculature; left: 3D display of vessels in a human lip, middle: corresponding MAP image, and right: PA image of the deeper region (264). (c) In vivo imaging of the palm vessels; left: 3D image of the vascular network; right: the imaging system schematics (263). (d) Dual-mode PA/ultrasound imaging of human thyroid. Left: right lobe papillary thyroid cancer image. Right: schematics of the imaging system (265). (e) Illustration of the MSOT of inflammatory bowel disease (267). (f) vMSOT of the human carotid artery, showing the carotid bifurcation (266). (g) MSOT of SLN using ICG as the exogenous contrast (268).

5. Conclusions and Future Prospects of PAT

In this concise Review article, we have introduced recent advances in PAT technologies, as well as their representative applications in life sciences and clinical studies. As a fast developing technology, PAT has a great potential to enable novel biomedical discoveries and generate significant clinical impact. PAT is a highly scalable imaging modality providing balanced spatial resolution over a wide range of penetration depth. Its functional and molecular sensitivity originates from the unique optical absorption contrast, and has enabled the measurements of a large number of physiological parameters. Enabled by multi-spectral signal analysis, PAT can image numerous endogenous and exogenous contrast agents, with high molecular specificity.

5.1. Technical challenges of PAT

Despite its powerful imaging performance and exciting new applications, PAT still has several major challenges remained to be addressed. First of all, the limited penetration is PAT’s most significant drawback when compared with other whole-body imaging modalities. At present, PAT has reached up to 5 cm using external light illumination (262), which is insufficient for many clinical applications such as kidney and liver imaging. Secondly, wavelength-dependent light attenuation in living tissues leads to unknown local optical fluence at depths, preventing accurate quantitative imaging. Thirdly, super-resolution PAM has made significant progress in the recent years, but its low signal-to-noise ratio still prevents single-molecular imaging. Fourth, due to the strong acoustic aberrations introduced by air cavities and bones (273), PAT has difficulties in imaging the lungs and brain (274), especially for human applications (275). Last but not least, image artifacts due to sub-optimal data acquisition (e.g. limited-view or limited-detection-bandwidth) often lead to reduced imaging quality.

5.2. Potential Future Prospects of PAT

The above challenges have also pointed out the potential future prospects of PAT. A number of technological breakthroughs are anticipated in the coming years. Currently, super-resolution PAM, discussed in Section 3.3.1, has achieved approximately 100 nm spatial resolution with various nonlinear signal generation mechanisms. Future development of SR-PAM using higher-order nonlinearity will achieve even better resolutions on the level of ~50 nm, and may provide single molecule sensitivity. Such a high resolution and sensitivity will enable PAM to compete with other super-resolution imaging modalities such as photo-activated localization microscopy and stimulated emission depletion microscopy (276). Secondly, PAT is highly promising for functional brain imaging of direct neural activities in small animal models, using NIR voltage- or calcium-sensitive indicators that are specially engineered for enhanced optical absorbance. PAT possesses intrinsic advantages over conventional optical microscopy in terms of penetration depth. The current voltage- and calcium-sensitive indicators are mostly designed for fluorescent imaging in the visible wavelength range, and thus not ideal for PAT’s deep-penetrating applications. Recent advances in protein engineering have shown great promise for developing NIR protein sensors (277), whereas the first NIR calcium-sensitive protein has been recently reported (278). Thirdly, multi-scale multi-parameter PAT as a single imaging device is possible. As discussed above, existing PAT systems have been individually applied for measuring functional and molecular information at different scales, ranging from organelles to organs. Future developments in PAT may integrate different imaging scales into one device and can measure multiple parameters simultaneously. Doing this, PAT’s multi-scale multi-contrast imaging capability can be truly realized, enabling the studies of the same biological process at different scales. Lastly, low-cost compact PAT systems may become highly beneficial for longitudinal monitoring of physiological parameters such as oxygenation, blood flow, and glucose level. They may also find use in detecting circulating tumor cells as an independent assessment of cancer metastatic potential and evaluation of treatment efficacy, such as immunotherapy and chemotherapy (279, 280). With the initial success on low-cost PAT using laser diodes, further advances in compressed sensing and machine learning can reduce the hardware complexity and overall system costs.