A high-speed and open-top ultraviolet photoacoustic microscope that can provide histological images intraoperatively for surgical margin analysis is demonstrated, including the system configuration, optical alignment, sample preparation, and experimental procedures.
Surgical margin analysis (SMA), an essential procedure to confirm the complete excision of cancerous tissue in tumor resection surgery, requires intraoperative diagnostic tools to avoid repeated surgeries due to a positive surgical margin. Recently, by taking the advantage of the high intrinsic optical absorption of DNA/RNA at 266 nm wavelength, ultraviolet photoacoustic microscopy (UV-PAM) has been developed to provide high-resolution histological images without labeling, showing great promise as an intraoperative tool for SMA. To enable the development of UV-PAM for SMA, here, a high-speed and open-top UV-PAM system is presented, which can be operated similarly to conventional optical microscopies. The UV-PAM system provides a high lateral resolution of 1.2 µm, and a high imaging speed of 55 kHz A-line rate with one-axis galvanometer mirror scanning. Moreover, to ensure UV-PAM images can be easily interpreted by pathologists without additional training, the original grayscale UV-PAM images are virtually stained by a deep-learning algorithm to mimic the standard hematoxylin- and eosin-stained images, enabling training-free histological analysis. Mouse brain slice imaging is performed to demonstrate the high performance of the open-top UV-PAM system, illustrating its great potential for SMA applications.
Surgical margin analysis (SMA), which requires an examination of tissue specimens under a microscope, is an essential procedure to determine whether all cancer cells are removed from a patient’s body in a resection surgery1. Therefore, a microscope that can rapidly provide histological images is vitally important for SMA to avoid repeated surgeries caused by incomplete removal of cancer cells. However, according to the current gold-standard method based on bright-field optical microscopy, the excised tissue is required to be fixed in formalin, embedded in paraffin, sectioned into thin slices (4-7 µm), and then stained by hematoxylin and eosin (H&E) before imaging, which is time-consuming (3-7 days) and laborious2,3. A frozen section is a rapid alternative for SMA by quickly freezing, slicing, and staining the tissue, which can provide histological images in 20-30 min4. However, the histological features are often distorted and required skillful training, which hinders the applicability of the technique to multiple types of organs5.
Optical microscopy techniques that can provide cellular images without or with a few steps of tissue processing have been developed for SMA. However, each of them suffers from different issues. For example, optical coherence tomography6 and confocal reflectance microscopy7 suffer from low specificity because of their low intrinsic scattering contrast. Although microscopy with ultraviolet surface excitation8 and light-sheet microscopy9 can provide high-resolution and high-contrast images for SMA, the toxic and volatile staining procedure usually cannot be performed in an operating room, which prolongs the turnaround time. Multi-photon microscopy10 and stimulated Raman microscopy11 can provide rich information for SMA. Yet, the high cost of the required ultrafast lasers that are used to generate nonlinear effects prevents their wide applicability.
Recently, by taking advantage of intrinsic optical absorption, label-free ultraviolet photoacoustic microscopy (UV-PAM) has been developed to provide high-resolution histological images12. In UV-PAM, the photon energy of the excitation UV light (e.g., 266 nm) is first absorbed by the DNA/RNA in cell nuclei13 and then converted into heat, inducing acoustic wave emission through thermal-elastic expansion14. By detecting the generated acoustic waves, two-dimensional (2D) UV-PAM images of cell nuclei can be obtained via maximum amplitude projection of the acoustic signals, providing histological information for SMA. To enable the clinical applications of UV-PAM, high-speed UV-PAM based on galvanometer mirror scanning has been developed to provide histological images for a brain biopsy sample (5 mm x 5 mm) within 18 min, showing great potential in time-sensitive applications15. To further validate the possibility of UV-PAM for thick tissue imaging, a reflection-mode UV-PAM system with a waterproof one-axis microelectromechanical systems scanner was proposed, successfully demonstrating intraoperative histopathological examination of human colon and liver tissues16. Since the original UV-PAM image is in grayscale while the gold standard H&E-stained image is in pink and purple colors, it is difficult for pathologists to interpret UV-PAM images directly. To address this issue, a deep-learning algorithm was proposed to transfer grayscale UV-PAM images into virtual H&E-stained images in near real-time so that pathologists can understand the images without any additional training17.
This work reports a high-speed and open-top UV-PAM system that can be operated similar to conventional optical microscopies, providing both original grayscale histological images and virtually stained images assisted by a deep-learning algorithm. A formalin-fixed and paraffin-embedded (FFPE) mouse brain slice is imaged by the UV-PAM system to demonstrate the similarity between our virtually stained UV-PAM and standard H&E-stained images, showing its potential for SMA applications.
All animal experiments performed in this work are approved by the Animal Ethics Committee at The Hong Kong University of Science and Technology.
1. Open-top UV-PAM system (Figure 1)
2. Sample preparation
3. Experimental procedures
Figure 1 shows the schematic of the high-speed UV-PAM system. In this setup, the optical excitation and ultrasonic detection paths are on the same side and below the sample, forming a reflective mode and open-top system. Thus, it is user-friendly and suitable for imaging thick samples.
Figure 2 shows the scanning trajectory of the UV-PAM system during imaging. Sub-image of each section (e.g., area of 5 mm x 30 µm) is first generated using a scattered interpolation algorithm, and then, a whole image (e.g., area of 5 mm x 5 mm) is obtained by stitching all the sub-images using custom image processing algorithm (see Table of Materials).
Figure 3A and Figure 3B show the UV-PAM and H&E images of an FFPE mouse brain slice, respectively, both of which have a field-of-view of 5 x 5 mm2. The image processing algorithm can be accessed from Github via the link provided in the Table of Materials. The image acquisition time was less than 18 min. Figure 3C–F show the zoomed-in images of the marked regions in Figure 3A, where the individual cell nuclei can be clearly resolved. More importantly, the corresponding cell nuclei can be found in the standard H&E-stained images (Figure 3G–J), showing the high accuracy of the current system for cellular imaging. To transfer the grayscale UV-PAM image to a virtual H&E-stained image, a deep-learning algorithm17 (see Table of Materials) was applied, which would take less than 30 s for an image with 8000 x 8000 pixels. The corresponding zoomed-in images are shown in Figure 3K–N. The virtually stained UV-PAM images provide almost the same structural information as the H&E-stained images, showing promise for clinical translation of the current UV-PAM system.
Figure 1: The schematic of the high-speed and open-top UV-PAM system. (A) The system setup. (B) Photograph of the sample holder and a water tank that is attached to a two-axis manual stage. (C) Photograph of the ring-shaped ultrasonic transducer fixed in the water tank and the sample tank that covers the hole of the sample holder. (D) Photograph of the sample tank with the membrane and sample that would be placed on the sample holder. UV: Ultraviolet; DAQ: Data acquisition card; GM: Galvanometer mirror. Please click here to view a larger version of this figure.
Figure 2: The scanning trajectory of the UV-PAM system during imaging. Please click here to view a larger version of this figure.
Figure 3: Experimental results of UV-PAM imaging with deep learning-based virtual staining. (A) UV-PAM image of an FFPE mouse brain slice. (B) Standard H&E-stained image of the same slice. Scale bars: 1 mm. (C–F) Zoomed-in images of the marked regions in A. (G–J) Corresponding H&E-stained images of the marked regions in B. (K–N) Corresponding virtually stained images using a deep-learning algorithm. Scale bars: 50 µm. Please click here to view a larger version of this figure.
In summary, a high-speed and open-top UV-PAM system has been demonstrated for histological imaging. The detailed instructions about the system configuration, optical alignment, sample preparation, and experimental procedures are presented. The image acquisition program can be accessed from Github via the link provided in the Table of Materials. The lateral resolution of the present system is ~1.2 µm which has been experimentally measured in a recent publication21. A mouse brain slice was imaged to demonstrate that the present system can obtain a histological image within 18 min for an area of 5 x 5 mm2, which is a typical size of brain biopsy22. Although the original image is in grayscale, with the assistance of a digital virtual staining tool enabled by a deep-learning algorithm, the present system can further provide virtually stained images in near real-time, ensuring easy adaptation for pathologists to interpret the images. As a label-free image technique, the present UV-PAM system can also provide histological images for unprocessed fresh tissue samples. More examples (including frozen-sectioned and fresh tissue samples) have been demonstrated in a previous publication17. The experimental results show the high potential of the present deep-learning-assisted UV-PAM system in SMA applications.
One of the advantages of the UV-PAM system is that the system is implemented in reflection mode, enabling imaging of thick tissues. Besides, the open-top UV-PAM system allows the sample to be placed on the scanning window (the membrane of the sample tank), which has a similar operation as traditional optical microscopies. Therefore, this system is more user-friendly than other systems that require samples to be sandwiched by two membranes11,14. Moreover, by using a 1D GM with a high-repetition-rate UV laser, the present UV-PAM system can achieve high imaging speed with higher cost-effectiveness when compared with the system using multifocal excitation23.
Currently, the imaging speed is mainly limited by the repetition rate of the laser and photon budget. With a laser that has a high repetition rate and high pulse energy, the imaging time can be further shortened. Another limitation of the system is that the sample can only be roughly adjusted to the focal plane of the objective lens by finding the maximum PA signals, instead of displaying a real-time image for users to visualize whether the sample is in focus. To display a near real-time image, 2D GM can be applied.
There are two critical steps in the protocol: (a) the confocal requirement of the optical and acoustic foci should be optimized to achieve a high detection sensitivity; (b) the scanning range of the GM on the x-axis should be smaller than the acoustic focal spot of the ring-shaped UT to maintain similar high detection sensitivity (in the current setup, the scanning range is ~30 µm ±15 µm). Otherwise, obvious vignetting effects around the edges would occur when multiple sub-images are stitched together to obtain a whole image.
The authors have nothing to disclose.
The authors would like to acknowledge the financial support from the Hong Kong Innovation and Technology Commission (ITS/036/19).
Alcohol | Sigma Aldrich | PHR1373 | Sample dehydration |
Amplifier | Mini Circuit | ZFL-500LN-BNC+ | Ultrasonic signal amplification |
Controller | National Instruments | NI myRIO | System controller |
Data acquisition card | Alazar Technologies | ATS9350 | Ultrasonic signal collection |
Deep-learning algorithm | For transfering the grayscale UV-PAM image to a virtual H&E-stained image; https://github.com/TABLAB-HKUST/Deep-PAM | ||
Formalin | Sigma Aldrich | R04586 | Sample fixation |
H&E staining kit | Abcam | ab245880 | Sample staining |
Histo-Clear II | National Diagnostics | HS-202 | Sample deparaffinization |
Image acquisition program | National Instruments | LabVIEW | Lab-built program using LabVIEW; https://github.com/TABLAB-HKUST/LabVIEW-program-for-UV-PAM |
Image processing algorithm | Mathworks | MATLAB | Lab-built algorithm using MATLAB; https://github.com/TABLAB-HKUST/ImageRec_GM-UVPAM |
Kinematic platform mounts | Thorlabs | KM200B | Adjust the sample to be flat |
Membrane | Glad | Cling wrap | Sandwiched in sample tank |
Microscope objective lens | Thorlabs | LMU- 5X-NUV | Objective lens |
Motorized stages | Physik Instrumente | L-509.10SD00 | Scanning stages |
One-dimensional galvanometer mirror | Thorlabs | GVS411 | Fast scanning mirror |
Oscilloscope | RIGOL Technologies | DS1102E | Ultrasonic signal readout |
Phosphate-buffered saline | Sigma Aldrich | P3813 | Sample washing |
Pinhole | Edmund Optics | #59–257 | Spatial filtering |
Plano convex lens | Thorlabs | LA-4600-UV | Focusing lens |
Plano convex lens | Thorlabs | LA-4663-UV | Collimating lens |
Pulser/receiver | Imaginant | DPR300 | Pulse echo amplifier |
Q-switch diode-pumped solid-state laser | Bright Solutions | WEDGE HF 266 nm | 266-nm laser |
Ring-shaped ultrasonic transducer | University of Southern California | Ultrasonic signal detection | |
Sample holder | Lab-made | Hold the sample tank | |
Sample tank | Lab-made | Hold biological samples | |
Single-axis Z-translational stage | Thorlabs | PT1 | Manual stage |
Two-axis manual stage | Thorlabs | LX20 | Manual stage |
Water tank | Lab-made | Ultrasonic signal transmission | |
Xylene | Sigma Aldrich | XX0060 | Sample clearing |