Photoacoustic (PA) imaging (PAI) is a biomedical imaging technique that exploits the PA effect to visualize the contrasts of optical absorption among biological tissues. The PA effect refers to the instantaneous thermal expansion of optically absorptive targets when they are illuminated by a laser pulse, followed by the generation of ultrasound (US) waves^[1,2]. A PAI system detects these waves using US transducers and then uses the signals to reconstruct PA images that show the targets in the tissue^[3]. PAI provides deeper imaging depth than optical imaging because US waves scatter less than light does within biological tissues. The major targets of biomedical PAI include hemoglobin (Hb) and melanin, using methods that exploit their high optical absorption in the visible and near-infrared (NIR) spectral ranges^[4–7]. Therefore, PAI can provide detailed images of vascular structures or melanin distribution. To expand PAI applications, PA images can be obtained using other spectral ranges, such as ultraviolet (UV) or even x-rays^[8,9]. Using multi-wavelength laser illumination, PAI can separate the mixed PA signals from various chromophores by applying spectral unmixing techniques to their unique absorption spectra^[10–12]. Spectral unmixing of oxy-hemoglobin (${\mathrm{HbO}}_2$) and deoxy-hemoglobin (HbR) is used to measure the hemoglobin oxygen saturation (${\mathrm{sO}}_2$) which serves as a biomarker for malignant tumors. Exogenous PA agents can be used to increase the spatial resolution, signal-to-noise ratio (SNR), and imaging depth, and enable target-specific drug delivery monitoring or theragnosis in combination with photothermal therapy^[13–16]. While most of the PAI applications are based on the linearity of PA signal amplitude with respect to the optical properties, the nonlinearity of the PA signal can be utilized to achieve super-resolution imaging or functional imaging such as thermometry^[17,18]. PAI is frequently combined with US imaging^[19–22] or with other optical imaging modalities [e.g., optical coherence tomography (OCT) or fluorescence imaging]^[23,24] to provide complementary structural or functional information. Further, deep learning has been applied to PAI to improve image reconstruction performance and spatial or temporal resolutions^[25,26].

These advantages make PAI suitable for various fields of pre-clinical research such as neuroimaging, ophthalmic imaging, skin imaging, cancer imaging, and cardiovascular or peripheral vascular imaging. Neuroimaging applications of PAI provide a high-resolution mapping of ${\mathrm{sO}}_2$ in brain vasculature and can monitor microhemorrhages, so the method is useful in pre-clinical studies of brain anomalies such as stroke^[27–31]. In ophthalmic imaging, PAI can show retinal vasculature and ${\mathrm{sO}}_2$ and melanin in the retinal pigment epithelium; this information is useful in diagnosing ocular diseases^[24,32–35]. PAI can also be used in studies to quantify vascular changes induced by medications^[36].

In some areas, PAI is also being applied to clinical studies as well as pre-clinical research^[37–39]. In cancer imaging, PAI is effective in visualizing tumor angiogenesis and hypoxia related to the malignancy of the cancerous lesion such as thyroid cancer and breast cancer^[40–47]. In cardiovascular imaging, the ability of PAI to detect lipids has been exploited to characterize and monitor atherosclerotic plaques^[48–51]. PAI can visualize vasculature and thus help to assess peripheral vascular diseases^[21,52–55]. PAI can visualize microvascular structure and dynamics, particularly in the skin^[56].

Among these clinical applications, PAI is especially useful for diagnosing and monitoring skin diseases, such as inflammations, wounds, and skin cancer, which are accompanied by abnormal changes in the vasculature or melanin distribution^[57–59]. For example, melanoma, psoriasis, dermatitis, and port-wine stains (PWS) have distinct structures of blood vessels around lesions, and melasma and melanoma are strongly associated with abnormal melanin distribution^[60–66].

Conventional diagnostic tools for skin diseases include skin biopsies and optical or US imaging, but these methods are invasive, require subjective evaluation, or have low sensitivity. Skin biopsy for histological examination is the gold standard for diagnosing many skin diseases, but it leaves a wound and can facilitate infection; it is also impractical for assessing large lesions^[67]. Optical imaging such as photography or fluorescence imaging (i.e., Wood’s lamp) can be performed in vivo to visually evaluate the color, size, or boundary of the lesions, but it can only assess superficial skin features to depths of $<1\mathrm{mm}$. Conversely, US imaging can image centimeters into the tissue but lacks the sensitivity to visualize pigmentation or microvascular structures effectively. Recently, the ultrafast Doppler (UFD) technique has been used for imaging highly vascular organs such as kidneys^[68], but it has not been used yet for investigating skin microvasculature. PAI is non-invasive, has a high spatial resolution, and is sensitive to Hb and melanin; therefore, the method can effectively overcome these limitations of conventional methods. PAI can visualize the microvascular structure and melanin distribution below the skin surface with high contrast and can provide quantitative evaluation of these features to enable distinction between lesions of different skin diseases.

In this review, we discuss clinical PAI techniques for imaging human skin to resolutions ranging from micrometers to millimeters. First, we introduce the principles of the PAI modalities used for skin imaging in three categories (Fig. 1) according to the scale: PA microscopy (PAM), PA mesoscopy (PAMes), and PA tomography (PAT). Next, we discuss PAI applications to image two types of imaging targets in the skin: vasculature only or vasculature plus melanin. To the best of our knowledge, there were only several recent PAI reports that discussed melanin imaging only, and thus we partially included the vascular imaging in the latter section. Each section is divided into three subsections that present uses of PAM, PAMes, and PAT, respectively, to compare their different imaging scales of the same imaging target. We review peer-reviewed papers published from 2018 to 2024, with a focus on imaging techniques for human skin, and on clinical indicators that consider vasculature and melanin. We then analyze the general relationship between spatial resolution and imaging depth or field-of-view (FOV) to guide the selection of an appropriate PAI modality for specific skin-imaging applications.

The major components of PAI systems are the optical excitation source (i.e., laser) and the acoustic detector (i.e., US transducer). Both the optical beam delivery and acoustic detection can be made either focused or unfocused (or diffuse), depending on the desired imaging depth and spatial resolution. In biological tissues, scattering is more than US waves, so each photon can ballistically travel up to only $\sim 1\mathrm{mm}$ on average (i.e., transport mean free path). Therefore, the optical beam can only be focused at $<1\mathrm{mm}$ depth in biological tissues; beyond that depth, the beam diffuses randomly. Contrarily, the generated US waves scatter much less and can be focused at depths from several millimeters to centimeters, but attenuation of the waves tends to increase as their frequency increases. Therefore, depending on the system configuration, PAI can provide either high spatial resolution at a shallow depth or low spatial resolution at a deep depth. Generally, PAI is categorized into three modalities: PAM, which uses a focused laser and a focused US detector to achieve the finest spatial resolution at shallow depths; PAMes, which uses an unfocused laser and a high-frequency focused US to achieve high resolution and less superficial imaging depth; and PAT, which uses an unfocused (diffused) laser and multiple low-frequency unfocused US transducers (i.e., a US transducer array) to achieve deep imaging depth but low spatial resolution.

A typical PAM system consists of a single-element focused US transducer, a laser source (usually emitting $\lambda =532\mathrm{nm}$) with its focusing optics, and an optical-acoustic beam-coupling module. PAM can be further divided into two types: optical resolution PAM (OR-PAM) and acoustic resolution PAM (AR-PAM).

OR-PAM uses a tightly focused laser beam to achieve high lateral resolution typically in the range of a few micrometers. Recent improvements in OR-PAM can achieve sub-micrometer resolution using a transparent US transducer to increase the numerical aperture (NA) of the optical configuration^[9]. However, beyond the depth limit ($\sim 1\mathrm{mm}$) of optical diffusion in biological tissues, optical scattering greatly reduces the focusing effectiveness, so lateral resolution degrades. AR-PAM can image to depths of several millimeters, so it is suitable for imaging relatively deep tissues while maintaining adequate resolution^[69].

AR-PAM relies on acoustic focusing by the US transducer to achieve lateral resolution typically in tens of micrometers, which is determined by the US frequency of the transducer, NA, and the corresponding focal spot size. In principle, AR-PAM and PAMes both use a diffuse laser in relatively deep depths, but AR-PAM may use a quasi-focused laser when switchably combined with OR-PAM. In such a switchable OR/AR-PAM configuration, the acoustic focus position can be shifted to either $<1\mathrm{mm}$ depth for OR-PAM or $>1\mathrm{mm}$ for AR-PAM, while sharing the same laser focusing at $<1\mathrm{mm}$ depth.

Both OR- and AR-PAM acquire time-resolved amplitude data (i.e., A-line data) for a single point per laser pulse. The axial resolution depends on the US frequency of the transducer. To obtain 2D or 3D images, multiple A-lines must be acquired by raster scanning using mechanical stages. PAM images are formed directly from the data acquired at the focal depth without additional reconstruction algorithms, but AR-PAM may use synthetic-aperture focusing techniques to improve lateral resolution at out-of-focus depths^[70]. Moreover, image post-processing techniques such as motion correction or vascular signal enhancement techniques can be applied to additionally improve the PAM images^[71,72]. Further, deep-learning algorithms can be used on PAM images for automatic microvessel segmentation or virtual staining^[73,74].

However, conventional mechanical raster scanning is time-consuming and susceptible to motion artifacts, so it has limited clinical applications. Scanning speed can be increased using advanced optical or acoustic scanning systems such as Galvo scanners, micro-electromechanical system (MEMS) scanners, or polygonal mirrors^[75–79]. Laser scanning OR-PAM systems have been also proposed to increase imaging speed^[80,81]. Instead of increasing scan speeds, some studies use deep learning to increase temporal resolution by achieving sufficient resolution while requiring less data than conventional methods^[82]. Alternatively, the spatial light modulation technique enables high frame rate PAM imaging without mechanical scanning^[83].

PAMes operates on a scale between AR-PAM and PAT, typically using wide-field diffuse laser illumination and a single-element focused US transducer. PAM is similar to AR-PAM in that it is an acoustic-resolution modality using a focused US transducer. However, AR-PAM is often configured with an optical lens to focus the laser to be switchable with OR-PAM. On the contrary, PAMes uses a diffused laser to deliver higher laser power to image deeper into the tissue. PAMes stacks multiple A-lines to create 2D or 3D images, but it also uses reconstruction algorithms to increase spatial resolution and SNR^[84]. The method usually achieves imaging depths of up to 10 mm with spatial resolutions in tens of micrometers. Particularly, raster-scan optoacoustic mesoscopy (RSOM) from iThera Medical GmbH, the most widely used PAMes system, uses ultra-broadband ($\sim 100\mathrm{MHz}$) and a high-frequency (tens of megahertz) US transducer, and divides the PA signals into two acoustic frequency bands to render small and large vessels separately^[85]. This feature provides higher axial resolution than typical PAM systems and thus provides the ability to distinguish the epidermis from the dermis within the skin tissue.

PAT targets the visualization of tissue structures within several centimeters in all directions. It uses wide-field laser illumination and a US transducer array (typically 64 to 256 elements) for tomographic image reconstruction and, therefore, can acquire a cross-sectional image (i.e., B-mode) with a single laser pulse^[86]. For 3D imaging, PAT can either use a 1D transducer array (e.g., linear and curved) with a motorized scanner or a 2D transducer array that acquires volume data at once^[87–89]. In addition, 3D handheld PAT systems have been developed for versatile imaging of healthy or diseased tissues positioned in narrow locations where conventional 3D PAI systems cannot fit in due to size^[90,91].

PAT usually uses a low-frequency (several megahertz) US array transducer, which facilitates visualization of deeper tissues than PAM or PAMes at the cost of a loss of spatial resolution (to hundreds of micrometers). Furthermore, most US transducer arrays have a limited signal reception aperture and may not fully acquire the PA waves that propagate omni-directionally, so limited-view artifacts may develop. Therefore, advancements in image reconstruction are needed to maximize the achievable image quality even in the limited signal reception environment^[92]. Recent studies have used deep learning to solve the low spatial resolution or limited-view artifacts of PAT systems^[93,94].

PAT uses a wide-field laser and, therefore, does not rely on optical focusing and does not require a very high-PRF laser, as PAM does. Therefore, multiple optical wavelengths can be used relatively easily to estimate the relative concentrations of chromophores (e.g., ${\mathrm{HbO}}_2$, Hb, and melanin); this method is often called multi-spectral optoacoustic tomography (MSOT).

We now compare the three PAI modalities described above (Fig. 1). The basic principle of PAI is that the laser is illuminated on the target tissue and the corresponding PA signal is generated according to the optical absorption coefficients of chromophores [Fig. 1(a)]. Typically, PAM and PAMes use the visible range to achieve high spatial resolution and sensitivity, whereas PAT can also use NIR when deep penetration is required. Many pulsed laser sources that use an optical parametric oscillator (OPO) support fast wavelength switching in the NIR range; this ability is suitable for multi-wavelength PAT or even some PAMes implementations.

The systems that use each modality have different constructions and applications, and they provide different types of images [Fig. 1(b)]. PAM and PAMes systems commonly use a focused US transducer with motorized raster scanners to acquire images, whereas PAT systems use unfocused US transducer arrays. In PAM, particularly for OR-PAM, the laser is focused at the epidermal depth ($<1\mathrm{mm}$) to achieve high lateral resolution, whereas PAMes and PAT systems use diffuse lasers. The high spatial resolution of PAM allows precise separation of each skin layer up to the superficial dermal layer and enables detailed delineation of the vasculature and melanin. PAMes can identify and analyze the structures up to deep dermal layers with a slightly degraded spatial resolution. PAT can image at a depth much deeper than the subcutaneous tissue and has a wider FOV than the other two modalities. It can show larger vessels at greater depths than the other modalities and can depict the melanin in a whole lesion across multiple layers of the skin.

Cheng et al. developed a typical OR-PAM system that can achieve centimeter-scale scan range (up to $40\mathrm{mm}\times 40\mathrm{mm}$) to image human skin microvasculature^[95]. The optical and acoustic beams are coaxially aligned by passing a 532 nm laser beam through a 3 mm central hole on a US transducer (center frequency: 40 MHz, bandwidth: 110%). The system yielded spatial resolutions of 31.2 µm axially and 5 µm laterally. One B-scan image in the 5 mm range can be acquired in about 0.1 s (step size: 5 µm, PRF: 10 kHz). To demonstrate human skin imaging, the back and palm of the hand areas of a healthy volunteer were imaged and qualitatively compared in terms of the number, depth, and diameter of dermal vasculature. The maximum PA signal detection depth was greater in the palm of the hand ($\sim 2\mathrm{mm}$) than in the back of the hand ($\sim 1.5\mathrm{mm}$) due to the higher concentration of melanin pigment in the back of the hand than in the palm. The distribution of vessel depths and diameters was more diverse in the palm than in the back of the hand; the difference was more obvious when using depth-encoded PA MAP images than when using regular PA MAP images.

Ahn et al. have combined OR-PAM and photoplethysmography (PPG) to track the movement of single blood microvessels and PPG signals simultaneously in a finger [Fig. 2(a)]^[96]. This implementation of OR-PAM used a galvanometer scanner, a flat ring-shaped US transducer, and a parabolic mirror. The galvanometer scanner allows fast B-mode scanning at a frame rate of 50 Hz, and the parabolic mirror steers both light and acoustic waves simultaneously with a short focal length and 2.5 µm lateral resolution. This system could simultaneously acquire cross-sectional vascular images and measure the blood volume of the finger, which were synchronous to the co-acquired PPG signals. The heart rates of three volunteers were measured using both OR-PAM and PPG, and the measurements by the two modalities agreed well.

Wang et al. have constructed a dual-modal PA/US dermoscope using a bimorph US transducer that uses lead zirconate titanate (PZT) (for US transmission) and polyvinylidene fluoride (PVDF) (for PA and US reception) [Fig. 2(b)]^[97]. PZT provided strong US wave generation, and the PVDF had wide signal reception bandwidth to provide high spatial resolution. The system achieved an axial resolution of 57.8 µm for both modalities, with a lateral resolution of 45 µm for US microscopy and 6.8 µm for PAM. A human palm could be imaged to a depth of $\sim 1.5\mathrm{mm}$, which allowed high-resolution images of the stratum corneum (SC) and superficial vessel plexus, but the signal from the dermal layer was not strong enough to analyze the structure. Complementarily, US images could visualize skin up to a depth of 5 mm, where fibrous connective tissue and large blood vessels are located. This work demonstrated that the dual-modal PA/US skin imaging could fully depict the skin tissue structure as well as the microvasculature.

To increase the visibility and quantitative accuracy of PA dermoscopic images, a 4D spectral-spatial computational model of the skin has been developed^[98]. It considers the heterogeneous optical and acoustic tissue parameters. The model was built with a Monte-Carlo simulation of optical propagation, k-Wave MATLAB toolbox for acoustic simulation, and detailed modeling of the multilayered skin structure in the 3D spatial domain as well as the spectral domain (i.e., optical wavelength). A computational skin model was used to generate in silico PA dermoscopy data pairs (low center frequency vs. high center frequency and strong vs. weak scattering) to train two U-Net neural networks to increase the spatial resolution and imaging depth. The trained network significantly increased the axial resolution and visibility of deep vessels in images of human palm skin.

Cheng et al. used an extended depth-of-field PAM (E-DOF-PAM) to monitor the revascularization process during the healing of a skin wound^[99]. An elongated focus lens was added to a conventional Gaussian beam PAM (GB-PAM) system to create Bessel beams that allowed relatively uniform spatial resolution at different depths. Compared to GB-PAM, the lateral resolution of the E-DOF-PAM was slightly degraded from 4.7 to 6.5 µm, but the depth-of-field was significantly increased to nearly 1 mm. Wounded skin on a volunteer was imaged using E-DOF-PAM, and the image separately displayed the SC, stratum basale (SB), and dermal vascular layer. PAM images of the wound showed increased thickness and PA amplitude in the SC layer, disappearance of signals in the SB layer, and development of small and dense microvasculature in the dermal layer.

Ma et al. developed a non-invasive photoacoustic microscopic biopsy (PAMB) system to visualize and quantify dermal angiopathy with high spatial resolution and penetration depth^[100]. The PAMB system integrates a 532 nm laser and a broadband (5–90 MHz) US transducer, and achieves axial resolutions from 34 to 57 µm and lateral resolutions from 1.5 to 104 µm (1.5 to 3.8 µm on the zero plane, 10 to 40 µm at 0.2 to 1.5 mm depth, and 40 to 140 µm at 1.5 to 3.0 mm depth in skin tissue), with a maximum imaging depth of 3 mm. This system uses an adjustable confocal opto-sono objective that allows switching among different NAs to optimize focus at various depths. The system quantified vascular parameters, including capillary-loop depth, diameter, and vascular volume, and it could distinguish healthy skin from PWS-affected skin (Fig. 3). PAMB demonstrated significant differences in vascular features among different PWS stages and, thereby, provided a quantitative assessment of lesion severity. The authors introduced a comprehensive parameter (PAFIND) that combines microvessel density and depth to enable objective evaluation of PWS severity and treatment efficacy. The study highlighted PAMB’s potential for non-invasive, label-free biopsies, and it may be a valuable tool for diagnosing and monitoring skin diseases and treatment responses.

Hindelang et al. used RSOM to differentiate between allergic and irritant skin reactions in patch testing^[101]. The RSOM images were analyzed to provide quantified vascular biomarkers such as blood volume, vessel fragmentation, and frequency-content ratio. Allergic reactions showed higher vessel fragmentation and lower frequency ratios than did irritant reactions. The RSOM system used in this study consisted of a 532 nm laser and a 55 MHz US transducer. The US transducer has an ultra-broadband bandwidth of 10 to 120 MHz, so the RSOM provided images that had spatial resolutions of 8 µm axially and 30 µm laterally, and an imaging depth of 1.5 mm. The researchers divided the detected PA signal into two frequency bands, i.e., 10 to 40 MHz to distinguish large vasculature, and 40 to 120 MHz to distinguish small vasculature [Fig. 4(a)]. Compared to dynamic optical coherence tomography (D-OCT), RSOM provided similar image quality in en face images but better quality in transverse cross-sectional images due to its axial resolution and penetration depth.

The RSOM technique has also been used to quantify skin sensitivity to UV radiation in healthy volunteers^[102]. Skin areas were exposed to increasing doses of UV radiation, and the dermal microvasculature was monitored in RSOM images. UV radiation induced vasodilation and recruitment of vessels, and the vessel fraction increased linearly with UV doses. These results suggest that RSOM can monitor and quantify UV-induced erythema with high precision.

Nau et al. used RSOM to measure biomarkers such as the mean capillary loop length and mean plexus width to assess the severity of atopic dermatitis (AD)^[103]. Yew et al. also used the method to investigate skin structure and vasculature changes in AD lesions during treatment^[104].

Li et al. used multispectral raster scanning optoacoustic mesoscopy (ms-RSOM) with four distinct wavelengths (532, 555, 579, and 606 nm) to obtain functional information of AD lesions^[105]. ${\mathrm{sO}}_2$ and epidermis thickness decreased in AD lesions after treatment. The same strategy has been used to monitor psoriasis and showed that the structural indicators (epidermal thickness and total blood volume) and functional indicators (${\mathrm{sO}}_2$) after treatment were decreased in a psoriasis lesion^[106].

Vascular indicators in the skin provided from RSOM images have also been used to assess diabetes-related changes in dermal microvasculature [Fig. 4(b)]^[107]. RSOM images of diabetes patients were compared to those of healthy volunteers; they were segmented and compared to derive six label-free PA biomarkers: epidermal thickness, signal density of epidermis layer, dermal blood volume, vessel diameter, small vessel number, and large vessel number. Microvascular parameters differed between diabetic patients and healthy people, with the difference being most significant in the number of vessels $<40\mathrm{\mu }\mathrm{m}$ in diameter.

However, the RSOM technique has the disadvantage of a long scan time. To solve this problem, He et al. developed a fast RSOM (FRSOM); it significantly reduces scanning time and, thereby, minimizes motion artifacts^[108]. The FRSOM system consists of a coaxial optical-acoustic probe that uses a 532 nm laser and a 50 MHz spherically focused ring transducer with an integrated pre-amplifier to increase the SNR. Thus, FRSOM could obtain the same image quality as conventional RSOM with lower laser-pulse energy; this trait allowed an increased laser repetition rate without exceeding the laser safety limit, so scanning can be accelerated. To scan a $4\mathrm{mm}\times 2\mathrm{mm}$ area, the FRSOM took $<15\mathrm{s}$, whereas the conventional RSOM system took 70 s. FRSOM can further minimize motion artifacts by scanning while the subject holds a single breath, so imaging of microvessel structures could be improved. FRSOM was used to visualize and quantify changes in the vessel diameter, blood volume, and vessel density during post-occlusive reactive hyperemia (RH) tests, and it detected significant differences in the maximum volume change, hyperemia ratio, and time to peak; these differences were used to quantify microvascular endothelial dysfunction in smokers and cardiovascular disease patients, compared to healthy individuals.

Systemic sclerosis (SSc) is an immune-mediated disease that involves fibrosis and vasculopathy of the skin and internal organs. SSc has a higher mortality rate than any other rheumatic disease^[109]. PAI has been evaluated as a method to detect changes in skin microvessels to enable early detection and quantitative assessment of these conditions. Masthoff et al. used MSOT Acuity Echo (iThera Medical, Munich) to analyze fingers that were afflicted with SSc^[110]. Each cross-sectional image was $30\mathrm{mm}\times 30\mathrm{mm}$, and the PA spatial resolution was $\sim 190\mathrm{\mu }\mathrm{m}$. Laser pulse repetition rates for single and multispectral imaging were 25 and 5 Hz, respectively, and each examination took about 10 min. For multispectral imaging, five wavelengths $700\le \lambda \le 850\mathrm{nm}$ were used to calculate ${\mathrm{HbO}}_2$, HbR, and total hemoglobin (HbT). SSc patients showed a significant decrease in ${\mathrm{HbO}}_2$ and HbT compared to healthy volunteers [Fig. 5(a)]. Additionally, the progressive SSc patients had lower HbR than healthy volunteers and stable SSc patients [Fig. 5(b)].

RH is a non-invasive method to assess peripheral microvasculature, which, in the simplest form, represents the magnitude of limb reperfusion following a brief period of ischemia induced by arterial occlusion^[111]. Using MSOT imaging, Rodrigues et al. showed that PAI can be used to observe and quantify RH in vivo^[112]. Superficial and deep vascular plexuses were measured at baseline, during occlusion, and after occlusion, and the corresponding HbR, ${\mathrm{HbO}}_2$, HbT, and ${\mathrm{sO}}_2$ were obtained from the MSOT image. Observations revealed that the occlusion collapsed the smaller superficial plexus while inducing stasis at the deeper plexus [Figs. 5(c1) and 5(c2)].

Zhang et al. used a PAI system that uses LEDs (AcousticX, CYBERDYNE, Inc., Tsukuba, Japan) to analyze the severity and treatment response of PWS^[113]. The system consists of a US transducer array with 128 elements and 9 MHz center frequency, and two LED arrays that had 400 µJ pulse energy, one attached on each side of the US transducer. The system provided mean axial resolution of 220 µm and mean lateral resolution of 460 µm, with an imaging depth of up to 28 mm. PWS legions were imaged during treatment using hematoporphyrin monomethyl ether photodynamic therapy. After treatment, PWS levels estimated from the PA signal intensity showed significant decrease.

Ma et al. reported a switchable OR-AR PAM system to image the melanin and vasculature in different skin layers of human skin [Fig. 6(a)]^[114]. A customized adjustable opto-sono objective lens was used to adjust the focal position in the axial direction, and it enabled switching between OR and AR modes. For instance, OR imaging was used when the opto-sono objective lens was positioned with its optical focus of $<1\mathrm{mm}$ deep in the skin, whereas AR imaging was used when acoustic focus was at deeper than 1 mm, where optical beams diverge. Thus, the system can visualize the different layers of human skin with high resolution at all depths, with OR-PAM used in the epidermis (0–180 µm depth) and epidermal–dermal junction (220–550 µm depth) layers, and AR-PAM used beneath the junction layer. Using the system, the melanin concentration in the SB and the vascular density below the epidermal–dermal junction were compared in the back and palm of the hand. Significantly, more melanin was detected on the back of the hand than on the palm, whereas more vessels were observed in the palm than on the back.

Instead of scanning with a fixed focal depth, an autofocusing opto-sono probe has been developed for dynamic confocusing along the skin contour [Figs. 6(b1)–6(b3)]^[115]. The main working component of the autofocusing probe was a liquid lens that exploits electrowetting; it consists of 22% NaCl solution as the conducting liquid and a mixture of glycerol and silicone oil as the insulating liquid. When electrical voltage was applied between the adjacent electrodes, the curvature radius of the interface between the two liquids was altered, so the optical and acoustic beams were refracted simultaneously to the same angle. Using real-time feedback of PA A-line signals, the autofocusing probe adjusted the focus to maximize the PA amplitude and, thereby, increased the SNR at deeper depths compared to fixed-focus scanning.

Ma et al. investigated the use of OR-PAM to measure the melanin component of café-au-lait macules (CALMs) in patients^[116]. The researchers developed a fast linear confocal scanning PA dermoscopy (FLCS-PAD) system, in which a dual PVDF transducer created a laterally elongated acoustic focal field ($\sim 0.8\mathrm{mm}\text{longitudinally}\times 3\mathrm{mm}\text{laterally}$), and the optical focus was linearly scanned within the field using a two-axis galvanometer scanner. The optical depth-of-focus was about 150 µm, which included the epidermal layer well, and enabled reliable quantitative PA measurement in relatively uniform optical and acoustic fields. FLCS-PAD scans of healthy volunteers with two different skin colors showed that the PA amplitudes of SC and SB were higher in dark skin than in light skin. Then a patient with CALM was scanned before and after Q-switched ruby laser treatment, and the epidermal thickness and the melanin concentration were compared. The CALM area had slightly thicker epidermal thickness and significantly higher melanin concentration than the normal skin area, and the treatment only reduced the melanin concentration. The researchers suggested a PAINX score proportional to the epidermal thickness and melanin concentration; the score agreed well with the conventional melanin index measured from digital photographs.

A multiscale confocal PA dermoscopy (MC-PAD) system has been developed using multiple objectives (e.g., $4\times$, $10\times$, and $20\times$) to observe features in CALM and PWS patients^[117]. Increasing the magnification (i.e., NA) by the objective lens refined the resolution for visualization of skin structures, at the cost of reduced imaging depth. Thus, the SC, SB, and dermal layers were imaged using $20\times /0.55\mathrm{NA}$, $10\times /0.3\mathrm{NA}$, and $4\times /0.1\mathrm{NA}$, respectively. The acoustic frequency of the PA signal was divided into different bands in the range of 5 to 90 MHz to optimize the visualization of skin layers. Quantitative results showed higher melanin concentration and thicker SB thickness in CALM-afflicted areas than in normal areas. The PWS study showed that the diameter and density of dermal vessels were greater in PWS areas than in normal areas. PAD analysis on PWS before and after photodynamic therapy (PDT) demonstrated that the diameter, depth, and density of lesion vessels were reduced after PDT but that the improvements were more significant in superficial dermis than in deep dermis^[100].

Wang et al. introduced bifocal dual-wavelength (532 and 1064 nm) PAM to improve the visualization of deep dermal vessels in PWS patients (Fig. 7)^[118]. The conventional 532 nm wavelength could visualize the epidermis and superficial vascular plexus, and the additional 1064 nm wavelength enabled tissue visualization up to $\sim 3\mathrm{mm}$ depth, which includes the deep vascular plexus and hypodermis. The lateral resolutions were 6.6 µm for 532 nm and 46.8 µm for 1064 nm, and their foci were 1.5 mm apart in depth. The epidermal melanin density was lower in PWS than in normal skin, due to abnormal expansion of dermal vessels in PWS. The superficial dermal vessels were significantly increased to about four times that of normal skin, and deep dermal vessels had more branches to transport nutrition to the superficial dermis in PWS than in normal skin. Further, the overall vascular diameter in the superficial dermis tended to be thicker in PWS than in normal skin.

A commercially available PAM system (PASONO-SKIN, Guangdong Photoacoustic Technology Co., Ltd., Foshan, China) has been used to classify melasma by quantifying the clinical indicators for melanin and blood vessels^[119]. The PASONO-SKIN system consists of a 532 nm nanosecond pulsed laser and a ring US transducer (center frequency: 35 MHz, bandwidth: 114%, focal length: 8 mm) and achieves a resolution of 50 µm axially and 7 µm laterally. The conventional classification of melasma has been complex due to various mixtures of pigment abnormalities and vascular formation. In this aspect, PAM measurements on the depth of melanin, mean vascular diameter, and vascular density were used to classify the melasma lesions quantitatively. The depth of melanin determined whether the lesion is the epidermal type (melanin only in the epidermis) or mixed (melanin in both epidermis and dermis), and the two vascular parameters further classified the lesions into M + V (i.e., increased vasculature) or M types (i.e., non-significant increase in the vasculature). PAM provided 3D structure data of melanin and blood vessels in the epidermis and dermis [Fig. 8(a)], and the images were quantified into the PA amplitude (arbitrary units, a.u.) of melanin and mean diameter (μm) and density (%) of vessels. As classified by the metrics, all types of lesions had a significant increase in the PA amplitude of melanin, but only the M + V type showed a significant increase in the vascular diameter and density [Fig. 8(b)]. The results showed that PAM could provide quantitative parameters for classifying melasma type.

He et al. showed that FRSOM can distinguish melanoma lesions from benign nevi by quantifying differences in vascular density, complexity, and tortuosity between the two types of lesions^[120]. FRSOM images (Fig. 9) distinguish small vasculature from large vasculature structures by reconstructing data from two frequency bands: high frequency (40–120 MHz, Fig. 9, green) and low frequency (10–40 MHz, Fig. 9, red). The FRSOM system showed that an irregular dermal vasculature appears in the tumor area and is resolved in the edge of the lesion. However, to separate the effects of the melanin signal, the boundaries must be segmented. Therefore, the authors implemented dual-wavelength (515 and 532 nm) FRSOM to separate the melanin and Hb using spectral unmixing. The dual-wavelength FRSOM took about 25 s to scan an area of $4\mathrm{mm}\times 2\mathrm{mm}$ and provided vascular features without having to segment the boundaries.

When the goal is to distinguish signals from multiple chromophores, a multi-wavelength laser must be used. Berezhnoi et al. developed ultrawide spectrum and bandwidth raster scan optoacoustic mesoscopy (UWSB-RSOM) to visualize the biomolecules in human skin using a laser that has a broad spectral range from visible to short-wave infrared (SWIR) [Fig. 10(a)]^[121]. The UWSB-RSOM system used an Nd:YAG OPO laser to generate $\lambda =650$, 1210, and 1450 nm and fast single-wavelength lasers that operated at 515 and 532 nm. The system had two focused transducers with center frequencies of 55 and 23.9 MHz. The scan with fast single-wavelength lasers and the 55 MHz transducer had spatial resolutions of 8.6 µm axially and 29.6 µm laterally. The scan with the SWIR laser and 23.9 MHz transducer had spatial resolutions of 33 µm axially and 53 µm laterally. Scans of forearm skins of two healthy volunteers obtained high-resolution images of subcutaneous fat, sebaceous glands, hair follicles, and microvasculature. UWSB-RSOM distinguished between melanin and melanoidins in the epidermis, identified lipid distributions in the SWIR range, and visualized water content at 1450 nm [Fig. 10(b)]. The UWSB-RSOM can visualize the skin’s molecular composition, including hairs that extend from the subcutaneous fat to the skin surface and beyond.

Much PAT research to image joint vasculature and pigment has targeted diagnosis of melanoma, which is a malignant tumor caused by the transformation of melanocytes, and that accounts for most deaths from skin cancer. A person with melanoma is most likely to survive when diagnosis is early and treatment is performed before the melanoma metastasizes. To identify the progression stage (according to the Breslow depth), the size and thickness of melanoma lesions must be quantified, and lesions must be treated or excised. Melanoma lesions contain high concentrations of melanin that produce a significantly stronger PA signal against normal skin and, therefore, can be distinguished from surrounding tissue.

Breathnach et al. used a commercial Vevo LAZR (VisualSonics Inc., Toronto, Canada) system with LZ 550 probe (VisualSonics Inc., Toronto, Canada) to scan melanomas^[122]. The system includes a 256-element linear array transducer with a high center frequency of 40 MHz. A tunable OPO laser (680–970 nm, OPOTEK, Carlsbad, USA) can be used with a fiber bundle to irradiate the laser from both sides of the transducer array at 30° to the imaging plane. $\lambda =680\mathrm{nm}$ was selected for single-wavelength PA scans, and five wavelengths (680, 700, 750, 850, and 900 nm) were used for spectral unmixing to distinguish melanin from other chromophores in the tissue. The full multi-wavelength scanning procedure took about 7 min. The PAT images clearly demarcated the location of the melanoma lesion and identified its depth [Figs. 11(a)–11(c)], and both single and multi-spectral PAT presented high correlation coefficients (0.98 for benign and 0.99 for melanoma) with histological thickness. However, PAT measurement overestimated melanoma thickness in all cases, mainly because of shrinkage caused by dehydration of the excised sample during histologic sectioning^[123]. Further, multi-spectral PAI effectively visualized the lesion. The in vivo images were not perfectly matched with the histology results and, thus, had large measurement errors.

Kim et al. imaged an excised melanoma skin sample using a clinical PA/US imaging system that combines a clinically viable research US system (ECUBE-12 R, Alpinion Medical Systems, Anyang, Korea) and an OPO laser (Phocus Mobile, OPOTEK, Carlsbad, USA) to generate three wavelengths (680, 800, and 1064 nm)^[124]. To image the whole lesion, a linear motorized scanning system moved the imaging probe along the elevation direction and then stacked cross-sectional PAT images. The spatial resolutions were 0.2 mm axially and 1.2 mm laterally, and the scanning range was $38.1\mathrm{mm}\times 75\mathrm{mm}$. The PA MAP image melanoma lesion boundary was acquired from displayed areas of melanoma that are not visible to the naked eye, and the thickness from the PA image was consistent with the histopathological result. A follow-up study used the same clinical PA/US imaging system to image melanoma lesions on patients in vivo^[125]. The authors used a lightweight ($<1\mathrm{kg}$) handheld probe scanner, which provided versatile 3D PAT imaging on various locations of melanoma lesions on patients in vivo. The FOV was $38.1\mathrm{mm}\times 31.4\mathrm{mm}$, and the imaging depth was up to 30 mm; data acquisition took about 57 s for multi-wavelength imaging using five wavelengths (700, 756, 796, 866, and 900 nm). Both single (700 nm) and multi-wavelength PA images identified the melanoma lesion boundary in 3D and provided its depth information [Figs. 11(d)–11(h)]. The lesion depth measured from PAI agreed well with histology, but the histopathological depth values were smaller than PAI depth values due to the dehydration and shrinkage of samples after excision. As a special case, they clearly visualized feeding blood vessels near a suspected melanoma in a patient who refused surgery.

PAT has also been evaluated for diagnosis of non-melanoma skin cancers (NMSCs), which constitute most skin cancers. The lesions infiltrate skin tissue to uneven depths, and their boundaries can be difficult to identify, so these carcinomas are treated using Mohs micrographic surgery, which performs repeated excision and examination. This consumes much time, so the boundary and the size of the lesion must be estimated accurately before the surgery.

Kukk et al. modified their integrated OCT-Raman spectroscopy device to perform PA imaging. The modified system uses a single-crystal US transducer (L22-14vX-LF, Vermon, France) that includes 128 elements with a central frequency of 18 MHz^[126]. An OPO laser (SpitLight Compact 400 OPO, InnoLas Laser GmbH, Germany) emits maximal pulse energy at 430 nm wavelength with 7 ns pulse duration and 20 Hz repetition. This system had resolutions of 200 µm axially and 100 µm laterally for US, and 300 µm axially and 200 µm laterally for PA, with a PA imaging depth of up to 5 mm. In vivo tests on human skin nevi demonstrated a good correlation between PA measurements and histological results [Fig. 12(a)], although tissue shrinkage or position-matching errors caused differences between the thickness measurements by histology and by PAI. The authors also imaged skin lesions that were suspected to indicate skin cancers^[127]. PA/US image measurements of lesion thickness correlated strongly (${\mathrm{R}}^2=0.93$) with histological measurements, especially for lesions that had histological thickness between 0.2 and 3 mm. These results show the potential of PAT to assist in determining the extent of excision and improve the accuracy at which skin cancers can be diagnosed.

Attia et al. imaged skin lesions suspected to be non-melanoma carcinomas using an inVision 512-echo MSOT system (iThera Medical GmbH, Munich, Germany)^[128]. The FOV of the system was $10\mathrm{mm}\times 10\mathrm{mm}\times 12\mathrm{mm}$ and achieved a nearly isotropic resolution of 80 µm, which allowed precise measurement of the tumor dimensions. Ten optical wavelengths (700, 715, 730, 760, 780, 790, 800, 825, 850, and 900 nm) were used for multi-spectral measurement, and the whole imaging procedure took about 5 min per patient. Tumor dimensions measured using the MSOT system correlated well with histological measurements (intraclass correlation coefficient is $>0.7$), Also, the system could provide vascular features of non-melanoma skin tumors. Chuah et al. used MSOT for a further study of different types of NMSCs on different skin phototypes^[129]. Dimensions of the NMSCs could be precisely imaged regardless of the skin phototype because MSOT can distinguish melanin from ${\mathrm{HbO}}_2$ and HbR. Measurements of shallow tumors near the skin surface showed relatively large errors, but the overall results from the study showed that the depth estimated using PAI agreed well with the histology results. These results indicate that PAI has the potential to increase the precision of tumor ablation and to decrease the recurrence rates of NMSCs.

PAI can also be used to differentiate between benign and malignant skin tumors, by exploiting its ability to distinguish between blood vessels and melanin. For instance, skin tags are commonly benign skin lesions, but they can be malignant lesions like squamous cell carcinoma, BCC, or melanoma. Therefore, a sensitive diagnostic modality is needed to distinguish the malignant lesions accurately. Fredman et al. evaluated reflectance confocal microscopy (RCM), OCT, and MSOT to analyze skin tags to provide lesion dimensions and micromorphological distributions of vascularization and melanin^[130]. The MSOT imaging used MSOT Acuity (iThera Medical GmbH, Munich, Germany), in which the US transducer array had 256 elements with 8 MHz central frequency and optical wavelengths of 700, 730, 760, 780, 800, 850, and 875 nm. The FOV was $15\mathrm{mm}\times 15\mathrm{mm}\times 20\mathrm{mm}$, and each scan took about 5 s. Histopathological analysis and dermoscopy measurements commonly detected no malignancy in all lesions, and all three imaging methods also showed features that correspond to benign lesions. OCT and RCM gave detailed information mainly about vessel distribution and the nest of melanocytes, respectively, but MSOT provided integrated information about both features. Blood vessels located below the melanin signal were clearly distinguished by their ${\mathrm{HbO}}_2$ signal; this result indicates that melanin was not present in the lower layer of the dermis. Von Knorring et al. added high-frequency ultrasound (HFUS) imaging to RCM, OCT, and MSOT, and imaged patients that had either benign or malignant tumors^[131]. PAI found abundant blood vessels inside most of the malignant tumors; this result indicates that the malignancy of the lesion was closely associated with the converging network of blood vessels and melanin [Fig. 12(b)].

In the papers reviewed here, both imaging depth (Fig. 13, left) and lateral FOV (Fig. 13, right) show a trade-off relationship with spatial resolution. Least squares line fitting showed an $\text{imaging}\text{depth of}\approx 92.6\times \text{spatial resolution}$, which agrees with previous reports^[69,132]. Similarly, $\text{lateral}\mathrm{FOV}\approx 236\times \text{spatial resolution}$. These relations can provide guidance on selecting an appropriate PAI modality for the skin imaging application of interest, depending on the desired precision and the size of the target.

Systems for skin imaging have a range of specifications PAM (Table 1), PAMes (Table 2), and PAT (Table 3). Generally, PAM and PAMes offer much higher axial/lateral resolutions ($\sim 5$ to 50 µm) than PAT ($\sim 200\mathrm{\mu }\mathrm{m}$). PAM can be implemented in either OR-PAM or AR-PAM, depending on the location of optical and acoustic foci. OR-PAM uses a focused light beam to achieve the best lateral resolution ($\sim 7\mathrm{\mu }\mathrm{m}$), which achieves detailed imaging in the epidermal layers (SC, SB) and superficial dermal layer. To image deep vasculature in the dermal layer, OR-PAM and AR-PAM are often used switchably, or dual-wavelength 532/1064 nm foci can be used.

AR-PAM techniques can be implemented with almost the same specifications as PAMes because they are both built on a US focus. However, PAMes systems, often known as RSOM, offer a fine axial resolution of $<10\mathrm{\mu }\mathrm{m}$ thanks to the ultrawide US bandwidth ($\sim 10$ to 120 MHz) and can also distinguish vessel sizes by dividing the bandwidth range. Both PAM and PAMes can clearly identify the vasculature or melanin structures in each skin layer, but multi-wavelength laser excitation is not widely used.

Especially for PAM, the lateral resolution coarsens as the optical wavelength increases, so $\lambda >532\mathrm{nm}$ is not preferred. Thus, the signals of microvasculature are not distinguished from those of melanin spectrally but instead are distinguished to a certain extent by identifying the skin layer that they occupy. Instead, PAM can provide quantitative analysis of detailed skin structures such as the depth, diameter, and density of microvessels and, thereby, enables sensitive detection of abnormalities in the skin layers.

PAT has much coarser spatial resolution PAM or PAMes methods, but it allows large transverse scan areas ($\sim 30\mathrm{mm}\times 30\mathrm{mm}$) with imaging depth of $<30\mathrm{mm}$, whereas those of PAM and PAMes are limited to $\sim 5\mathrm{mm}\times 5\mathrm{mm}$ and $<5\mathrm{mm}$. Therefore, PAT cannot readily distinguish fine capillary structures, but vessels as small as arterioles or venules can be consistently identified, imaged, and analyzed to track changes. Furthermore, PAT systems usually deploy a wavelength-tunable laser source, so multi-wavelength measurements enable Hb to be distinguished spectrally from melanin chromophores. For example, MSOT systems provide quantitative measurement of functional biomarkers (e.g., HbR, ${\mathrm{HbO}}_2$, and ${\mathrm{sO}}_2$) using NIR excitation wavelength characteristics to each chromophore, thereby providing an objective and quantitative diagnosis of disease activities.

The three PAI modalities use different analysis metrics related to skin vasculature (Table 4) or vasculature plus pigment structures (Table 5). The differences in the diagnostic information can be summarized as follows: PAM and PAMes focus on analyzing the detailed micro-scale vascular or melanin structures layer by layer, whereas PAT focuses on measuring the dimensions of a whole lesion or relative concentrations of Hb or melanin. In skin diseases including allergies and atopy, accompanying superficial changes such as inflammatory reactions, PAM or PAMes could be useful for detecting changes in skin thickness and vascular structure and determining the effectiveness of treatments. They can also be used in other skin pigment disorders such as PWS to analyze and quantify the characteristic features of pigment and vasculature. For skin diseases with macroscopic lesions such as melanoma, it is important to assess a whole lesion dimension and determine whether a lesion is benign or not. In this sense, PAT can provide a full scan of an entire lesion for a holistic view of melanin and vessel distribution changes reaching the deep subcutaneous layer under the skin. It is relatively easy to apply multi-wavelength imaging, which can be used to specifically differentiate between melanin and vascular signals. The different analytic roles of PAI modalities are most apparent when looking at studies on target disease common to all three modalities. For instance, in PWS imaging studies, PAM had the advantage of imaging capillaries in a small FOV and can quantify the density and depth of blood vessels in lesions. Using the microstructural information, PAM can also provide quantitative diagnostic indices that are sensitive to alterations in the shallow skin layers. PAT can cover large skin areas that include almost an entire PWS lesion, whereas PAM can only analyze a small sample area of a lesion. PAT has poor spatial resolution and, therefore, cannot identify and analyze vascular geometry, but it can distinguish lesions by considering their relative signal intensities and indexing their severity.

While reviewing the recent literature, we found that most of the clinical skin PAI research was based on vascular imaging, and not many weights were put on the melanin analysis. PAI is able to visualize the non-superficial morphology and distribution of skin pigment, and we think it is worth investigating its clinical potential to provide a critical triaging modality for various pigmentation diseases such as melanoma or hypopigmentation disorders. To expand the range of application, more skin biomarkers having optical contrast (e.g., collagen and lipid) should be discovered with the use of multi-wavelength laser sources supporting up to the NIR-II range.

In terms of clinical application, niche positioning in the current clinical flow would be important. Generally speaking, many skin diagnoses are made from qualitative visual inspection often aided by optical techniques (e.g., Wood’s lamp) or histologically analyzed by biopsy. In this aspect, PAI has great potential to fit between the two diagnostic methods, providing detailed visualization of non-superficial skin structures and quantitative analysis of the structural features. Still, the reliability and repeatability of PAI modalities should be validated in comparison to conventional techniques, and especially the effect of skin color on the PAI results should be investigated thoroughly. Due to the nature of optical and acoustic beams, the existence of thick scars, scabs, or calluses could also limit the application of PAI. Furthermore, PAI systems should be designed with extra care for hygiene because they require direct contact with the skin.

Overall, PAI technology can diagnose human skin non-invasively and potentially meet a variety of needs in dermatology such as staging, progression, and treatment monitoring. PAI also provides quantifiable metrics by analyzing the PA signal amplitudes that represent Hb or melanin contents observed in various scales using PAM, PAMes, or PAT. By adopting a fit-for-purpose PAI system, dermatologists will be able to effectively diagnose and test skin diseases, and reduce the number of repetitive treatments.