1.

Introduction

Surgical margin assessment (SMA) is considered to play a significant role in reducing the local recurrence rate of a variety of cancer types.^1–7 In the histological analysis performed for SMA, morphological features of cell nuclei provide critical information to identify cancer cells through microscopic examination.^8,9 Currently, postoperative histological analysis of excised and stained specimens is widely performed and regarded as the standard procedure. However, this analysis typically includes laborious and time-consuming processes before imaging, such as tissue processing, embedding, sectioning, and staining. Comprehensive initial detection is another concern. In the case of breast conserving surgery, for instance, over 20% of patients are required to have a second surgery based on the results of postoperative analysis.^10–18 Clearly, this undesirable re-excision surgery adversely affects the patients’ quality of life by increasing health care costs, adding physical and psychological burdens, and potentially delaying adjuvant therapy. To avoid re-excision surgeries, an intraoperative SMA as reliable as the conventional postoperative SMA has been highly sought. One rapid sectioning technique, frozen sectioning along with staining, can be performed intraoperatively. However, this technique has not been widely used, mainly due to the technical difficulty of freezing adipose-rich tissue, including breast tissue, and the low quality of slices compared with those obtained from a conventional sectioning method, such as paraffin-embedding sectioning.^19,20 Another, more recent option for rapid histological analysis, microscopy with ultraviolet surface excitation (MUSE), requires only a brief staining time and inexpensive excitation light source and can image specimens without tissue sectioning.^21,22 However, MUSE can provide only superficial images of specimens, less than $10\mu \mathrm{m}$ below the surface,²² and, based on current SMA criteria, such a thin examined layer may be inadequate to secure a low local recurrence rate of cancer.^1,7 Other than histological methods requiring staining, several label-free optical imaging techniques can be performed intraoperatively, including reflectance confocal microscopy,^23–25 third-harmonic generation microscopy,^26,27 multiphoton autofluorescence microscopy,^28,29 and stimulated Raman scattering microscopy.³⁰ The contrast of reflectance confocal microscopy relies on the refractive index, which is not specific to the cell nucleus; thus, a comparable background signal can hamper a reliable SMA. The same nonspecific contrast issue confronts third-harmonic generation microscopy as well. Multiphoton autofluorescence microscopy currently can provide only a negative contrast of DNA/RNA, making it innately challenging to confirm the present or exact shape of a nucleus. Stimulated Raman scattering microscopy can provide contrast-specific imaging of the cell nuclei by spectral decomposition. However, when the typical rough surface of an excised specimen is to be imaged without tissue sectioning, due to its few-micrometers depth of focus, stimulated Raman scattering microscopy may need multiple scanning sessions, shifting the focal plane along the depth direction for each session, to acquire a complete surface image of the specimen. To address all of these challenges, in previous studies, our lab developed ultraviolet photoacoustic microscopy (UV-PAM), which provides label-free and high contrast images of DNA/RNA in cell nuclei with greater imaging depth than a conventional histological image.^31–34 UV light absorption at 266 nm by DNA/RNA is an order of magnitude higher than that by other biomolecules, such as lipids or proteins.^31,32 We demonstrated that UV-PAM was able to provide analysis of the same quality as postoperative histological analysis and that it required less time.³³ This demonstration suggested that UV-PAM could potentially provide accurate intraoperative SMA, reducing the rate of re-excision in cancer surgeries. Further technical advances in UV-PAM can help this technology achieve its full potential. The most sought-after improvement is the imaging speed. The current UV-PAM system acquires photoacoustic (PA) signals on a point-by-point basis with raster scanning, so its imaging speed is limited by the repetition rate of the laser, a critical bottleneck to practical clinical applications such as intraoperative SMA.³³

In this study, we developed a multifocal UV-PAM (MF-UV-PAM) system to mitigate this limitation. By exciting the biological specimens with multiple optical spots, and subsequently reconstructing the induced PA signals at these points, imaging can be accelerated by increasing the number of excitation foci.^35–37 Thus, the imaging speed can be improved drastically without being solely limited by the laser repetition rate, while the lateral resolution remains optically defined.

2.

Methods and Materials

2.1.

Multifocal UV-PAM System

Based on the PA phenomenon, our MF-UV-PAM system [Fig. 1(a)] produced PA images by detecting acoustic waves generated using UV light. A nanosecond pulsed UV laser (266-nm wavelength, 10-kHz laser repetition rate; WEDGE HF 266 nm, Bright Solutions Srl.) was employed for PA excitation. The laser beam diameter was first expanded by a set of lenses (LA4647 and LA4663, Thorlabs Inc.). The laser beam width in one direction was then reduced by a set of cylindrical lenses (LJ4709RM and LJ4281RM, Thorlabs Inc.), which was followed by converging the width further by another cylindrical lens (LJ4147RM, Thorlabs Inc.) to illuminate a one-dimensional (1-D) microlens array [$250\text{-}\mu \mathrm{m}$ lens pitch, $835\text{-}\mu \mathrm{m}$ radius of curvature, 40 micro lenses, 0.075 numerical aperture (NA); SUSS Microtec Se.]. By means of the 1-D microlens array, the laser beam was transformed into multifocal spots focused on the bottom side of an imaging sample, which was placed on the bottom of a water tank filled with water. Part of the generated acoustic waves propagated inside the sample and water and then were detected by a focused 1-D array ultrasonic transducer (50-MHz center frequency, 256 elements; MS 700, FUJIFILM VisualSonics, Inc.). The received acoustic pressure was converted to an electric signal, amplified, and sent to a 256-channel data acquisition system (DAQ), which consisted of two PXIe-1085 chassis managed by PXIe-8880 controllers (National Instruments), each configured with 16 PXIe-5170R data acquisition boards. The DAQ digitalized the 256-channel amplified signals at $250\mathrm{MS}/\mathrm{s}$ and then transferred the data to a computer for recording. In the computer, using MATLAB (Mathworks, Inc.),³⁸ the PA image was reconstructed based on the recorded signals and, finally, was displayed on a screen. Cross-sectional images or volumetric images could be produced, respectively, by linear or raster scanning of the sample using a motorized scanning stage. The stage scanning, data acquisition trigger, and laser emission were all synchronized by a microcontroller (Mega 2560, Arduino), and the control interface was programmed in LabVIEW (National Instruments).

Fig. 1

Schematic of multifocal ultraviolet laser photoacoustic microscopy (MF-UV-PAM) system and a comparison with conventional ultraviolet laser photoacoustic microscopy (UV-PAM). (a) Schematic of MF-UV-PAM. (b) The conventional UV-PAM system with a single optical focal point. (c) MF-UV-PAM system with multiple optical focal points.

3.

Results

3.1.

System Performance

Figure 2(a) shows a representative result of the PA MAP image of a gold nanoparticle, and Fig. 2(b) shows the same PA MAP data with isotropic 2-D Gaussian fitting in a 3-D rendering, with its $z$-axis as the normalized PA amplitude. The calculated mean value of the FWHM of the lateral PSF was $1.6\pm 0.2\mu \mathrm{m}$ ($\text{mean value}\pm \text{standard deviation}$, 10 samples), which was reasonably close to the theoretical value, $1.8\mu \mathrm{m}$ $\lambda /2\mathrm{NA}$, where $\lambda$ is the wavelength of the laser beam. Figure 2(c) shows a reconstructed 1-D image along the depth direction of the gold nanoparticle in Fig. 2(a) and the absolute value of its Hilbert transform (envelope of the curve). The calculated mean value of FWHM of the axial PSF was $40.9\pm 6.9\mu \mathrm{m}$ (10 samples). Figure 3(a) shows the PA MAP image of the crossing hairs in the clear medium, and Fig. 3(b) shows the corresponding 2-D depth information of each hair, which shows that the hairs located at different depths can be clearly distinguished.

Fig. 2

Lateral and axial resolutions of the MF-UV-PAM system. (a) A representative MAP image of the photoacoustic signal of a gold nanoparticle. (b) Isotropic 2-D Gaussian fitting of the MAP image in (a). (c) Reconstructed 1-D image along the depth direction of the gold nanoparticle in (a) and its Hilbert transformation.

Fig. 3

MAP image of the PA signal and 2-D depth information of crossed hairs located at different depths. (a) PA MAP image and (b) 2-D depth information.

3.2.

Mouse Brain Slice Imaging

Figure 4(a) shows a PA MAP image of a coronal section of the fixed mouse’s brain. The total scanning time of this image was around 16 min, which was 40 times faster than possible with the older imaging system using a single-element ultrasonic transducer and a stepping motor for scanning. Figures 4(b)–4(d) show close-up images of the hippocampal region, the olfactory region, and the retrosplenial region of the cerebral neocortex in Fig. 4(a), respectively. Even in the dense region of the nuclei, the individual nuclei are clearly differentiated. Figure 4(e) shows a representative PA MAP image of a single nucleus with diameter of around $5\mu \mathrm{m}$, from Fig. 4(d).

Fig. 4

MAP image of the PA signal of a fixed mouse’s brain slice: (a) PA MAP image of the whole area of the brain slice; (b) close-up image of the hippocampal region; (c) close-up image of the olfactory region; (d) close-up image of the retrosplenial region of the cerebral neocortex; and (e) close-up image of a single nucleus.

4.

Discussion

Compared with the earlier UV-PAM employing a single objective, the lateral resolution of the MF-UV-PAM is generally degraded by a few times because the NA of the commercially sourced microlens array is smaller than that of commercially available single objectives. However, the lateral resolution achieved in this study could sufficiently differentiate individual cell nuclei. To further improve the performance of MF-UV-PAM for SMA, it is more important to improve the axial resolution, in other words, to achieve optical sectioning capability. Several techniques have been developed previously, including a technique utilizing Grüneisen relaxation,⁴⁰ extracting the higher orders of the PA signal,⁴¹ or utilizing multiview excitation.⁴² Providing high lateral and axial resolutions can reduce false-positive margins³³ as well as provide information about intranuclear structures, which may prove useful in achieving a more accurate SMA.

To increase the imaging speed further, we can use a microlens array with an even higher microlens density within the same size. A limiting factor capping this density is the acoustically determined lateral resolution of the 1-D array ultrasonic transducer, $75\mu \mathrm{m}$ in our system. If the microlens density is so high that the distance between the neighboring microlenses is comparable to or smaller than this lateral resolution, the reconstructed images of the neighboring focal spots will start to overlap, and the neighboring signals will affect each other (cross-talk). In our system, to guarantee the image quality by minimizing cross-talk, we used a microlens array with a pitch of $250\mu \mathrm{m}$. However, a microlens array with a smaller pitch could possibly be used if the pitch is above the limiting factor. The maximum number of the available microlenses is also limited. Due to the limitation of the available laser energy, as the number of the microlens increases, the laser energy available for a single microlens decreases; thus, the SNR also decreases accordingly. Assuming that the SNR is proportional to the laser energy per single microlens and a cell nucleus image with an SNR of 5 provides adequate quality for SMA, then, given the SNR of 73.6 for a single nucleus in our system [Fig. 4(e)], the maximum possible number of microlenses would be 588 using a laser with the same total energy. In designing a real system under the two limiting factors above, if we employ a 1-D microlens array with the smallest possible pitch, the maximum number of microlenses would be only 133. However, we can also consider a system using a 2-D microlens array with a 1-D array ultrasonic transducer. A 2-D PA image can be reconstructed if the PA sources produced at the multiple points illuminated by the 2-D microlens array are located within the acoustically sensitive area of the 1-D array ultrasonic transducer. This configuration can be achieved with side detection of the PA signal by the 1-D array ultrasonic transducer. Because the acoustically sensitive area of our 1-D array, ultrasonic transducer is around 7 mm (depth direction) by 10 mm (lateral direction), 588 microlenses are easily accommodated, even with a $250\mu \mathrm{m}$ pitch 2-D microlens array. Accordingly, the imaging speed can be improved in proportion to the number of microlenses. In addition, if a high energy laser and/or a 1-D array ultrasonic transducer with a higher acoustic resolution is available in the future, more microlenses could be employed to improve the imaging speed.

Based on our estimate, instead of using the microlens array, simply replacing the motorized stage with a fast scanning device, such as a galvo-mirror, MEMS mirror, or voice-coil motor, would improve the imaging speed by around two to four times at best. This improved speed, however, is considered not to be significantly better than that of our proposed method, and is not sufficient for intraoperative SMA. However, in future studies that employ a laser operating at a higher repetition rate beyond the maximum response speed of the motorized stage, integrating a fast scanning device into the system would help to maximize the imaging speed. Such an integration could be readily implemented for our MF-UV-PAM.

Last, we should consider the possibility of revising the current transmission-mode setup to enable in vivo imaging, including reflection-mode imaging. The simplest way would be to place the 1-D array ultrasonic transducer at the same side of the 1-D microlens array, but at a tilted angle to make sufficient space to place the 1-D microlens array right above/below the sample. As in our current setup, the tilted 1-D array ultrasonic transducer and the 1-D microlens array must keep the optical focal points and the acoustic focal line confocally aligned. This way, the PA signals excited at the multiple optical spots can be detected by the 1-D focused ultrasonic transducer. In implementing the system in reality, we will have to optimize the space carefully so that the 1-D array ultrasonic transducer does not interfere with the microlens array. Also, the optimal tilt angle of the 1-D array ultrasonic transducer needs to be carefully investigated to achieve sufficient acoustic sensitivity. Instead of tilting only the 1-D array ultrasonic transducer, we could tilt the optical axis of the microlens array, or we could tilt both. Second, we can use a 2-D microlens array and a ring-shaped or 1-D array ultrasonic transducer. The combination of a 2-D microlens array and a ring-shaped array ultrasonic transducer would be similar to the design used in a previous study by our lab.³⁵ This design may not be categorized as reflection-mode; however, an intact live animal can be placed under the 2-D microlens array with no need for slicing, enabling in vivo imaging. A combination of a 2-D microlens array and a 1-D array ultrasonic transducer would basically follow the same configuration as discussed above in estimating the maximum possible number of microlenses. This configuration could be potentially used for in vivo imaging as well.

In summary, we demonstrated that UV-PAM employing a 1-D microlens array and 1-D array ultrasonic transducer can successfully achieve an imaging throughput several tens of times higher than that of the conventional point-by-point scanning method, while keeping satisfactory image quality. This demonstration is an important milestone for the practical implementation of UV-PAM for intraoperative SMA.