1. Introduction

Surgery is a front-line procedure for the removal of cancerous solid tumors. Nevertheless, complete removal of tumor tissue is challenging. Tumor tissues are normally resected along with a rim of healthy tissue; however, if histopatholocial examination finds residual tumor tissues in this rim (a positive margin), an additional surgery may be required to remove more tissue. Such repeat surgeries are required in 20-40% of solid tumor resections in which complete removal of cancerous tissues in the first surgery is critical to prognosis, cost reduction, and mitigation of additional procedures [1,2]. Assessment of margin status involves lengthy histological analysis, not typically available during the surgical procedure. This is further complicated by an inability to image thick tissue samples requiring additional processing time. Currently, the gold standard of histological evaluation is hematoxylin and eosin (H&E) staining, a procedure that has remained relatively unchanged for the last 100 years.

Recently, several optical imaging approaches have been developed for the purpose of virtual histology as a replacement for H&E staining in solid tumor margin assessment. Each method offers its own benefits and drawbacks with regards to image acquisition time, the potential for interoperative imaging, depth of imaging in the tissue, and lateral resolution.

Optical coherence tomography (OCT) offers millimeter-scale depth imaging but as it is a scattering based contrast imaging modality, it lacks specificity to the cell nuclei and other subcellular components relevant for pathology. Additionally, OCT has difficulty with increasing sample variability and contamination proving increasingly difficult to interpret in clinical applications [3,4].

Microscopy with UV Surface Excitation (MUSE) uses added contrast agents in the form of fluorescent markers for wide field UV imaging as a low-cost alternative for nuclei and protein imaging at superficial depths. MUSE attempts to image superficial layers by taking advantage of limited UV depth penetration, however, since this is tissue dependent, imaging thickness is not always uniform. Optical sectioning capabilities are dependant on the illumination incidence angle but cannot exceed the depth of focus (DOF) of the objective limiting MUSE to low and moderate numerical aperture (NA) objectives. This imaging modality would not be suitable for intraoperative imaging but still demonstrates promise for ex vivo margin assessment with thin samples [5–8].

Non linear microscopy (NLM) encompasses imaging modalities such as two-photon (2P), three-photon (3P), second-harmonic generation (SHG), third-harmonic generation (THG), stimulated Raman spectroscopy (SRS), and coherent anti-stokes Raman scattering (CARS) microscopy. NLM has proven itself with excellent results shown in images of gross tissue sections as it can provide high resolution, high sensitivity, and reject stray light. Several groups have reported NLM optical sectioning up to a 100µm depth with an axial resolution up to 80nm with comparable image quality to H&E stained sectioning and the highest molecular sensitivity reported of all microscopy modalities to date [9–12]. However, these techniques typically require expensive equipment including femtoseconds lasers and specialized scanning optics which can cost significantly more than the other imaging modalities discussed here. Another, drawback of NLM is imaging speed. For example, multicolor SRS with linear decomposition may require 10–20s in a high-resolution image of 512 × 512 pixels to retrieve protein and DNA contrast with both high sensitivity and high spatial resolution, using a high NA objective lens [12].

Optical resolution photoacoustic microscopy (OR-PAM) uses optical absorption contrast to produce high resolution images of cell nuclei and cytochromes ex vivo and vasculature in vivo [13–16]. Unlike the aforementined all-optical imaging techniques, OR-PAM requires physical acoustic coupling with the sample which can limit clincal use. OR-PAM has achieved lateral resolution up to 0.7µm in reflection mode with DNA/RNA contrast in cell nuclei [17]. Typically, OR-PAM is reported to achieve an imaging rate of 10-100kHz, with up to 500kHz having also been demonstrated, showing its versatile imaging capabilities to image in vivo blood oxygenation in real-time [18]. Further work has been completed with a dual wavelength OR-PAM and AR-PAM multimodal system which uses the two wavelengths 532-nm and 1064-nm [19]. Cell nuclei imaging has been demonstrated in human cancer samples with strong agreement between OR-PAM and H&E stains. Further histological analysis has also demonstrated the ability to discern cancerous cells from lymphocytes [14].

Alternatively, photoacoustic remote sensing (PARS) is a label-free non-contact reflection mode imaging modality that operates by measuring the back-reflection modulation intensity of a continuous-wave, near infrared (NIR), low-coherence interrogation laser in response to a high energy pulsed laser. PARS demonstrates specificity for various endogenous contrast agents and can discern biomolecules based on optical absorption. PARS utilizes absorption contrast present within tissue by pulsing a nanosecond laser on a chromophore which undergoes a sudden change in temperature leading to thermoelastic expansion and the production of localized pressure waves. These pressure waves produce modulations in the refractive index of the tissue which is detected in the back-reflection of the NIR interrogation beam. The intensity of the PARS signal is directly related to the optical absorption which corresponds to the concentration of the chromophore [20–22]. PARS demonstrates advantages over OR-PAM by removing the need for physical acoustic coupling from the sample by using an all optical approach. With regards to imaging speed, PARS is only limited in repetition rate from confinement stresses allowing it to operate in real time with a higher repetition rate excitation laser [23]. Our approach using UV-PARS can potentially achieve DNA contrast with MHz pixel readout rates, enabling a 512 × 512 image to be acquired in a fraction of a second, thus nearly two orders of magnitude faster imaging speeds compared to NLM and OR-PAM. This is essential for intra-operative imaging and has been demonstrated with in vivo imaging and thick tissue sections [20,25]. Galvanometer and mechanical scanning has been utilized with PARS as two methods of imaging with a field of view varying from 0.01mm² to 0.16mm², respectively [24,25].

PARS was first introduced in 2017 as a novel imaging modality that detects endogenous contrast agents with previous work reporting the ability to detect and image nueclic acids and hemoglobin as well as differentiate between oxygenated and deoxygenated blood [20,24,26]. Improvements to PARS technology has increased resolution substantially from when it was first introduced as well as enhanced stability and decreased acquisition times whilst imaging a larger variety of chromophores. Specific wavelengths are created using a variety of methods including direct laser sources, frequency doubling, and stimulated Raman scattering. By exciting DNA/RNA at 266-nm the measured absorption is used to image cell nuclei ex vivo and is called ultraviolet-PARS (UV-PARS). This technology has been demonstrated as a virtual histology method to obtain a comparable image to a hematoxylin stained sample, one of the two stains used in H&E for histological analysis, in a matter of seconds compared to weeks [24,25]. The first UV-PARS system used an off-axis parabolic mirror for focusing, along with mechanical scanning [24]. A next-generation UV-PARS system was introduced using a reflective objective for higher resolution, down to 0.39µm [25,27]. Abasi et al., subsequently demonstrated UV-PARS imaging of human tissues with images obtained at depths of up to 50µm [28]. With regards to hemodynamic imaging, oxygenated, and deoxygenated hemoglobin have been excited using 580-nm and 565-nm, respectively, to measure sO2 from vasculature in vivo in real time by using faster laser repetition rates [23,26]. This was the first demonstration of dual-wavelength PARS with subsequent systems imaging ex vivo samples of cell nuclei and hemoglobin using a 266-nm and 532-nm excitation system [29]. Multiple wavelengths were previously considered in PARS imaging of histological tissue samples; however, lateral UV spatial resolution was reported as 1.2µm, which is courser than the needed submicron resolution for accurate histological analysis. This could in part be due to poor UV beam quality, and a moderate numerical aperture objective.

In this work we first demonstrate imaging capabilities in phantoms and tissue samples followed by in vivo imaging with a dual-wavelength system. This paper is the first to demonstrate UV-PARS based virtual histopathology of living subjects combined with 532-nm light for hemoglobin contrast whilst remaining close to the ANSI mandated maximum permissible exposure (MPE) limits. This work is important for assessing neovascularization, Mohs surgery, depth of invasion of tumors relative to vascular beds, and for the assessment of margin status in tumor resection procedures.

2. Methods

2.1 Ex vivo sample preparation

All biological samples were acquired in accordance with the Univeristy of Alberta’s Animal Care and Use Committee ethics guidelines and regulations. Mouse organ tissue sections were prepared from a nude mouse (Charles River, NU/NU). Tissues of intereset were disected following euthanization of the mouse. These tissues were then formalin fixed, embedded in paraffin blocks, and sectioned to the desired thickness between 4-30µm and laid on glass microscope slides. Prior to imaging, paraffin was removed from the tissue by first heating the slides at 60°C for 1 hour, followed by several 2-minute-long washes in 2 changes of xylene, 2 changes of 100% ethanol, 95% ethanol, and finally de-ionized water. Additionally, 100nm gold nanoparticles (742031, Sigma-Aldrich) were imaged for characterization of the system’s lateral resolution. Both tissue and nanoparticle samples were imaged between a microscope slide and a quartz UV-transparent coverslip (CFQ-2220, UQG Optics) wetted with DI water to keep the sample hydrated during imaging.

2.2 In vivo live animal imaging

Charles River nude mice were used for this study. All laboratory animal protocols were approved by the Univeristy of Alberta’s Animal Care and Use Committee. Isoflurane was used at ∼2% for anesthesia during the imaging session. The animal was positioned on a custom-made laboratory animal holder. The ear was flattened between a microscope slide and coverslip before imaging to reduce topographical abnormalities.

2.3 System operation

Figure 1 shows the system setup diagram for UV-PARS. A 532-nm pulsed fiber laser (GLP-10, IPG Photonics) operating at 20kHz delivers nanosecond pulses. The excitation beam is split towards a 350-MHz bandwidth photodiode (DET10A, Thorlabs) for triggering the data acquisition card. The 532-nm path undergoes Galilean beam expansion using a set of planoconvex (LA1986-A-ML, Thorlabs) and planoconcave (LC1120-A-ML, Thorlabs) lenses to expand the beam to the desired width of 5mm to maximally fill the aperture of the reflective objective. Along the SHG path, the excitation beam is focused through a 5×5×6-mm Caesium Lithium Borate (CLBO) non-linear crystal (Eksma Optics) by a planoconvex lens (LA1464-A-ML, Thorlabs). Separation of the generated 266-nm from the 532-nm wavelength light is accomplished using a UV transparent CaF₂ prism (PS862, Thorlabs). The diverted 532-nm light is collected in a beam dump while the 266-nm excitation beam is expanded with a planoconvex (LA4327-UV-ML, Thorlabs) and planoconcave (LC4796-UV, Thorlabs) lens system to the desired beam width of 5mm. The 532-nm and 266-nm beams are co-focused with the 1310-nm linearly polarized low coherence continuous super luminescent diode laser (SLD1018PXL, Thorlabs). This diode laser is used as the interrogation beam by fiber coupling it through a zoom collimator (ZC618APC-C, Thorlabs) to allow the adjustment of the beam size from 1.07 - 3.27mm. The interrogation light passes through a half wave plate (WPH05-1310, Thorlabs) to adjust the linear polarization angle then passes through a planoconvex (LA1433-C-ML, Thorlabs) and planoconcave (LC1715-C-ML, Thorlabs) lens system for further beam expansion. This interrogation beam then passes through a polarized beam splitter (PBS, CCM1-PBS254, Thorlabs) followed by a quarter wave plate (QWP, WPQ10M-1310, Thorlabs) to convert the beam from linear to circular polarization. The 1310-nm beam is combined with the 532-nm beam using a dichroic mirror (DMSP1000R, Thorlabs) followed by combination with the 266-nm beam using a harmonic beam splitter (HBSY134, Thorlabs) which transmits the 1310-nm and 532-nm beams and reflects the 266-nm beam. All three beams continue through to the galvanometer mirror set (GVS412, Thorlabs) and were co-focused and raster scanned onto the sample with an aluminum coated reflective objective with an NA of 0.5 (LMM-40X-UVV, Thorlabs). By using a reflective objective, we eliminate chromatic aberrations thereby ensuring the focal points of all three wavelengths of light used occur at the same depth. Pulse energies for the 532-nm and 266-nm light were 80nJ and 5nJ respectively with an average interrogation power of 7mW on the sample with the 1310-nm light. A retractable mirror was used to enable additional brightfield imaging using an electron-multiplying charge coupled device (EMCCD) camera (DV-885K-CSO-#VP, Andor) with transillumination.

 

Fig. 1. System diagram of PARS microscope. Components can be identified with: beamsplitter (BS), second harmonic generator (SHG), beam dump (BD), photodiode (PD), galvanometer mirrors (GM), reflective objective (ROBJ), objective lens (OBJ), half-wave plate (HWP), polarized beam splitter (PBS), collimator (C), quarter-wave plate (QWP), balanced photodiode (BPD), avalanche photo diode (APD), and electron multiplied charge coupled device (EMCCD) camera.

Download Full Size | PDF

The back-reflected 1310-nm beam returns through the same path but with a reversal in its circular polarization direction. This beam then passes through the QWP and is converted from circular back to linear polarization. It is then redirected as it passes through the PBS and focused with a semi plan objective (3.2/0.10 160/-, Zeiss) onto a 75-MHz bandwidth balanced photodiode (PDB420C-AC, Thorlabs). The amplified RF voltage signal exiting the photodiode is bandpass filtered using a 1.8-MHz high-pass filter (EF509, Thorlabs) and a 11-MHz low-pass filter (BLP-10.7+, Mini-circuits). Voltage data from the galvanometer was acquired during the entirety of the scan to determine the laser spot position for each excitation pulse. The mirrors operated at approximately 60-Hz and 0.05-Hz with a 1° optical scan angle. 200,000 data points were taken at the laser pulse repetition rate (PRR) of 20kHz for an acquisition time of 10s for each wavelength and an overall imaging time of 20s. All data was acquired using the data acquisition card (CSE1242, Gage Applied). The signal intensity at a given interrogation point was produced by taking the maximum of the Hilbert transform of the voltage signal from the balanced photodiode. Data was filtered using 1.8 MHz to 11 MHz Gaussian response bandpass filters, then interpolated using Delaunay triangulation onto a Cartesian grid in MATLAB where the image was further processed with averaging, median and gaussian spatial filtering. In vivo images were acquired using a mosaic strategy with both optical scans of ∼100µm x100µm tiles and mechanical scanning to step between tiles.

2.4 Resolution characterization

To characterize the lateral resolution of both systems, 100-nm gold nanoparticles were imaged with both 266-nm and 532-nm excitation. The nanoparticles imaged were suspended in deionized water then selected based on finding two nanoparticles with the shortest distance between them and both with a distinguishable FWHM. The resolution was determined using Gaussian fit plots of the nanoparticle’s intensity with smallest resolved distance between two nanoparticles peaks. The resolution was determined to be 0.44µm for the 266-nm wavelength and 1.2µm for the 532-nm excitation as shown in Fig. 2.

 

Fig. 2. Resolution characterization. a) Image of 100-nm gold nanoparticle using 266-nm excitation. Scale bar is 0.1 µm. b) Cross-section profile used to determine lateral resolution by selecting two nanoparticles and fitting a gaussian distribution over the nanoparticle profile intensity. Peak to peak distance was measured between the closest two nanoparticles with a discernible FWHM. c) Image of 100-nm gold nanoparticle using 532-nm excitation. Scale bar is 1 µm. d) Cross-section profile used to determine lateral resolution by measuring the distance between the two peaks.

Download Full Size | PDF

3. Results

Several different tissue samples were taken from the kidney, lung, and heart with thickness ranging from 4-30µm to demonstrate the versatility of this system. Initial presentation of UV-PARS is shown in comparison with a hematoxylin stain, the gold standard of histological analysis, to demonstrate the excellent agreement between these two histological methods to resolve cell nuclei in 4µm thin tissue sections. Figure 3(a) shows a hematoxylin stained brightfield image of the sample compared to UV-PARS image in Fig. 3(b) of lung alveoli. Furthermore, both images show some background signal from the cytoplasm which is due to the presence of some UV absorbing components such as RNA and proteins in the cytoplasm and intracellular fluid.

 

Fig. 3. Sectioned mouse lung tissue 4 µm thick. Scale bar is 20 µm. a) Hematoxylin stain brightfield image; b) UV-PARS image comparison with 266-nm excitation with demonstration of the excellent agreement between the two imaging modalities.

Download Full Size | PDF

Figure 4 contains images taken from a 30µm section of thick mouse heart (Fig. 4(a-d)) and kidney (Fig. 4(e-h)) tissue. The section was first scanned with the 266-nm excitation beam to acquire an image of cell nuclei (Fig. 4(a, e)). This was followed by imaging with the 532-nm excitation for hemoglobin contrast (Fig. 4(b, f)) and finally imaged with an EMCCD for a brightfield comparison as a reference (Fig. 4(c, g)). These dark patches have good colocalization with the PARS hemoglobin signal and help provide insight into the areas with little or no cell nuclear contrast. Although hemoglobin and cell nuclei are the target molecules, with large absorptions peaks at the excitation wavelengths, other biomarkers will absorb these wavelengths as well which can provide additional cell morphology information but at the cost of less defined nuclei and vasculature boundaries. Differences in image registration between the brightfield microscopy images and PARS hemoglobin contrast may be due to insufficient fluence to excite a detectable absorbance response (Fig. 4(c, g)). The two PARS images were superimposed onto one another and assigned separate color maps with red and blue for hemoglobin and nucleic acids contrast respectively (Fig. 4(d, h)). Gaps between cell nuclei in the 266-nm excitation image are realized to be vasculature with the additional information gained from the hemoglobin contrast. Cell nuclei in sectioned tissues are well defined with a SNR of 42dB while red blood cells were identified with green light excitation demonstrating an SNR of 41dB. SNR was calculated by comparing the noise standard deviation (${\sigma _N}) $ to the mean signal of the sample ($\bar{S})\; $: $SN{R_{dB}} = 20lo{g_{10}}({\bar{S}/{\sigma_N}} )$.

 

Fig. 4. Images were taken from a 30 µm sectioned mouse heart (a-d) and kidney (e-h) tissue. Scale bar is 10 µm for all images. UV-PARS image taken using 266-nm excitation wavelength in both a) and e); PARS image taken using 532-nm excitation wavelength in both b) and f); brightfield image with a purple outline of 532-nm PARS image added for reference in both c) and g); both excitation wavelengths of PARS are superimposed onto each other with DNA as blue and hemoglobin as red in d) and h).

Download Full Size | PDF

In vivo vasculature and cell nuclei images were obtained from a mouse ear. These images were obtained sequentially showing the vasculature images obtained using the 532-nm excitation (Fig. 5(a)), and cell nuclei imaged at the same depth with the 266-nm excitation but with a smaller field of view (Fig. 5(b)). Vasculature images were obtained with an SNR up to 45dB, comparable to our previous work [30]. Cell nuclei images were obtained by imaging in a mosaic pattern and demonstrated an SNR up to 49dB.

 

Fig. 5. In vivo images taken from a mouse ear. (a) Vasculature images were taken using a 532-nm excitation source. Scale bar is 60 µm. (b) Cell nuclei images were taken in a mosaic pattern using a 266-nm excitation source. Scale bar is 10 µm.

Download Full Size | PDF

4. Discussion

We have presented a dual-wavelength photoacoustic remote sensing microscopy system using 532-nm and 266-nm excitation wavelengths to image hemoglobin and cell nuclei at a subcellular scale in both in vivo and ex vivo models. We achieved a 0.44µm lateral resolution with 266-nm light and tight optical sectioning and improved the lateral resolution with the 532-nm excitation source to 1.2µm compared to the previously reported 1.5µm [29]. Together with nuclei visualization using UV-PARS, blood oxygenation can also be used to provide complementary information during tumor resection surgeries. Images have an inherent sectioning capability due to the DOF which has been characterized to be 6-12µm with UV-PARS, less than the average thickness of a cell layer (10µm), allowing optical sectioning to be possible with PARS in both thick tissue and animal models [24]. Single cell layer imaging capability allows histologists to have a more precise diagnosis of cancer boundaries by specifying the depth of penetration. Further applications may include decreasing Mohs surgical times by decreasing tissue analysis time. Mohs surgery is a precise treatment with exceptional success rates for skin cancer where the surgeon will progressively remove the visible cancer for examination until no cancer remains. Typically, the wait time for examination takes up the majority of the procedure time as it requires freezing, cutting, staining with H&E then analysis of the tissue [31]. PARS could reduce the time for analysis allowing faster procedure times leading to a reduction in cost and an improvement patient care.

This system demonstrates acceptable fluence rates for 1310-nm, 532-nm and 266-nm which are below the ANSI mandated maximum permissible exposure (MPE) limits [32]. Currently, our UV fluence is similar to that used in UV-PAM, where no obvious tissue damage was observable after a single scan [15]. Mouse ear images were taken at the same depth with both 266-nm and 532-nm excitation wavelengths. The first layer of tissue in a mouse ear with both vasculature and cell nuclei present is the dermis layer well below the surface of the epidermis or about 13.3-16µm in nude male mice around 1 year old [33,34]. MPE limits can be determined from this to extrapolate laser fluence on the skin surface for live subjects. UV fluence levels are within ANSI limits for a single pulse MPE with UV wavelengths on the skin by utilizing the moderate NA objective and low pulse energy. With our optical focus >13.3µm below the skin and with a focal spot diameter of 0.44µm, the focal spot on the skin will be at least 15.44µm in diameter. The average energy deposited on the skin surface can be calculated to be 2.67 mJ/cm² which is below the photochemical damage threshold of 3mJ/cm². However, in this application, light is delivered to a small localized area therefore the MPE for a pulse train must be considered as well. The average number N of overlapping laser pulses can be calculated by dividing the surface spot size by scan step size: $N = 15.44\mathrm{\mu }\textrm{m}/0.22\mathrm{\mu }\textrm{m} = 70$. UV thermal damage for 266-nm is equal to $MP{E_{train}} = 0.56{t^{0.25}} = 136mJ/c{m^2}$ with t equal to the exposure time, $t = N/PRR = 70/40k$, of 3.5ms and the MPE_average equal to $MP{E_{train}}/N = 1.94mJ/c{m^2}$. These calculations demonstrate that a single UV pulse is within acceptable limits, however, with our high repetition rate laser, our fluence is 37.6% above the ANSI limit [32]. This could be decreased to be within ANSI limits by increasing the scan area or step size but at the cost of decreased lateral resolution or reducing the PRR at the cost of imaging speed. Even if higher pulse energies are used in other applications such as solid tumor surgery (eg: Mohs surgery), such tissue layers would be surgically removed and therefore not subject to the above ANSI limits. By using similar calculations to the above UV limits, 532-nm light will have an average energy deposited on the skin of 2.33mJ/cm². This is under the ANSI MPE limit of 185mJ/cm² for visible light as well as the MPE_average as calculated to be 2.64mJ/cm² [32].

5. Conclusion

This paper presents a dual-wavelength PARS system capable of imaging cell nuclei and hemoglobin in both ex vivo and in vivo samples with subcellular resolution. Our fluence was determined to be within MPE limits for 532-nm light and slightly above for 266-nm UV light. Without the requirement for exogenous contrast and the non-contact nature of this imaging modality, dual-wavelength PARS has the potential for a simpler, faster intraoperative margin analysis of resected cancerous tissue. Optical imaging for visualizing nucleated cell layers and vascular beds with complimentary contrast was demonstrated with a 0.44µm lateral resolution of cell nuclei and a 1.2µm lateral resolution of hemoglobin. This work represents the first documentation of in vivo dual-wavelength PARS of cellular nuclei and hemoglobin. Images were acquired in both unstained thick and thin mouse tissue sections of kidney, lung and heart samples as well as a mouse ear in vivo. PARS demonstrates the ability to co-align multiple excitation wavelengths and the enormous potential to resolve multiple endogenous contrast agents. Future work is needed to improve imaging speeds, decrease UV fluence within MPE limitations, validate against frozen section pathology in tumor excision applications, and expand contrast to include proteins and lipids.