Open Access
Add to BibSonomy
Abstract
There is an unmet need for fast virtual histology technologies that exhibit histological realism and can scan large sections of fresh tissue within intraoperative time-frames. Ultraviolet photoacoustic remote sensing microscopy (UV-PARS) is an emerging imaging modality capable of producing virtual histology images that show good concordance to conventional histology stains. However, a UV-PARS scanning system that can perform rapid intraoperative imaging over mm-scale fields-of-view at fine resolution (<500 nm) has yet to be demonstrated. In this work, we present a UV-PARS system which utilizes voice-coil stage scanning to demonstrate finely resolved images for 2×2 mm2 areas at 500 nm sampling resolution in 1.33 minutes and coarsely resolved images for 4×4 mm2 areas at 900 nm sampling resolution in 2.5 minutes. The results of this work demonstrate the speed and resolution capabilities of the UV-PARS voice-coil system and further develop the potential for UV-PARS microscopy to be employed in a clinical setting.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
The primary treatment option for solid-tumor cancers is surgical resection, in which surgeons attempt to remove all malignant tissue while retaining as much healthy tissue as possible. Upon removal of the tissue volume, the sample will be inspected for positive surgical margins (PSMs), which is indicative of incomplete resection. The gold-standard histological method for margin analysis is via bright-field imaging of hematoxylin and eosin (H&E) stained formalin-fixed paraffin-embedded (FFPE) tissue sections. However, this process is both laborious and time-consuming, hence necessitating post-operative margin analysis. An estimated 20-40 % of patients who undergo breast-conserving surgeries will require re-excision due to the post-operative identification of PSMs [1,2], placing additional burden on both the patient and the healthcare system. Frozen section (FS) procedures have also been employed for histopathological margin analysis. FS techniques can report results within 20 minutes [3] and is thus a viable option for margin analysis within the surgical suite. However, FS techniques suffer from freezing artifacts during sample preparation, yielding an accuracy of only 84 % for margin identification in lumpectomies [4]. Moreover, prolonged operation is associated with increased risk of surgical-site infections, with the likelihood of infection increasing to 13 %, 17 %, and 37 % for every additional 15 minutes, 30 minutes, and hour of operation, respectively [5]. As such, there is an unmet desire for highly-accurate diagnostic techniques that can produce H&E-like staining contrast within expedited intraoperative time-frames.
Over recent decades, there has been a drive to realize virtual histology images using optical imaging modalities, including imaging techniques such as stimulated Raman scattering [6–8], multi-photon fluorescence [9–11], confocal fluorescence microscopy [12], microscopy with ultraviolet surface excitation [13], and light sheet microscopy [14]. Recently, a novel label-free imaging technique known as photoacoustic remote sensing (PARS) microscopy has been developed [15–18]. In PARS microscopy, a pulsed excitation beam is co-focused with a continuous wave detection beam to scan over a tissue region. By using the photoacoustic-induced modulations of the back-scattered detection beam, an absorption-contrast image can be formed. PARS microscopy has demonstrated the ability to image a variety of endogenous biological structures, including vasculature [15,17] with 532 nm excitation, lipids with 1225 nm excitation [19], and cell nuclei with ultraviolet 266 nm excitation (UV-PARS) [20–23]. Since UV-PARS microscopy can image distributions of nuclei with fine resolution, it is a promising modality for virtual histopathological applications. Furthermore, depth-resolved nucleic imaging has been demonstrated [23] with an axial resolution of 1.6 $\mu$m, allowing for UV-PARS depth sectioning of cellular layers up to the penetration depth of the 266 nm beam in biological tissue - tens of microns from the surface of the sample. Virtual histology images using UV-PARS were first demonstrated in Restall et al. [24], in which UV-PARS images were combined with simultaneously acquired 1310 nm scattering images, where the UV-PARS scan corresponds to a nucleic contrast (virtual hematoxylin stain), and the scattering scan corresponds to cytoplasmic contrast (virtual eosin stain). Enhanced scattering resolution was later achieved by measuring the average intensity of the back-scattered 266 nm excitation pulse, thus demonstrating both virtual hematoxylin and eosin stains with matched resolution shown in Haven et al. [25]. Further work has demonstrated both radiative and non-radiative contrast after PARS excitation [23,26,27]. Moreover, UV-PARS histology can be combined with imaging modalities such as optical coherence tomography (OCT) [28,29].
Early iterations of UV-PARS microscopy employed two galvanometer mirrors for scanning the co-focused beams through the objective lens, yielding 250×250 $\mu$m$^2$ UV-PARS images [21]. However, this technique requires extremely long scanning times to enable imaging over large areas due to the requirement for inter-tile translation. Large-area virtual histology images of whole-slide FFPE sections with UV-PARS microscopy have been demonstrated [30] but were generated at slow rates of 1.5 minutes per mm$^2$ with 500 nm sampling resolution, scaling to a 1 cm$^2$ area in 2.5 hours. Increases in imaging speed were demonstrated in Restall et al. [31], in which mosaic strip scanning was used to generate 5×10 mm$^2$ images in 3.5 minutes. While this system is capable of rapid imaging speeds over large areas, it requires the use of a smaller numerical aperture (NA) objective lens, and so the degraded optically-focused lateral resolution of 0.86 $\mu$m limits the diagnostic utility of the generated nuclei images. Accurate histopathological diagnosis requires an equivalence of 200X magnification [32] such that both internuclear and intranuclear details can be delineated. As such, the hybrid-optomechanical UV-PARS scanning technique demonstrated in [31] is suitable for gross scans only.
Voice-coil stage scanning has been employed previously in photoacoustic microscopy (PAM) [33–35]. In these works, the photoacoustic imaging probe was mounted to the voice-coil stage, with a B-mode PAM image being generated for each sweep of the voice-coil stage. The dual ultrasound-photoacoustic voice-coil system demonstrated by Harrison et al. [33] can perform 10 Hz bi-directional scans over a 9 mm travel range, yielding B-mode images over a 9 mm travel range at 20 frames per second (fps), albeit at a course lateral resolution of 180 $\mu$m. Advancements were shown in Wang et al. [34], in which 40 fps and 20 fps PAM imaging rates were demonstrated over 1 mm and 9 mm travel ranges, respectively, with an optically focused lateral resolution of 3.4 $\mu$m. Enhancements in voice-coil PAM imaging speed were demonstrated at 40 fps over 9 mm ranges by using bright-field illumination, albeit with a degraded acoustically focused lateral resolution of 44 $\mu$m [35].
Previous iterations of UV-PARS microscopy have demonstrated either fine resolution (<500 nm sampling) images or rapid imaging speeds, but no UV-PARS system to date has achieved the combined feat of performing fine resolution scans within intraoperative time-frames. Figure 1 compares the maximum speed and resolution capabilities of recent UV-PARS iterations with the scanning technique presented in this work. We demonstrate a UV-PARS microscopy system based on voice-coil stage scanning, which can achieve sub-micron resolution and rapid scanning rates over mm-scale regions, thus demonstrating the potential of UV-PARS microscopy for translation to a clinical setting. In this work, we show voice-coil UV-PARS imaging on both thin slide FFPE sections of human prostate as well as fresh murine liver.
Fig. 1. Comparison diagram between UV-PARS scanning techniques. Scale bars = 50 $\mu$m. a) Hybrid opto-mechanical scanning described in Restall et al. [31]. The galvanometer mirror sweeps the beam in the y-direction as the stage traverse the x-direction. Mosaic strips are concatenated to create a large field-of-view. b) Line scanning method utilized in Haven et al. [25]. The stage scans horizontal lines, stepping vertically between line scans. c) Voice-coil based UV-PARS scanning. The voice-coil stage oscillates in the horizontal direction while the stepper stage traverses the vertical direction to create a sinusoidal scanning trajectory. OL - Objective lens; GM - galvanometer mirror.
2. Methods
2.1 System design
The system diagram is displayed in Fig. 2, similar to that shown in previous UV-PARS systems [24,25], albeit with a novel scanning approach. In the excitation path, the beam from a 532 nm 10 kHz - 2 MHz pulsed laser (SPFL-532-40, MKS) is beam reduced prior to being frequency doubled to generate 266 nm radiation via a CLBO crystal (CLBO-501S, EKSMA) and then spectrally separated through a prism (PS863, Thorlabs). The UV beam is then expanded and directed through a half-waveplate (WPH05M-266, Thorlabs) and polarizing beam-splitter (PBS) (10SC16PC.22, Newport). The reflected portion of the beam is used for optical event triggering via a silicon photodiode (PDA10A, Thorlabs), in which the generated electrical pulse is then passed through a digital delay generator (DG645, SRS) for digital event recognition. The transmitted portion of the 266 nm beam is then directed towards a harmonic beam-splitter (HBSY134, Thorlabs) for beam combination. In the detection path, a 1310 nm super luminescent diode (SLD) (SLD1018S, Thorlabs) is fiber-coupled to a zoom-collimator (ZC618APC-C, Thorlabs) prior to beam expansion. The beam is then directed through a half-waveplate (WPH05-1310, Thorlabs) and polarizing beam-splitter (CCM1-PBS254, Thorlabs). The transmitted portion of the beam passes through a quarter-waveplate (WPQ10M-1310, Thorlabs) which converts the 1310 nm beam to circular polarization. Upon back-scattering, the 1310 nm detection beam will have the opposite polarization and thus be reflected by the polarizing beam-splitter, which directs the beam towards a 75 MHz balanced photodetector (PDB420C-AC, Thorlabs) for extraction of the PARS and scattering signals. After photodetection, the PARS signal is passively band-pass filtered through an 11 MHz inline low-pass filter (BLP-10.7+, Mini-Circuits) for signal elongation and a 1.8 MHz high-pass filter (EF509, Thorlabs) for removal of scanning artifacts. At the harmonic beam-splitter, the 266 nm and 1310 nm beams are combined and co-focused onto the sample through a 0.5 NA reflective objective lens (LMM40X-UVV, Thorlabs) for sample scanning. A vertical translation stage (X-VSR20A-E01, Zaber) is utilized to adjust the sample height relative to the beam foci. The lateral and axial optical resolution for UV-PARS was previously characterized as 0.39 $\mu$m and 1.2 $\mu$m, respectively [24].
Fig. 2. System diagram for the voice-coil based UV-PARS system. SLD - superluminescent diode; ZC - zoom collimator; L - lens; HWP - half-wave plate; M - mirror; PBS - polarizing beam splitter; PD - photodiode; P - prism; BD - beam dump; QWP - quarter-wave plate; HBS - harmonic beam splitter; RO - reflective objective; DG - delay generator; DAQ - data acquisition system; CLBO - Caesium lithium borate crystal.
2.2 Scanning methodology
The scanning methodology is displayed in Fig. 3. A voice-coil stage with 100 nm encoder resolution (X-DMQ12L-AE55D12, Zaber) oscillates in the x-direction and is mounted atop a stepper stage (PLS-85, PI), which traverses the y-direction at a constant velocity. To track absolute position in the x-direction, two analog quadrature encoder channels from the voice-coil stage are captured and digitized, with the successive states of each signal (’0’ or ’1’) being recorded at each rising or falling edge. By analyzing the successive encoder state changes, an x-position trajectory is incrementally constructed in MATLAB. The y-position trajectory can be constructed by simply recording the times at which each encoder state change occurs and multiplying by the known constant velocity of the stepper stage. The combination of the x- and y-trajectories yield a sinusoidal scanning pattern in the focal plane. The digital trigger rising edge event is linearly interpolated between the nearest two encoder edges in time for mapping the PARS event to an absolute position, as shown in Fig. 3.
Fig. 3. Diagram of the voice-coil scanning methodology. Left: The combined slow-axis and voice-coil stage motion yield a sinusoidal scanning trajectory across the tissue sample. Middle: Sampling resolution characterization of the sinusoidal scanning trajectory. Right: Time-resolved depiction of the acquired signals in a particular segment of the scan trajectory. Laser excitation events are linearly interpolated between the two nearest encoder edges. PARS and scattering signals are processed over a 500 ns window.
2.3 Data acquisition
For signal acquisition, a 100 MHz 14-bit digitizer (CSE8389, GaGe) streams five data channels continuously at 25 MS/s, which are the PARS and scattering signals, two quadrature encoder channels for position tracking, and a trigger signal from the digital delay generator. In previous works [25], PARS signals were acquired at a 125 MS/s (8 ns/S) to sufficiently sample the rapidly developing PARS waveform over the $\sim$125 ns modulation window [15], with data segments being acquired only at trigger events. However, for absolute position tracking with voice-coil scanning, all data channels must be streamed continuously for acquisition systems that use a single digitizer card. Performing a continuous data stream of data at 125 MS/s for minute-duration scans is extremely demanding on system resources, and so the reduced sample rate of 25 MS/s (40 ns/S) was chosen as to be able to sufficiently sample the encoder waveforms along with both virtual histology signals. However, sampling the unfiltered PARS signal at 40 ns intervals is not adequate to capture the temporal modulations. By applying a passive inline 11 MHz low-pass filter, the PARS signal becomes broadened enough temporally to be sampled sufficiently over a 500 ns time window at 25 MS/s. It should be noted that filtering the PARS signal will remove spectral energy, and thus ultimately degrade the signal-to-noise ratio (SNR). However, in the presence of the above filters, there was no difference in signal amplitude between fast and slow scanning modes with the voice-coil stage. For each laser excitation event, a single data point for each of the UV-PARS and scattering data channels are extracted, obtained by integrating the absolute value of the PARS signal over a 500 ns window and averaging the scattering channel over the same temporal window. For scanning a $\sim$1 cm$^2$ area in 3.5 minutes at 1 $\mu$m mean sampling resolution or 8.8 minutes at 400 nm mean sampling resolution, this would require a total of 51 GB and 127.5 GB of data to be acquired for processing, respectively.
2.4 Image reconstruction
After extracting the position-based PARS and scattering data, the data sets are then interpolated on a 2D grid via Delauney triangulation, with the point-spacing of the grid set to be the mean lateral sampling resolution for the scan in question. After interpolating both data sets to obtain independent PARS and scattering images, the images are then histogram equalized in MATLAB to maximize contrast.
2.5 Scanning parameters
While the unloaded voice-coil stage is capable of oscillating the 1.2 cm full-travel range (FTR) at 25 Hz, the stage load will place an upper-limitation on scanning frequency. For a sinusoidal acceleration curve, the maximum acceleration can be expressed by
where $W$ is the image width and $f$ is the bi-directional voice-coil stage frequency. Given that the voice-coil stage can exert a maximum constant force-over-travel of $F=$15 N, this allows for the stage to drive a mass of $m = m_s + m_l = \frac {F}{a_{max}}$, where $m_s$ = 95 g is the mass of the moving stage-top and $m_l$ is the secured load mass. We can express the maximum scanning frequency of the voice-coil stage at a given image width and load mass asThe chosen speed of the slow-axis (stepper) stage is determined by the bi-directional scanning frequency and the desired sampling resolution in the y-direction. This can be expressed as
where $v_s$ is the constant velocity of the slow-axis stage and $D_y$ is the desired mean sampling resolution in the y-direction. The laser pulse repetition rate (PRR) is chosen to match the x-resolution to the y-resolution, such that the time between sequential laser pulses will yield the desired sample resolution when the voice-coil stage is at peak velocity. With the peak velocity of the voice-coil stage written as $v_p = \pi f W$, the minimum required laser repetition rate will be where $D_x$ is the desired sampling resolution in the x-direction. Lastly, assuming matched x and y sampling resolutions ($D = D_x = D_y$), the total scanning time can be calculated using the slow-axis stage speed and the desired image height $H$ asAs an example, for a 10 g sample load, the stage could oscillate the full 1.2 cm range at 25 Hz and thus create a 1.2 × 0.83 cm$^2$ ($\sim$1 cm$^2$) image at fine 400 nm sampling resolution in 7 minutes. Additionally, gross scans could be performed at 1 $\mu$m sampling resolutions in 2.8 minutes. Performing voice-coil scans with these parameters could address the unmet needs of speed, resolution, and field-of-view for intraoperative virtual histology.
2.6 Tissue preparation and sample mounting
To demonstrate the virtual histology capabilities of the voice-coil UV-PARS system, imaging was performed on both thinly sectioned FFPE slides as well as fresh thick murine liver. Quartered FFPE thin sections of human prostate tissue were obtained from radical prostatectomy patients as per approved ethics (HREBA (Cancer) / HREBA.CC- 20-0145). Thin FFPE sections were obtained only after the relevant pathology cases were closed and the tissue flagged for disposal, with all identifying patient information being redacted. For fresh thick tissue imaging, murine liver was extracted from a Swiss Webster mouse (CFW, Charles River) which were procured in accordance with the University of Alberta’s Animal Care and Use Committee ethics guidelines and regulations. Methods for securing both FFPE thin tissue and fresh thick tissue samples to the imaging platform are displayed in Fig. 4. After sample mounting, the stage system was adjusted so as to align the surface of the sample to the focal plane of the 266 nm beam for imaging. To achieve this, the vertical lift stage was used to first lower the sample below the imaging plane, and was then raised until sufficient photoacoustic signal was generated. The tip-tilt of the stage system was then adjusted at this stage height for imaging. When imaging thin tissue, this implied that the depth-of-focus (DOF) of the 266 nm beam was contained to be within the $\sim$5 $\mu$m thin section over the entire field-of-view (FOV). For thick tissue imaging, the DOF of the 266 nm beam was contained to be within tens of microns from the sample surface due to the penetration depth of 266 nm radiation in biological tissue.
Fig. 4. a) Photo of the voice-coil stage with a 3D printed custom slide container and a thin-slide FFPE section. b) Left: Photo of the materials used to construct a thick tissue container. Right: Side-view of the fresh tissue container construction. The custom 3D printed container is super-glued to the glass slide. The thick tissue is then loaded into the container, where it is secured by super-gluing the UV coverslip to the top of the container. The thickness of the 3D printed tissue containers ranged from 3-6 mm, and was chosen to be small enough to ensure the thick tissue sample is pressed to the surface of the coverslip after assembly.
3. Results
3.1 Thin slide H&E validation
To validate the voice-coil UV-PARS system scanning capabilities, imaging was performed on a thin-slide FFPE section obtained from radical prostatectomy patients. Demonstrated in Fig. 5 is a $\sim$2×2 mm$^2$ UV-PARS image of human prostate tissue (left) alongside the true bright-field H&E image of an adjacent tissue section (right). The UV-PARS scan was completed at a 25 Hz bi-directional voice-coil scanning frequency and a 313 kHz laser PRR to obtain a 2×2 mm$^2$ scan at 500 nm mean sample resolution in 1.33 minutes. Images were taken using pulse energies of $\sim$5 nJ, with the UV-PARS image demonstrating an SNR of 30.1 dB, where SNR was calculated by taking the ratio of the mean value over a representative nuclei to the standard deviation of the background noise. It should be noted that since the true H&E image is an adjacent section to the UV-PARS image, there will be close similarities in nucleic structure, but the images will not have one-to-one concordance. As shown in the red inset of Fig. 5, the UV-PARS scan can distinguish both intranuclear and internuclear structure. The green and red arrows of the adjacent H&E image display two variations of cell nuclei. It can be seen that the solid dark-purple nuclei of the adjacent section corresponds to the brightest nuclei in the UV-PARS image, while the translucent light-purple nuclei corresponds to the lighter-gray nuclei in the UV-PARS image. Moreover, the fine resolution UV-PARS scan is not only capable of distinguishing between the two variations of nuclei, but can also resolve intranucleic structure seen in the nuclei highlighted by the red arrows.
Fig. 5. Comparison between the voice-coil UV-PARS image taken at 500 nm mean sample resolution (left) and the adjacent true H&E thin section (right). Top: $\sim$2×2 mm$^2$ UV-PARS image of an FFPE thin section of human prostate tissue, and the adjacent true H&E thin section. Scale bar = 500 $\mu$m. Middle: UV-PARS image of the blue-blowout alongside the true H&E region. Scale bar = 250 $\mu$m. Bottom: UV-PARS image of the red-blowout region alongside the adjacent H&E section. The UV-PARS image demonstrates the variations in nucleic structure, shown by the red and green arrows. Scale bar = 100 $\mu$m.
3.2 Fresh murine tissue imaging
Demonstrated in Fig. 6 and Fig. 7 are 4×4 mm$^2$ UV-PARS images of fresh murine liver at 900 nm mean sample resolution in approximately 2.5 minutes. Figure 6 was imaged with pulse energies of $\sim$2 nJ, a laser PRR of 209 kHz, and displays an SNR of 37.9 dB, while Fig. 7 was imaged with pulse energies of $\sim$1 nJ, a laser PRR of 209 kHz, and displays an SNR of 34.2 dB. Reduced pulse energies were employed for fresh tissue scanning since it has been found experimentally that UV-PARS modulations from fresh tissue are much larger than for thin FFPE sections, and so less energy deposition is required to obtain sufficient signal. As can be seen, the 4×4 mm$^2$ images of Fig. 6 and Fig. 7 display variation in nuclei distributions over the entire FOV. Moreover, the insets of each figure demonstrate how individual nuclei can be distinguished with 900 nm mean sampling resolution. While the ability to resolve intranuclear structure is degraded relative to the finely resolved scan of Fig. 5, it should be noted that the use of a coarser sampling resolution allows for faster imaging rates, a trade-off which will be subsequently discussed.
Fig. 6. a) 4×4 mm$^2$ UV-PARS image of a region of fresh murine liver. Scale bar = 500 $\mu$m. b) Red inset region from a). Scale bar = 250 $\mu$m. c) Green inset region of b). Scale bar = 100 $\mu$m. d) Blue inset region of b). Scale bar = 100 $\mu$m.
Fig. 7. a) 4×4 mm$^2$ UV-PARS image of a region of fresh murine liver. Scale bar = 500 $\mu$m. b) FOV of the red-inset shown in a). Scale bar = 250 $\mu$m. c) Magnified view of the green rectangular inset shown in b). Scale bar = 100 $\mu$m.
4. Discussion
4.1 Present limitations
The scanning time, frequency, and amplitude characteristics of the voice-coil system can be understood by inspecting Fig. 8. Shown in Fig. 8 is a heat map of the voice-coil operation where the cm$^2$ imaging time is color-encoded as a function of scanning width and oscillation frequency, assuming a moving stage mass of 162 g. Displayed by the pink and blue cross-hairs are the voice-coil operating points at which the 2×2 mm$^2$ image (Fig. 5) and 4×4 mm$^2$ images (Fig. 6, 7) were taken, respectively. As shown in the heat map, maximum imaging rates for the voice-coil UV-PARS system can only be attained at the largest scanning amplitudes. The 2×2 mm$^2$ UV-PARS image at 500 nm mean point spacing of Fig. 5 was demonstrated at a scanning frequency of 25 Hz, approximately half of the maximum oscillation frequency. While the voice-coil stage is capable of generating the driving force to produce the 50 Hz scanning frequency, it was found that at frequencies greater than 25 Hz, the rapid motion of the load mass resulted in mechanical instability at the mounting connection between the voice-coil stage and the vertical lift stage. For the 4×4 mm$^2$ scan at 900 nm mean point spacing, the operating frequency was 15 Hz, less than two times the maximum oscillation frequency. In this case, the reduced frequency of 15 Hz was chosen to ensure that the fresh tissue sample did not move over the duration of the scan.
Fig. 8. A heat-map of the cm$^2$ imaging time as a function of scanning width and frequency, calculated using the moving stage mass of 162 g. Imaging times for 400 nm and 1 $\mu$m sampling resolutions are shown on the right-hand side. The pink and blue cross-hairs indicate the operating points for the widths and scanning frequencies demonstrated in Fig. 5 and Fig. 6 and 7 , respectively.
Another limitation of the present voice-coil system is the difficulty in achieving UV-PARS scans over the entire 12 mm travel range. To achieve this, the tip-tilt of both the voice-coil and stepper stage axes must be carefully adjusted to ensure that the sample surface remains within the $\sim$1 $\mu$m DOF of the 266 nm excitation beam. It was found that even with methodical alignment of stage axes’ tip-tilt, the scanning surface could not be contained to the DOF of the 266 nm beam. This was made apparent by performing tip-tilt adjustment using a reflective sample, and then observing the relative change in spot-size of the back-reflected beam when the sample surface is scanned laterally near the beam focus. The inability to perform absolute alignment of the stage systems tip-tilt to the focal plane of the 266 nm beam is primarily due to the flatness specification of both the stepper and voice-coil stages, specified as $\pm$2 $\mu$m and $\pm$6 $\mu$m, respectively. The flatness specification is the difference between the maximum and minimum heights of the stage top over the full travel range, and thus will result in the scanning surface moving in and out of the 266 nm DOF over the duration of a scan. This problem is likely exacerbated by sample mounting considerations. For example, if the microscope coverslip is not parallel to the stage-top, careful alignment of the stage-systems tip-tilt would be required to image a single cellular layer. Moreover, when mounting thick tissue to the imaging platform, the fresh tissue will be compressed between the glass slide and coverslip as shown in Fig. 4(b), which could buckle the surface of the coverslip beyond the penetration depth of the 266 nm beam and thus result in a loss of signal during thick tissue imaging. Alongside the present inability to achieve UV-PARS scans over the FTR, the process of tip-tilt alignment is both time-consuming and requires trained personnel for system operation, limitations that must be overcome before the presented system could provide clinical utility.
4.2 Future directions
Subsequent iterations of the voice-coil system will address the aforementioned limitations towards obtaining both optimal scanning rates as well as maximum imaging sizes. Our ability to image at maximum scanning frequencies was limited in various aspects. It was found that at high frequencies (e.g. >25 Hz at 2 mm scan widths), the rapid oscillation of the voice-coil stage mass resulted in mechanical vibrations throughout the stage system, thus preventing reliable image reconstruction. In this work, the voice-coil stage was mounted atop a vertical lift stage of a comparable size and mass. The mounting connection between these stages served as the source of mechanical instability during high frequency voice-coil oscillations, which was determined by testing high-frequency operation when the voice-coil was mounted atop the lift stage and comparing this to when the voice-coil was mounted directly to the stepper stage. A design with improved stability is needed in future work to mitigate these problems. In doing so, this would enable voice-coil operation at maximum scanning frequencies. At the current moving stage mass of 162 g, the voice-coil could be made to oscillate 300 $\mu$m, 500 $\mu$m, and 1 mm image widths at peak frequencies of 125 Hz, 97 Hz, and 68 Hz, respectively. At these frequencies, fine sampling resolution (400 nm) UV-PARS images of 300×300 $\mu$m$^2$, 500×500 $\mu$m$^2$, and 1×1 mm$^2$ areas could be scanned in 3 s, 6.5 s, and 18.3 s, respectively.
A limiting factor to achieving maximum scanning rates with fresh tissue was sample motion over the scanning trajectory. This limitation could be overcome by designing a more secure sample container. The current sample container was constructed using the materials specified in Fig. 4, in which the fresh tissue sample was contained laterally by a custom 3D printed sample container and vertically by both the microscope glass slide and cover slip. Future work is needed to ensure a no-slip condition of the tissue-glass interface at maximum oscillation frequencies.
To enable voice-coil UV-PARS scans over the full 12 mm travel range, future iterations could employ vertical compensation during scanning. This would eliminate the need for the system goniometers and would reduce the total moving stage mass to approximately 105 g, allowing for a maximum oscillation frequency of $\sim$25 Hz at the FTR of 1.2 cm. Assuming an image width of 1.2 cm and height of 0.83 cm, a full cm$^2$ region could be imaged at gross 1 $\mu$m sampling resolution in 2.8 minutes and at fine 400 nm sampling resolution in 7 minutes. Moreover, small FOVs of 300×300 $\mu$m$^2$, 500×500 $\mu$m$^2$, and 1×1 mm$^2$ could then be imaged in 2.4 s, 5.2 s, and 14.7 s, respectively. These rapid scanning frequencies would permit imaging of 1 cm$^2$ coarse resolution (1 $\mu$m) images in less than 3 minutes and fine resolution (400 nm) images of small FOVs in mere seconds, displaying imaging rates, resolutions, and image sizes sufficient for intraoperative applications. Vertical scanning could also be utilized in future works to explore multi-layer depth-resolved UV-PARS voice-coil scanning, similar to that shown in [23].
Additionally, this work did not compensate for pulse-to-pulse energy fluctuations, since they were less than 5 %. Future UV-PARS voice-coil imaging systems could incorporate a custom peak detector circuit [25,36] and stream an additional data channel to correct for pulse-to-pulse energy variation during imaging, thus enhancing image SNR.
5. Conclusions
The voice-coil UV-PARS scanning system demonstrated is an advancement of previous UV-PARS iterations, capable of providing fine-resolution scans while demonstrating high imaging speeds. In this work, we showed 2×2 mm$^2$ UV-PARS imaging at 500 nm sample resolution in 1.33 minutes as well as 4×4 mm$^2$ images of fresh murine liver at 900 nm sampling resolution in 2.5 minutes. While the demonstrated system has shown both the ability to image rapidly at both fine and coarse sample intervals, further work will be required to achieve the full potential of the voice-coil system. Future iterations of the UV-PARS voice-coil system will address both speed-based limitations associated with sample and mechanical system stability, as well as scan width limitations imposed by uneven sample surface morphology and unintended vertical motion of the stage tops over the scan trajectory. Successfully addressing these limitations would allow for cm$^2$ areas to be rapidly imaged with the voice-coil system without the need for specialized operator training, and could allow for future translation of UV-PARS scanning to an intraoperative setting.
Funding
Canadian Institutes of Health Research (PS 168936); Natural Sciences and Engineering Research Council of Canada (2018-05788); Canada Foundation for Innovation.
Disclosures
R.J.Z. is co-founder and shareholder of illumiSonics Inc. and Clinisonix Inc. which, however, did not support this work.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. T. L. Huston, R. Pigalarga, M. P. Osborne, and E. Tousimis, “The influence of additional surgical margins on the total specimen volume excised and the reoperative rate after breast-conserving surgery,” The Am. J. Surg. 192(4), 509–512 (2006). [CrossRef]
2. L. E. McCahill, R. M. Single, E. J. A. Bowles, H. S. Feigelson, T. A. James, T. Barney, J. M. Engel, and A. A. Onitilo, “Variability in reexcision following breast conservation surgery,” Jama 307(5), 467–475 (2012). [CrossRef]
3. D. Novis and R. Zarbo, “Interinstitutional comparison of frozen section turnaround time. a college of american pathologists q-probes study of 32868 frozen sections in 700 hospitals,” Archives of pathology and laboratory medicine 121(6), 559–567 (1997).
4. J. C. Cendán, D. Coco, and E. M. Copeland, “Accuracy of intraoperative frozen-section analysis of breast cancer lumpectomy-bed margins,” J. Am. Coll. Surg. 201(2), 194–198 (2005). PMID: 28832271. [CrossRef]
5. H. Cheng, B. P.-H. Chen, I. M. Soleas, N. C. Ferko, C. G. Cameron, and P. Hinoul, “Prolonged operative duration increases risk of surgical site infections: A systematic review,” Surg. Infect. 18(6), 722–735 (2017). [CrossRef]
6. D. Orringer, B. Pandian, Y. Niknafs, et al., “Rapid intraoperative histology of unprocessed surgical specimens via fibre-laser-based stimulated raman scattering microscopy,” Nat. Biomed. Eng. 1(2), 0027 (2017). [CrossRef]
7. T. C. Hollon, B. Pandian, A. R. Adapa, et al., “Near real-time intraoperative brain tumor diagnosis using stimulated raman histology and deep neural networks,” Nat. Med. 26(1), 52–58 (2020). [CrossRef]
8. B. Zhang, M. Sun, Y. Yang, L. Chen, X. Zou, T. Yang, Y. Hua, and M. Ji, “Rapid, large-scale stimulated raman histology with strip mosaicing and dual-phase detection,” Biomed. Opt. Express 9(6), 2604–2613 (2018). [CrossRef]
9. L. Cahill, M. Giacomelli, T. Yoshitake, H. Vardeh, B. Faulkner-Jones, J. Connolly, C.-K. Sun, and J. Fujimoto, “Rapid virtual hematoxylin and eosin histology of breast tissue specimens using a compact fluorescence nonlinear microscope,” Lab. Invest. 98(1), 150 (2018). [CrossRef]
10. Y. K. Tao, D. Shen, Y. Sheikine, O. O. Ahsen, H. H. Wang, D. B. Schmolze, N. B. Johnson, J. S. Brooker, A. E. Cable, J. L. Connolly, and J. G. Fujimoto, “Assessment of breast pathologies using nonlinear microscopy,” Proc. Natl. Acad. Sci. 111(43), 15304–15309 (2014). [CrossRef]
11. M. G. Giacomelli, B. E. Faulkner-Jones, L. C. Cahill, T. Yoshitake, D. Do, and J. G. Fujimoto, “Comparison of nonlinear microscopy and frozen section histology for imaging of mohs surgical margins,” Biomed. Opt. Express 10(8), 4249–4260 (2019). [CrossRef]
12. J. Kang, I. Song, H. Kim, H. Kim, S. Lee, Y. Choi, H. J. Chang, D. K. Sohn, and H. Yoo, “Rapid tissue histology using multichannel confocal fluorescence microscopy with focus tracking,” Quant. Imaging Med. Surg. 8(9), 884–893 (2018). [CrossRef]
13. F. Fereidouni, Z. T. Harmany, M. Tian, A. Todd, J. A. Kintner, J. D. McPherson, A. D. Borowsky, J. Bishop, M. Lechpammer, S. G. Demos, and R. Levenson, “Microscopy with ultraviolet surface excitation for rapid slide-free histology,” Nat. Biomed. Eng. 1(12), 957–966 (2017). [CrossRef]
14. A. K. Glaser, N. P. Reder, Y. Chen, E. F. McCarty, C. Yin, L. Wei, Y. Wang, L. D. True, and J. T. Liu, “Light-sheet microscopy for slide-free non-destructive pathology of large clinical specimens,” Nat. Biomed. Eng. 1(7), 0084 (2017). [CrossRef]
15. P. Hajireza, W. Shi, K. Bell, R. J. Paproski, and R. J. Zemp, “Non-interferometric photoacoustic remote sensing microscopy,” Light: Sci. Appl. 6(6), e16278 (2017). [CrossRef]
16. K. L. Bell, P. Hajireza, W. Shi, and R. J. Zemp, “Temporal evolution of low-coherence reflectrometry signals in photoacoustic remote sensing microscopy,” Appl. Opt. 56(18), 5172–5181 (2017). [CrossRef]
17. P. H. Reza, K. Bell, W. Shi, J. Shapiro, and R. J. Zemp, “Deep non-contact photoacoustic initial pressure imaging,” Optica 5(7), 814–820 (2018). [CrossRef]
18. K. L. Bell, P. Hajireza, and R. J. Zemp, “Coherence-gated photoacoustic remote sensing microscopy,” Opt. Express 26(18), 23689–23704 (2018). [CrossRef]
19. P. Kedarisetti, N. J. Haven, B. S. Restall, M. T. Martell, and R. J. Zemp, “Label-free lipid contrast imaging using non-contact near-infrared photoacoustic remote sensing microscopy,” Opt. Lett. 45(16), 4559–4562 (2020). [CrossRef]
20. N. J. M. Haven, K. L. Bell, P. Kedarisetti, J. D. Lewis, and R. J. Zemp, “Ultraviolet photoacoustic remote sensing microscopy,” Opt. Lett. 44(14), 3586–3589 (2019). [CrossRef]
21. N. J. M. Haven, P. Kedarisetti, B. S. Restall, and R. J. Zemp, “Reflective objective-based ultraviolet photoacoustic remote sensing virtual histopathology,” Opt. Lett. 45(2), 535–538 (2020). [CrossRef]
22. P. Kedarisetti, B. S. Restall, N. J. Haven, M. T. Martell, B. D. Cikaluk, J. Deschenes, and R. J. Zemp, “F-mode ultraviolet photoacoustic remote sensing for label-free virtual h&e histopathology using a single excitation wavelength,” Opt. Lett. 46(15), 3500–3503 (2021). [CrossRef]
23. B. S. Restall, P. Kedarisetti, N. J. Haven, M. T. Martell, and R. J. Zemp, “Multimodal 3d photoacoustic remote sensing and confocal fluorescence microscopy imaging,” J. Biomed. Opt. 26(09), 096501 (2021). [CrossRef]
24. B. S. Restall, N. J. M. Haven, P. Kedarisetti, M. T. Martell, B. D. Cikaluk, S. Silverman, L. Peiris, J. Deschenes, and R. J. Zemp, “Virtual hematoxylin and eosin histopathology using simultaneous photoacoustic remote sensing and scattering microscopy,” Opt. Express 29(9), 13864–13875 (2021). [CrossRef]
25. N. J. M. Haven, M. T. Martell, B. D. Cikaluk, B. S. Restall, E. McAlister, S. Silverman, L. Peiris, J. Deschenes, X. Li, and R. J. Zemp, “Virtual histopathology with ultraviolet scattering and photoacoustic remote sensing microscopy,” Opt. Lett. 46(20), 5153–5156 (2021). [CrossRef]
26. B. R. Ecclestone, K. Bell, S. Sparkes, D. Dinakaran, J. R. Mackey, and P. Haji Reza, “Label-free complete absorption microscopy using second generation photoacoustic remote sensing,” Sci. Rep. 12(1), 8464 (2022). [CrossRef]
27. M. Boktor, B. R. Ecclestone, V. Pekar, D. Dinakaran, J. R. Mackey, P. Fieguth, and P. Haji Reza, “Virtual histological staining of label-free total absorption photoacoustic remote sensing (ta-pars),” Sci. Rep. 12(1), 10296 (2022). [CrossRef]
28. M. T. Martell, N. J. Haven, and R. J. Zemp, “Fiber-based photoacoustic remote sensing microscopy and spectral-domain optical coherence tomography with a dual-function 1050-nm interrogation source,” J. Biomed. Opt. 26(06), 066502 (2021). [CrossRef]
29. Z. Hosseinaee, N. Abbasi, N. Pellegrino, L. Khalili, L. Mukhangaliyeva, and P. Haji Reza, “Functional and structural ophthalmic imaging using noncontact multimodal photoacoustic remote sensing microscopy and optical coherence tomography,” Sci. Rep. 11(1), 11466 (2021). [CrossRef]
30. B. R. Ecclestone, D. Dinakaran, and P. H. Reza, “Single acquisition label-free histology-like imaging with dual-contrast photoacoustic remote sensing microscopy,” J. Biomed. Opt. 26(5), 056007 (2021). [CrossRef]
31. B. S. Restall, B. D. Cikaluk, M. T. Martell, N. J. M. Haven, R. Mittal, S. Silverman, L. Peiris, J. Deschenes, B. A. Adam, A. Kinnaird, and R. J. Zemp, “Fast hybrid optomechanical scanning photoacoustic remote sensing microscopy for virtual histology,” Biomed. Opt. Express 13(1), 39–47 (2022). [CrossRef]
32. A. M. DeMarzo, W. G. Nelson, W. B. Isaacs, and J. I. Epstein, “Pathological and molecular aspects of prostate cancer,” The Lancet 361(9361), 955–964 (2003). [CrossRef]
33. T. Harrison, J. C. Ranasinghesagara, H. Lu, K. Mathewson, A. Walsh, and R. J. Zemp, “Combined photoacoustic and ultrasound biomicroscopy,” Opt. Express 17(24), 22041–22046 (2009). [CrossRef]
34. L. Wang, K. Maslov, J. Yao, B. Rao, and L. V. Wang, “Fast voice-coil scanning optical-resolution photoacoustic microscopy,” Opt. Lett. 36(2), 139–141 (2011). [CrossRef]
35. L. Wang, K. I. Maslov, W. Xing, A. Garcia-Uribe, and L. V. Wang, “Video-rate functional photoacoustic microscopy at depths,” J. Biomed. Opt. 17(10), 1 (2012). [CrossRef]
36. L. Snider, K. Bell, P. Hajireza, and R. J. Zemp, “Toward wide-field high-speed photoacoustic remote sensing microscopy,” in Photons Plus Ultrasound: Imaging and Sensing 2018, vol. 10494A. A. Oraevsky and L. V. Wang, eds., International Society for Optics and Photonics (SPIE, 2018), p. 1049423.
