1. Introduction

Reconstruction-based Acoustic-Resolution Photoacoustic Microscopy (AR-PAM) systems offer several advantages over traditional point-scanning AR-PAM systems that use focused transducers [1–13]. In particular, reconstruction-based methods utilize depth information more effectively, providing a wider depth of field (DOF) and enhanced signal-to-noise ratio (SNR), thereby enabling high-resolution imaging across a broader range of depths [5,14–16]. Physical system development has primarily focused on improving transducer configurations, with recent studies implementing high-NA transducers and focused transducers with holes for fiber optic illumination [17–19]. Complementing hardware advances, significant progress has been made in signal processing techniques, including the implementation of compressed sensing and fluence compensation methods combined with virtual point detectors, as well as advanced beamforming techniques such as Delay-Multiply-and-Sum (DMAS) for improved image reconstruction [3,5,11]. These technical advances have enabled reconstruction-based AR-PAM to find applications in various fields such as medical imaging, biological tissue analysis, and microvascular imaging [2,4,6,8,13,19–29]. However, they still fall short of reaching the imaging depths required to cover the full range achievable by photoacoustic principles due to the inherent limitations of the transducer's restricted focal region [12].

To achieve imaging over a greater DOF than AR-PAM systems, photoacoustic tomography (PAT) systems have been developed [2,4,30]. Conventional PAT systems typically employ array transducers, enabling high-speed volumetric imaging [19,25,26]. However, the fabrication of array transducer-based systems is complex and costly due to the need for precise element alignment, numerous data acquisition system, synchronization, and signal processing. Although array transducers can be designed with an extended focal length to accommodate deep imaging, this approach inherently limits sensitivity to signals originating from out-of-focus regions, thereby restricting overall imaging flexibility [21,31]. Meanwhile, a Fabry-Pérot (FP) sensor-based PAT system acquires PA signals using an interrogation beam, where the diameter of the interrogation beam determines the lateral resolution, enabling high-resolution 3D imaging similar to array transducer-based systems [24,32,33]. However, FP sensor-based PAT systems detect ultrasound signals through an optical interferometric method, which may introduce sensitivity limitations in deep tissue imaging [32]. In particular, the multi-layered optical structure of the FP sensor can lead to signal distortion or reduced detection efficiency due to acoustic impedance mismatches [34]. Additionally, shear stress within the FP sensor medium can further degrade resolution, resulting in values exceeding the theoretical limit defined by the interrogation beam diameter [35].

To address the limitations of conventional AR-PAM systems, we propose a needle hydrophone (NH)-based AR-PAM system that enables deep imaging with improved resolution consistency. Compared to focused transducer-based AR-PAM, NH-PAM provides an extended DOF due to its wide acceptance angle, enabling imaging across a broader axial range with reduced dependence on mechanical refocusing. Furthermore, NH-PAM offers a simpler and more robust design compared to FP-PAM and array transducer-based PAT systems, eliminating the need for precise optical alignment or complex fabrication processes. Also, by leveraging a broader bandwidth, NH-PAM has the potential to improve signal-to-noise ratio (SNR) in deep tissue imaging.

To validate the proposed system, we conducted phantom imaging and ex Ovo chick embryo imaging. By imaging tissue-mimicking phantoms, we verified a depth of focus greater than 20 mm and demonstrated that lateral resolution comparable to the diameter of the needle hydrophone could be achieved at various depths [36]. Additionally, we performed imaging of a chick embryo covered with a 1 cm-thick intralipid hydrogel and successfully visualized organ structures within the embryo, demonstrating the system’s capability for deep imaging in vivo-like conditions [37–39].

Like other AR-PAM systems, our approach enhances image resolution and SNR through reconstruction processes. Typically, post-processing methods based on SAFT are employed, with additional weighting factors such as coherence factor (CF) used to further improve image quality [1,14]. Another approach involves using the sensor's impulse response function (IRF) as a weighting function in image reconstruction [15]. However, in previous studies, IRFs were obtained through simulations, which may not accurately reflect the sensor’s actual response. To overcome this limitation, we experimentally acquired the IRF and applied it to image reconstruction [40]. As a result, side artifacts were significantly reduced compared to SAFT, and resolution consistency across different depths was improved.

2. Methods

2.1. System design

Figure 1 illustrates the schematic of the needle hydrophone photoacoustic microscopy system. The optical source is provided by a laser system (Radiant 532 LD, OPOTEK) with a wavelength of 700 nm and a pulse repetition rate of 20 Hz. To minimize the loss of the optical source at the prism due to the needle hydrophone (HNC-400, Onda), the optical source first passes through an axicon, creating a doughnut-shaped beam. The beam is then expanded and focused using a concave and a convex lens to produce a broad incident area. To ensure stable signal acquisition, the optical system and the needle hydrophone were mounted on an x-y stage (VT-80, PI) capable of auto-scanning. Since the optical source area and the needle hydrophone move together, the optical source area remains consistent relative to the needle hydrophone. The incident energy density on the sample surface is maintained below 5 mJ/cm². The x-y stage scanned the imaging area using a raster scan. The total size of the main optical system is 70 × 70 × 350 mm, as indicated by the green dashed line in Fig. 1(b).

 

Fig. 1. Schematic of the needle hydrophone PAM system. AX, axicon, CC, concave lens, CV, convex lens, NH, needle hydrophone, Pre-Amp, pre-amplifier, WT, water tank, SMC, step motor controller, Amp, amplifier. (b) Picture of scanning part. (c) Tip part of needle hydrophone. (d) The surface of needle hydrophone. Scale bars, 200 µm in (c-d).

Download Full Size | PDF

The photoacoustic (PA) signal generated by the sample is detected by the needle hydrophone, which has a frequency detection range of 1-10 MHz (±6 dB). The signal is first amplified using a 20 dB preamplifier (AG-2010 Preamplifier, Onda), followed by a DC block, and then further amplified by 30 dB using an ultrasound pulser-receiver (Olympus 5073, Olympus) with a 1 MHz high-pass filter (EF517, Thorlabs). A data acquisition (DAQ) computer (PCI-5124, NI) subsequently records the amplified and filtered signal with a sampling rate of 200 MS/s.

The matching layer between the needle hydrophone and the imaging sample is designed using a plastic water tank and transparent vinyl wrap. The water tank, which has a 72 mm diameter hole, is covered with transparent vinyl wrap. To remove any air layers, ultrasound gel is applied to the target before placing it onto the vinyl-covered water tank.

2.2. Image reconstruction

Unlike focused transducers, needle hydrophones do not have a physical focal point [1]. The needle hydrophone is a device that can receive a broad range of signals without a specific physical focus when collecting photoacoustic signals, indicating that the signal is not focused onto a single point, but rather spreads broadly without directional convergence. Therefore, the concave-shaped distorted signals observed with a focused transducer acting as a virtual point detector are not present [12]; instead, only convex-shaped distorted signals are observed. Due to this characteristic, the observed signals contain only temporal information without directional information. Thus, the observed signal can be assumed to propagate uniformly to all points equidistant from the source in spherical coordinates.

First, the signal obtained at position r₀ with θ=0 and φ=0 is defined as S(r₀,0,0). This signal can be assumed to propagate uniformly in all directions, resulting in the following form where it is scattered in multiple directions:

Here, the polar angle θ represents the acceptance angle of the needle hydrophone, and the azimuthal angle φ ranges from 0 to 2π. By repeating the process of scattering the signal in multiple directions and then superimposing these scattered signals, we can reconstruct the signals using the SAFT. This process can be expressed with the following equation:

Here, N_θ and N_φ represent the number of discretized steps of θ and φ, respectively.

Another reconstruction method involves using the CF. The CF yields a value between 0 and 1 by phase-matching PA signals with bipolar characteristics. This value can be multiplied by the SAFT result to improve signal quality. Reconstruction using the CF is expressed as follows:

The final signal can be obtained by multiplying the CF value by the SAFT result:

By multiplying the 3D data reconstructed through the SAFT process with the CF calculated for each pixel on a pixel-by-pixel basis, distorted signals can be eliminated.

In addition, signal reconstruction can be performed using the IRF of the needle hydrophone as a weighting factor in the SAFT process:

Angular aliasing can lead to incorrect wavefront summation in SAFT-based reconstruction, distorting the final image. To prevent this, the angular resolution was set to below than 0.2° at 20 mm depth from NH tip, ensuring that signals from adjacent scanning positions were accurately interpolated before integration.

The experimental IRF (MIRF) was obtained by focusing a 700 nm pulsed laser with a pulse repetition rate of 20 Hz into a water tank and recording the PA response using the NH (Fig. 2(a)). The NH was positioned 5 mm from the optical focal point, and the PA signal was acquired. To obtain spatially resolved IRF data, a 20 mm × 50 mm area was scanned in a raster pattern with a step size of 100 µm, ensuring accurate mapping of the hydrophone’s impulse response. The MIRF was then deconvolved from the acquired signals to correct for the NH’s frequency response.

 

Fig. 2. The scheme of image reconstruction with IRF data. (a) Experimental method of obtaining IRF of needle hydrophone. PL, pulse laser, CV, convex lens, NH, needle hydrophone, WT, water tank (b) The B-mode data from scanning divided each frequency (1-3, 3-6, 6-10 MHz) using FFT and inverse FFT. The filtered B-mode data was reconstructed using SAFT, CF, and IRF. Then each reconstructed data was assembled. Scale bars, 4 mm in (a-b).

Download Full Size | PDF

For the simulation of the hydrophone impulse response function (SIRF), the k-Wave toolbox in MATLAB was used to replicate experimental conditions (Fig. 2(d-f)). The simulation domain was set to 20 × 50 mm (800 × 2000 grid points) with a lateral resolution of 0.02 mm and axial resolution of 0.025 mm, covering the hydrophone’s scanned area (20 × 50 mm). The propagation medium was defined with a sound speed of 1500 m/s, an absorption coefficient of 0.75 dB/(MHz^y·cm), and a power law exponent of 2. The time step was set to 5 ns, with a total of 6000-time steps for stable wave propagation. The sensor’s effective size was 0.4 mm, and its frequency response was set to 5.5 MHz center frequency with 180% bandwidth.

Fourier transform was applied to decompose the acquired PA signals into different frequency bands, allowing frequency-dependent hydrophone sensitivity variations to be accounted for during reconstruction. This was necessary because, as shown in Fig. 2(b), the extent to which both lateral and depth signals are detected varies across different frequency bands, and the hydrophone exhibits different sensitivity at each frequency. Therefore, based on acoustic wave propagation in water, we categorized the signals into three bands: 1–3 MHz for wavelengths larger than the NH diameter, 3–6 MHz for wavelengths comparable to the NH diameter, and 6–10 MHz for wavelengths smaller than the NH diameter. Similarly, the impulse response function (IRF), which was used as a weighting function, was also divided into these frequency bands before proceeding with reconstruction (see Fig. S1 of Supplement 1). Finally, the frequency-specific reconstructed results were combined to generate the final output. By incorporating a MIRF into the reconstruction process, this study further refines NH-PAM's imaging performance, ensuring enhanced lateral resolution stability and reduced artifacts compared to conventional SAFT-based methods.

3. Results

3.1. Phantom imaging

To validate the performance of the NH PAM system, we imaged a 100 µm tungsten wire and evaluated the lateral and axial resolution, as well as the SNR, at different depths through reconstruction. First, a mold was prepared to position the tungsten wire at 5 mm depth intervals, and 1% (w/w) intralipid mixed with 10% hydrogel was poured into the mold to mimic tissue properties (Fig. 3(a)) [36]. The mold was then hardened at room temperature for 24 hours. The resulting phantom measured 70 × 70 × 50 mm, with tungsten wires placed at depths ranging from 10 mm to 30 mm from the surface. Subsequently, the region of 70 × 70 mm in the x-y plane was scanned using raster scanning.

 

Fig. 3. The tungsten wire B-mode data. (a) Top-view picture of a tungsten wire in intralipd hydrogel. (b-h) B-mode image from raw data and image reconstructed using SAFT, SAFT + MIRF, SAFT + SIRF, SAFT + CF, SAFT + CF + MIRF, and SAFT + CF + SIRF, respectively. Scale bars, 10 mm in (b-h). (i-k) The graph of lateral resolution, axial resolution, and SNR of the SAFT (blue line), SAFT + MIFR (green line), SAFT + SIRF (green dashed line), SAFT + CF (yellow line), SAFT + CF + MIRF (purple line), and SAFT + CF + SIRF (purple dashed line), respectively. N = 10.

Download Full Size | PDF

Figure 3(c-h) display the B-mode images obtained using SAFT, SAFT + MIRF, SAFT + SIRF, SAFT + CF, SAFT + CF + MIRF, and SAFT + CF + SIRF, respectively. The performance metrics related to resolution and SNR at different depths are presented in the graphs shown in Fig. 3(g-i).

In the B-mode image before reconstruction, the tungsten wire exhibited convex-shaped artifacts as well as reverberation artifacts along its axial direction (Fig. 3(b)). The convex-shaped artifacts, after applying reconstruction, were reduced, and the signal intensity at the actual position of the tungsten wire was amplified. In both SAFT + MIRF (Fig. 3(d)) and SAFT + SIRF (Fig. 3(e)), noticeable axial artifacts appear below the tungsten wire (see yellow arrow in (d)), likely originating from partial reflections or reverberation that are not entirely eliminated during reconstruction. When only the IRF-based reconstruction was applied, some noise and spurious echoes were still interpreted as valid signals and were not completely removed. In contrast, applying CF more effectively suppressed these noise components. Additionally, when comparing the use of IRF to CF, the B-mode images with CF applied decreased artifacts in the lateral direction.

Figure 3(i) shows the lateral resolution at depths ranging from 10 to 30 mm. The SAFT data (blue line) exhibits generally lower resolution, similar to SAFT + SIRF (green dashed line), whereas SAFT + MIRF (green line) provides a modest improvement over both SAFT and SAFT + SIRF. However, when CF is applied (yellow and purple lines), the lateral resolution improves even further. Although SAFT + CF + MIRF yields better overall lateral resolution than SAFT + CF, the latter slightly outperforms it at 30 mm depth, presumably because signal attenuation is more pronounced at greater depths. The detailed numerical values are provided in the Supplement 1 (see Table. S1 of Supplement 1).

Regarding axial resolution (Fig. 3(j)), one notable observation is that SAFT + MIRF (green line) and SAFT + SIRF (green dashed line)—both without CF—maintain relatively consistent resolution across the entire depth range. This suggests that using SAFT combined with the IRF, without CF, can adequately account for the laser beam profile and hydrophone response to preserve uniform axial resolution. On the other hand, the curves with CF applied (yellow and purple lines) fluctuate more with increasing depth, indicating that while CF strongly suppresses noise and enhances lateral resolution, it does not specifically address or mitigate axial reflections or reverberations.

From an SNR perspective (Fig. 3(k)), applying CF (yellow and purple lines) substantially increases SNR overall. However, the deep-region signals did not recover as effectively as shallow signals under CF. In other words, although CF clearly improves noise suppression and raises the SNR, it may not provide uniform reconstruction across all depths, thereby limiting its effectiveness in scenarios requiring broad depth coverage. In contrast, applying the IRF to SAFT yielded higher SNR at deeper positions compared to SAFT alone. Notably, MIRF proved more consistent than SIRF, likely because the simulated IRF does not fully capture the actual characteristics of the needle hydrophone.

We attribute the gradual SNR decrease after a slight rise at around 20 mm to two key factors. First, the laterally restricted IRF acquisition covers most of the hydrophone’s acceptance angle in shallower regions, but may fail to encompass all relevant angles at greater depths. While extending the lateral measurement range could enhance reconstruction accuracy, it would also significantly increase computational cost and processing time. Second, photoacoustic signals generated at deeper layers generally have reduced amplitudes due to tissue attenuation, leading to lower SNR upon reconstruction.

Overall, the SAFT + MIRF approach delivers a balanced improvement in lateral resolution, axial resolution, and SNR, consistently performing well at various depths. In contrast, while CF substantially boosts lateral resolution and SNR, it can introduce instability in axial resolution at larger depths. Therefore, if stable axial resolution over a wide depth range is paramount, one should consider applying CF with caution.

Although a 100 µm tungsten wire was employed, the measured lateral resolution was lower, largely because the 400 µm diameter of the needle hydrophone sets a physical limit. Moreover, since the tungsten wire forms a two-dimensional linear structure rather than a true point source, the reconstruction efficacy may be restricted in directions parallel to the wire.

Having characterized the basic depth-dependent resolution and SNR of the proposed system using a tungsten wire, we proceeded to evaluate its reconstruction algorithms on a more complex phantom such as metal bee phantom (Fig. 4(a)). Unlike the simple linear geometry of the tungsten wire, this phantom feature multiple surfaces and curvatures, making it possible to observe lateral and axial artifacts or reconstruction errors more clearly. Furthermore, the entire phantom—measuring 27.5 mm high—was embedded in 37.5 mm of intralipid gel to tissue properties. The 70 × 70 mm region in the x-y plane was scanned using raster scanning. This setup allows us to assess system performance in concurrently visualizing structures located at different depths.

 

Fig. 4. Bee shape metal phantom MAP images. (a) Metal phantom shape. (b-d) MAP image from SAFT, SAFT + MIRF, and SAFT + SIRF, respectively. (e-g) Zoom-in MAP image from (b-d) at white dashed box. (h-j) MAP image from SAFT + CF, SAFT + CF + MIRF, and SAFT + CF + SIRF in logarithmic scale, respectively. Scale bars, 10 mm in (b-d, h-j) and 2 mm in (e-g).

Download Full Size | PDF

Figure 4(b–d) present maximum amplitude projection (MAP) images obtained using SAFT, SAFT + MIRF, and SAFT + SIRF, respectively, with the region within the white dashed rectangle enlarged in (e–g). Near the upper wing area, SAFT, SAFT + MIRF, and SAFT + SIRF exhibit comparable resolution and SNR. However, in the deeper region marked by the white dashed box (Fig. 4(e–g)), the yellow arrow highlights a marked difference in signal intensity: as previously noted in the tungsten wire experiment, when CF is not applied, SAFT + MIRF yields the strongest signal in deeper sections.

Figure 4(h–j) show results for SAFT + CF, SAFT + CF + MIRF, and SAFT + CF + SIRF. Under CF, certain points (blue arrow) appear overly bright, making it difficult to discern structure using the same color map without CF (see Fig. S2 of Supplement 1). Hence, a logarithmic scale was used, rendering the MAP images more visually akin to the non-CF data. However, log scaling has inherent drawbacks, such as making it harder to distinguish differences among strong signals and exaggerating weak signal regions, thereby highlighting noise or artifacts more than in linear scale. Consequently, these factors can reduce interpretational clarity. Therefore, caution is advised when applying CF to analyze signals spanning a wide depth range.

3.2. Ex Ovo imaging

To further evaluate the proposed system’s performance in a biologically relevant scenario, we conducted ex Ovo photoacoustic imaging on a chick embryo (FOV: 33 × 45 mm). Figure 5(a) and 5(b) present photographs of a 11-day-old chick embryo and the 10 mm-thick intralipid-based turbid layer placed atop the embryo, respectively. The dashed rectangle in Fig. 5(a) outlines the region shown in the close-up.

 

Fig. 5. Chick embryo PA images. (a) The picture of 11 days chick embryo. (b) The picture of the turbid layer on the chick embryo. (c) SAFT MAP image. B, brain, E, eye, H, heart, AL, allantois, S, somites, FL, forelimb, I, intestine, HL, hindlimb. (d) SAFT + MIRF MAP image. (e) SAFT + SIRF MAP image. (f-h) SAFT, SAFT + MIRF, SAFT + SIRF B-mode image at white dashed box in (c), respectively. Scale bars, 4 mm in (c-h).

Download Full Size | PDF

After placing the turbid layer, the region of interest was raster-scanned, and the resulting MAP images using SAFT, SAFT + MIRF, and SAFT + SIRF are displayed in Fig. 5(c)–5(e), respectively. Interestingly, the anatomical structures of chick embryo are visible [39]. The green arrowhead highlights somites, while the dashed box indicates the B-mode cross-section presented in Fig. 5(f-h).

In the B-mode images (Fig. 5(f-h)), we can observe the somites depth profile more clearly. Compared to SAFT (Fig. 5(f)), SAFT + MIRF (Fig. 5(g)) appears to yield a slightly higher signals (blue arrow), suggesting that the experimentally measured IRF helps recover the deeper signal with less artifact. By contrast, SAFT + SIRF (Fig. 5(h)) shows a somewhat weaker somites signal in that same region, likely due to the SIRF not fully capturing the needle hydrophone’s actual frequency response and acceptance angle.

Under CF, the eye region signal was excessively amplified, making it difficult to observe adjacent structures (see Fig. S3 of Supplement 1). As in our earlier experiments, we applied a logarithmic scale to better visualize the data; although the log-scale images appeared similar to those without CF, they still suffered from limited differentiation among high-intensity signals.

Overall, these ex Ovo results highlight a trade-off between maximizing SNR and maintaining clarity across different anatomical structures. While CF can greatly amplify certain signals—such as the eye and brain region—its tendency to overshadow neighboring features poses challenges when visualizing complex, multi-layered tissues. Employing a logarithmic scale partly mitigates signal saturation, but makes it difficult to distinguish adjacent strong signals. In contrast, the MIRF-based reconstruction consistently yielded higher signal levels at deeper regions without excessively amplifying localized features, suggesting it may be more appropriate for imaging scenarios requiring uniform resolution and SNR across a broad depth range. Moving forward, an adaptive approach—one that selectively applies CF or modulates the IRF weighting depending on local signal intensity—could help overcome the limitations observed here, offering a more versatile reconstruction framework for intricate biological samples such as chick embryo.

4. Discussion

In this study, we developed an AR-PAM system incorporating NH and evaluated its performance through wire phantom, metal bee phantom, and ex Ovo imaging of chick embryo. The results demonstrate that applying an IRF—especially MIRF—can improve image quality over standard SAFT-based reconstructions, notably at deeper regions. However, certain trade-offs emerged when combining IRF with a CF, most evidently in how they affect resolution and SNR across varying depths.

A key discrepancy arose between the wire phantom (Fig. 3), where CF substantially improved lateral resolution and SNR, and subsequent metal phantom or ex Ovo data, where CF did not always yield a proportional improvement in deeper areas [41]. This contrast is primarily attributed to two factors: First, complex scattering or absorption in more realistic phantoms attenuates signals with increasing depth, limiting the effectiveness of CF’s phase-coherent summation. Second, CF can over-amplify localized strong reflectors—such as metal edges or certain embryonic structures (e.g., the brain)—thereby obscuring adjacent weaker signals. Consequently, while CF excels at boosting SNR in uniform or high-contrast scenarios (like the wire phantom), it may be less beneficial in more heterogeneous samples. In contrast, MIRF reconstruction provided a more balanced improvement across depths by accurately accounting for the NH’s acceptance angle and frequency response.

Compared to FP sensors, which often require precise optical calibration and suffer from sensitivity variations due to environmental fluctuations, NH-PAM offers greater robustness and simpler alignment. However, the primary trade-off is lateral resolution: while FP-PAT achieves sub-100 µm resolution, NH-PAM is constrained by the hydrophone diameter (400 µm in our setup), making it more suitable for applications prioritizing imaging depth over lateral precision [42]. This broader depth of focus is beneficial in scenarios where a large axial range must be visualized without refocusing. However, the deterioration of lateral resolution at deeper regions compared to shallower depths remains a persistent issue. This effect arises because signals originating from deeper regions undergo greater distortion and attenuation before reaching the hydrophone. Furthermore, the broader acceptance angle of NH introduces potential blurring at larger depths, necessitating an approach that compensates for these effects, such as spatially adaptive reconstruction techniques or frequency-specific processing. To further contextualize the performance of our system, Supplement 1 Table S2 compares the features and specifications of the current system with those of the previous AR-PAM and PAT systems.

Despite demonstrating imaging depths up to ∼30 mm, our system still faces several limitations. First, attenuation and scattering challenges remain in thicker, more optically turbid media, which can degrade signal quality at greater depths. Second, since the NH is mounted by drilling a hole into the prism, light loss occurs at the hole region, reducing illumination efficiency, particularly at greater depths. Lastly, computational complexity increases when applying MIRF-based reconstruction, which may limit its feasibility for real-time imaging applications.

In addition, we attempted in vivo imaging of a nude mouse abdomen; however, obtaining interpretable results proved challenging (see Fig. S4, Supplement 1). The primary obstacle was motion artifacts caused by breathing and heartbeats, which significantly disrupted the mapping process during raster scanning by preventing spatial coherence between adjacent scan points. To ensure the effective operation of our system in vivo, it is essential to implement motion artifact compensation algorithms for real-time imaging or to replace the current laser source with a high-repetition-rate to substantially reduce scanning time.

Moving forward, several enhancements are envisioned. First, reducing the needle hydrophone diameter or employing alternative materials (e.g., PMN-PT) could further improve lateral resolution and detection sensitivity [43–45]. Second, expanding the lateral range over which the IRF is measured could mitigate inaccuracies at greater depths but must be weighed against increases in computational load. Lastly, integrating higher-repetition-rate laser diodes directly alongside the NH has the potential to reduce energy loss and scanning time, making real-time in vivo imaging more feasible [46–48]. These refinements would not only enhance NH-PAM’s imaging capabilities in preclinical settings but also pave the way for its translation into clinical applications, particularly in fields requiring deep-tissue visualization with stable resolution across extended imaging depths.

5. Conclusion

In this study, we developed an NH-PAM system to enhance imaging depth and resolution consistency in acoustic-resolution photoacoustic microscopy. By incorporating MIRF-based reconstruction, we demonstrated a depth-of-focus exceeding 20 mm and an imaging depth beyond 30 mm while maintaining a lateral resolution comparable to the NH diameter (∼400 µm). The proposed system was validated through phantom imaging and ex Ovo chick embryo imaging, confirming its feasibility for deep-tissue imaging. The integration of MIRF-based reconstruction significantly improved lateral and axial resolution while reducing side artifacts, demonstrating NH-PAM’s potential as a high-resolution, deep-tissue imaging modality.