1. Introduction

Detection of cancers at early stages can reduce mortality and morbidity significantly and alleviate a significant burden on the healthcare system. Many human cancers emerge in epithelial tissues, triggering metabolic changes at the cellular level initially, before large-scale morphological differentiations occur [1–4]. However, the current gold standard cancer diagnosis is based on visual inspection and biopsy collection for histopathology to analyze morphological changes only [5,6]. Imaging modalities assisting visual inspection – such as endoscopy, colposcopy, and colonoscopy – are similarly limited to provide morphological information from the tissue surface only. Moreover, these modalities cannot offer cellular-level resolution needed for detection of precancerous changes. Histopathology itself is problematic in that it is invasive, labor-intensive, and time consuming, since it involves multiple steps, such as fixing, staining, and sectioning of the biopsy samples [7]. Individual sections are evaluated, one at a time, by a specialist, who has access only to morphological information. Since morphological changes associated with cancer development can be caused by benign conditions also, specificity of the traditional diagnosis pipelines is compromised. Sensitivity can be reduced also, since realistically only a limited number of samples can be collected and analyzed with current methods.

Diagnosis speed and accuracy of cancers can be increased by using additional sources of information, such as metabolic states of cells [8,9]. It is a well-studied phenomenon that cancer incidences are linked to alterations in cellular metabolism: In many cases, emergence of cancers is marked by a tendency of increased aerobic glycolysis rates [9–11]. Clinical relevance of metabolic analysis modalities is evident in current use of PET and MRI in service to cancer diagnosis and treatment assessment [12,13]. However, PET and MRI lack the cellular level resolution needed for cancer diagnosis at early stages. In fact, the only label-free metabolic imaging method capable of cellular level resolution is multiphoton, such as 2-photon (2p), autofluorescence microscopy. Being a nonlinear optical imaging method that uses near infrared laser pulses, multiphoton microscopy also provides the ability to image into intact tissue at depths covering the entirety of most epithelial tissues and further to offer depth-dependent characterization [14,15]. Label-free, two photon excited fluorescence imaging studies performed with engineered epithelial tissue models of cervical pre-cancers as well as freshly excised human cervical tissues demonstrate the sensitivity of this approach to metabolic reprogramming [4]. Specifically, it was shown that depth dependent alterations in the optical redox ratio, defined as the ratio of FAD/(NAD(P)H + FAD) TPEF images, and mitochondrial organization assessed from Fourier analysis of NAD(P)H images is sensitive to changes in the relative levels of oxidative phosphorylation, glycolysis, and glutaminolysis, associated with the over expression of the HPV E6 and E7 oncoproteins [16–18].

Studies with 2p autofluorescence microscopy of excised biopsy samples yielded promising results for early cervical cancer detection [4,19]. These results motivate the development of a hand-held probe, that miniaturizes 2p imaging modality, for assessment of multiple suspicious-appearing areas on the cervix, following colposcopic inspection. Clinical use of complete probe systems designed around this goal can potentially eliminate the need to collect biopsies altogether and perform on-spot diagnosis [20]. One significant design challenge impeding clinical translation of 2p autofluorescence imaging probes for cancer diagnosis is the challenging task of autofluorescence collection, which is inherently weak [21,22]. In clinical settings, collection of autofluorescence would be further hindered by tissue scattering, particularly at higher imaging depths. Therefore, it is crucial to implement high efficiency autofluorescence collection approaches in 2p autofluorescence imaging probes. Several probe systems with 2p autofluorescence imaging capability demonstrated in the literature handle autofluorescence collection mainly in two ways [22–33]. The first way uses double clad fibers (DCFs) in a way that the fiber core is dedicated to delivery of the excitation pulses and the inner cladding is reserved for autofluorescence collection [22–24,26–28,31,33]. The advantage here is to have the collection element axially aligned with the focal spot. However, sizes of commercially available DCFs limit the overall achievable collection area, which can limit the collection efficiency and the imaging depth [28]. The necessity that the collection elements must interface with the tissue through the distal optics can put additional complexity in design of the distal optics also. Additional care in laser coupling to fiber core is also needed due to the requirement to direct the collected signal coming from the second cladding in the proximal end of the fiber with dichroic mirror [34]. The second approach to autofluorescence collection is to couple the autofluorescence signal to another fiber dedicated to collection, using optical elements in front of or behind the objective, which can create additional design and alignment challenges [25,29,30].

In this work, we present a 2p autofluorescence probe uniquely designed to target a high collection efficiency to enable metabolic imaging for early detection of cervical cancers in a clinical setting [35,36]. For collection, our design employs 12 0.5 NA multi-modal fibers arranged around the probe distal optics in a radially symmetric manner. This configuration increases the overall collection area significantly while reducing complexity since it renders collection and excitation paths completely independent. Cleaving the collection fibers at a finite angle also increases the directivity and collection cone angle of the collection and increases collection efficiency per fiber, particularly at high scattering or high imaging depths. Our probe is also capable of collecting images at different depths, owing to axial actuation via a dc servo motor. This is significant, since published examples of metabolic tissue characterization for cancer detection often relies on depth-dependent variation of the parameters of interest [14,37,38]. Other features of the probe include flexible delivery of ultrashort pulses with minimal nonlinear effects using a hollow core photonic bandgap fiber and focusing to sub-micron spot size by designing a miniaturized objective consisting of commercially available lenses. In the upcoming sections, we will discuss the design requirements to address and simulation results to demonstrate fundamental advantages of the fluorescence collection method we propose here. Following a detailed description of the overall optomechanical design and assembly steps, we will present characterization of probe properties and 2p autofluorescence images of fresh porcine vocal fold epithelia.

2. Probe design and simulations

2.1 Design considerations

Our approach to design of a hand-held, 2p cervical tissue imaging probe prioritizes increasing the collection efficiency to achieve autofluorescence detection in a clinical setting, provided that the following design considerations are met. 1) Maintenance of collection efficiency over the entire depth of imaging in a highly scattering tissue. 2) Delivery of ultrashort pulses of high peak intensities to the tissue over a flexible path, namely through optical fibers, without initiating nonlinear effects and, thus, damaging the fiber [30,39]. 3) Compatibility with the 2p excitation wavelengths of endogenous intracellular fluorophores, that are considered important metabolic markers, such as NAD(P)H and FAD [19,40–43]. 4) A miniaturized excitation objective to focus the laser pulses coming out of the fiber into the tissue with a sub-micron imaging resolutions to enable depth-dependent characterization of the tissue metabolism at sub-cellular levels. 5) A miniaturized beam scanning mechanism to achieve the lateral scanning of the focal volume over a field of view (FOV) of > 100 µm without significant deterioration in resolution. 6) Axial scanning mechanism to allow a depth scanning up to 200 µm for characterization of cervical epithelia with increased thicknesses. 7) Image acquisition rates > 1 fps that would be compatible with a clinical operation.

All items constituting the hand-held probe must be packaged within an optomechanical design, allowing for easy manipulation of the probe and access to the region of interest by the clinician. The invasive elements of the probe must be covered with a detachable and sterilizable cap to isolate the axially actuated probe elements from the tissue and provide sanitization.

2.2 Key design concepts and elements

For improved fluorescence collection efficiency, our probe uses 12 multimode collection fibers with 0.5 NA and 735 µm core diameter (Edmund Optics), which yields a total collection area of 5.1 mm². We show the arrangement of the multitude of collection fibers around the excitation objective with a finite bend and termination angle at the distal end in Fig. 1(a). This configuration increases the optical etendue by increasing the total collection area and collection directivity, since collection cone angles shift towards the fluorescence source with angled termination, while offering a radially symmetric collection scheme. With increased etendue, we aimed our probe to collect much of the scattered fluorescence photons even at high imaging depths. The radial symmetry of the system offers an independence of collection efficiency throughout the FOV. Moreover, since collection fibers do not interact with excitation arm at all, we have more freedom in design of collection and excitation arms and face less challenges in maintaining accuracy in fabrication and assembly.

 

Fig. 1. Opto-mechanical design of the probe. a) Side cross-section of the distal optics, showing the angled bend and termination of collection fibers. b) Rendered 3D view of the distal optics, enclosed by SLA printed fiber holder (yellow). c) Side cross-section of the probe design. d) Axial actuation changing imaging depth (ID), which is the distance from the cap window to the focal spot position. Since the working distance (WD) of the excitation objective is fixed, axial actuation of the distal optics causes the focal spot to move with respect to the cap. The red triangle represents laser focusing, with an exaggerated WD for clarity. Other elements are drawn to scale.

Download Full Size | PDF

A stereolithography (SLA) printed piece holds the collection fibers and houses the excitation objective and excitation fiber at its center, thus packaging the entire distal optics, as shown in Fig. 1(b). To minimize losses due to fiber bending, the collection fibers around the excitation objective are laid out in soft curvatures. The excitation objective consists of commercially available lenses with 3 mm diameter to keep probe diameter within reasonable limits. We use a piezoactuator tube (Custom design, EBL Products) for lateral scanning of our excitation fiber, which is anchored to the piezoactuator tube like a resonant cantilever beam, as demonstrated in previous probes [25,39,44,45]. The excitation fiber is a hollow core photonic bandgap fiber (HC-800, NKT Photonics) with its zero-dispersion wavelength at 775 nm, coincides with 2p excitation bands of both NAD(P)H and FAD molecules. Emission photons yielded by these molecules are collected by a low noise, high gain photomultiplier tube (PMT) module (H7422PA-40, Hamamatsu), through a fluorescence filter compatible with the emission band of interest (BG-39, Schott).

Overall optomechanical design of the probe is shown in Fig. 1(c). The design consists of two main halves. The proximal half, serving as a handle, houses the dc servo motor (Z806, Thorlabs) addressing our needs with 6 mm axial actuation, sub-micron movement resolution, and enough torque to translate stainless steel material load. The distal half, intended to be the invasive element of the probe coming in contact with cervical tissue, is made of tubing sliding within each other to translate the axial actuation to the distal optics with assistance of commercially available springs and ball sliders. The distal half of the probe is covered by a sterilizable cap with an aperture on the distal end, covered by a circular sapphire window.

The aim of axial actuation is to be able to image tissue at different depths, as visualized in Fig. 1(d). With the distal aperture of the cap held stationary and in contact with the tissue, dc servo motor actuation causes axial movement of the distal optics, through translation of actuation by tubing within distal half of the probe. Since the working distance (WD) of the excitation objective is fixed, such actuation causes an axial movement of the laser focal spot within the tissue. With the axial actuation of whole distal optics inside the cap, distance between the collection fibers and the focal spot within tissue remains constant, which helps preservation of collection efficiency at different imaging depths.

2.3 Design of the excitation objective

We tested different commercially available lenses of 3 mm diameter in Zemax environment to design an excitation objective with the highest focusing NA possible while achieving the largest WD and imaging FOV. We set the design wavelength at the zero-dispersion wavelength of our excitation fiber at 775 nm and used its measured divergence of 0.1 rad (0.10 NA) at this wavelength. Figure 2 shows the finalized excitation objective design, which consists of a 9-mm focal length achromatic doublet lens (45-090, Edmund Optics) and a 2-mm focal length aspheric lens (355151, Lightpath). The design included a 110 µm thick c-plane sapphire window in contact with the tissue, modeled as sea water. With an optimized design, we were able to achieve a 1/e² IPSF diameter of 1.34 µm, corresponding to an objective NA of 0.52 with simulated lateral and axial 2p resolutions of 0.56 µm and 4.42 µm, respectively [46]. Clear aperture ratings of the selected lenses set the maximum attainable FOV of the system as 130 µm, which is achievable with a fiber deflection of 330 µm from the axis in one direction. Figure 2(b) additionally shows spot diagrams for imaging depths of 50 µm and 200 µm on axis and at the edge of FOV, which we determined as 120 µm. This choice of FOV is based on the Strehl ratio dropping to 0.8 at radial position of 60 µm for the imaging depth of 200 µm.

 

Fig. 2. Zemax simulation results of excitation objective. (a) The finalized configuration of lenses with input and output parameters indicated. (b) Spot diagrams at the center and edge of the 120 µm FOV for imaging depths of 100 µm and 200 µm.

Download Full Size | PDF

2.4 Collection arm simulations

To study the fluorescence collection efficiency of the proposed fiber bundle method, we performed Zemax simulations by changing relevant parameters and comparing this method against DCF collection. In this paper, we use the term probe collection efficiency to signify the percentage of generated photons that can be captured by the collection fiber(s). Figure 3(a) shows the simulation model, that includes seawater representing tissue, sapphire window, a single collection fiber, excitation objective and the DCF selected for this study. Between the sapphire window and collection fiber / excitation objective set is an air gap that would remain within the cap. In these simulations, an isotropic point source models the fluorescence volume, embedded in a medium that scatters according to Henyey-Greenstein model with g = 0.9. We model the medium with sufficient thickness to take into consideration the effect of back-scattered photons. While the ray count is set to 10⁶, fluorescence photon wavelength is assumed to be 500 nm, representing the average emission wavelength for NAD(P)H and FAD molecules. The axial position of the point source within the scattering medium representing the imaging depth, we simulated probe collection efficiency at different imaging depths and scattering length (l_s, always referenced to excitation wavelength of 775 nm in this paper) and for various fiber termination angles, namely different bending angles. We selected two l_s, values of 25 µm and 100 µm, representing high and normal scattering in tissues [47–50]. Based on the λ^-1.2 dependence of scattering length for soft tissues [49,50], at the average emission wavelength of 500 nm, we used the emission scattering lengths of 15 µm and 59 µm in the simulation model. For comparison, we also simulated a hypothetical scenario where our system employed DCF fluorescence collection for the objective design finalized above in conjunction with a DCF (F-SMM900, Fibercore), used in previous studies [24,28].

 

Fig. 3. Zemax simulation results estimating probe collection efficiency. a) Simulation model used to estimate probe collection efficiency for the presented collection method as well as for a DCF collection method, based on the commonly used DCF F-SMM900 and the objective design presented in Fig. 2. Detectors at different positions show the percentage of the sourced photons collected within an area of interest, at an imaging depth of 2l_s (where l_s= 25 µm at 775 nm) and fiber angle of 25°. b) Probe collection efficiencies of proposed collection approach, considering different fiber angles, and DCF collection scenario as a function of imaging depth. In the case of the proposed method, probe collection efficiency plots include the contribution of all 12 fibers. c) Effect cut angle (Θ) on collection directivity. Angle limits of accepted rays (Φ_t from axis to top and Φ_b from axis to bottom) depend on whether these rays can undergo total internal reflection at the core-cladding interface. The graph plots the change of these limit angles with respect to Θ, which indicates that a finite Θ increases the directivity of collection, considering the position of the fluorescence source. Two example cases of Θ are also illustrated with specified Φ_t and Φ_b show how fiber angles change collection directivity. Illustrations are not to scale.

Download Full Size | PDF

Figure 3(a) presents percentages of the sourced photons collected within areas of interest at different positions in the model, when imaging depth is 2l_s (l_s = 25 µm) and collection fiber angle is 25°. Looking at DCF collection path, while the excitation objective can collect a significant percentage of emitted photons (31.8% at the front), only 1.39% of photons can reach the second cladding of the DCF. This is an expected result since the objective can mainly direct to DCF tip the photons coming from the focal spot and its close vicinity only, due to the bidirectionality of the optics. This result implies DCF collection generally relies on ballistic photon collection, within a tolerance determined by the rather limited fiber area and NA of the second cladding. Therefore, we should expect DCF collection to suffer in high scattering or high imaging depth conditions, as suggested by other work in the literature [22,28]. In contrast, large NA collection fibers directly interfacing with the tissue accept photons over a wide range of incidence angles, even after they scattered multiple times as evident in the collection efficiency of 0.52% for each fiber, adding up to 6.24% for all 12 fibers positioned around the objective.

Figure 3(b) plots the probe collection efficiency achieved by either collection method against different imaging depths for different fiber angles and two different scattering lengths. Collection efficiency of 12 fibers increases for larger angles, however, with diminishing returns. This effect is more visible when the scattering is high, but nevertheless becomes negligible beyond 25°. Motivated by this trend, we chose to bend and terminate collection fibers at 25° in our probe design, considering that increased angles add to the total diameter of the probe design. Figure 3(c) illustrates how increased fiber angles increase directivity of collection. Here, we show a conventional optical fiber with termination angle Θ and acceptance limit angles in top and bottom directions, Φ_t and Φ_b. The acceptance limit angles are determined such that rays incident on fiber surface at these angles are refracted into fiber core at angles that correspond to critical angles for the subsequent refraction at core-cladding interfaces. Using the refractive index values for our collection fibers and Snell’s Law, we can analytically calculate Φ_t and Φ_b change with respect to Θ as plotted in Fig. 3(c). When Θ is zero, both Φ_t and Φ_b have equal magnitudes as would be expected from a 0.5 NA fiber. As Θ increases, both Φ_t and Φ_b shift, effectively altering the directivity of collection towards the position of the fluorescence source. Acceptance limit angles for two example cases of Θ = 0° and Θ = 20° are also illustrated for clarity. For high collection efficiency, it is desired to have a wide and well-directed span between Φt and Φ_b that can address scattered rays coming from the source.

Comparing the proposed method with DCF collection, we see that the proposed method clearly outperforms DCF collection, particularly at high scattering and high imaging depth conditions. At shallow depths (l_s < 1) probe collection efficiency at surface reaches 7.1% and 7.6% for high and medium scattering conditions, respectively, whereas DCF collection efficiency is limited to 2.0%. Importantly, the DCF collection shows an exponential decay with depth while the proposed method follows a more linear decrease, with decay rates of -0.058l_s and -0.066l_s for medium and high scattering conditions, respectively. Further simulations assuming a non-scattering medium suggest that, for the results presented in Fig. 3(b), ballistic photons constitute up to 45% of the photons collected by DCF, whereas the contribution of ballistic photons do not exceed 5% for the proposed method. This result further supports the idea that the proposed method is better suited for collection at high scattering lengths and high imaging depths.

The radial symmetry of the proposed collection method can compensate for the deflection of the focal spot from the neutral axial position during fiber scanning. Additional simulations indicate statistically insignificant changes in collection efficiency with the proposed method with spot deflection to the edge of maximum expected FOV. In contrast, collection efficiency reduction with DCF scenario can be as high as 64% when the focal spot is positioned at the maximum expected FOV edge at low imaging depths.

To efficiently direct all collected photons from 12-fibers into the detector, we designed a 5-lens collection system. For the detection of fluorescence, we use a low-noise, high sensitivity PMT module, which houses a thermoelectric cooler unit, that increases the distance between the PMT photocathode and front aperture. Delivering collected fluorescence to a photocathode buried inside a device using a bundle of 0.5 NA fibers creates a challenge in efficient coupling of fluorescence signals to photocathode. We expect the coupling efficiency of individual fibers to be different based on where they are located within a packed bundle of 12 fibers as demonstrated in Fig. 4(a), yielding a certain overall coupling efficiency. According to Zemax simulations shown in Fig. 4(b), which consider the presence of the 1” diameter, 3 mm thick fluorescence filter placed in front of the PMT aperture, the overall coupling efficiency from fiber bundle to photocathode can be as low as 8.7%. To address this challenge, we designed a set of collection optics to maximize the overall coupling efficiency from fiber bundle to photocathode to 65%, which is shown in Fig. 4(c).

 

Fig. 4. Zemax simulation results showing the coupling efficiency (CE) between the collection fiber bundle and the PMT photocathode. a) Schematic showing a packed bundle of 12 collection fibers, color coded based on their distance from the optical axis, with blue fiber being on-axis and red fibers being furthest away. b) Overall coupling efficiency is only 8.7% when no additional set of optics is used between the fiber bundle and the PMT, except for the fluorescence filter. c) Optimized collection optics set consisting of 5 off-the-shelf spherical lenses with diameters up to 2”, increasing the overall coupling efficiency to 65%.

Download Full Size | PDF

2.5 Optomechanical design, assembly, scanning, and read-out

The finalized optomechanical design of the probe is shown in Fig. 5(a), along with the external dimensions. The outer diameter and the length of the removable cap are 14.8 mm and 170 mm, respectively, in compliance with the clinical requirements for cancer diagnosis in the cervix. A circular sapphire window of 5.5 mm diameter and 110 µm thickness covers the distal aperture on the cap, conforming to the excitation optics design and the collection fiber arrangement diameter requirements.

 

Fig. 5. Optomechanical design of the probe. a) External dimensions of the finalized design. b) Photograph showing the fully assembled probe used in subsequent bench-top experiments.

Download Full Size | PDF

Tubing parts remaining inside the cap are machined out of stainless-steel material at the workshop of the University of Texas at Austin to maintain rigidity and alignment accuracy within the cap. The design choice to implement the tubing inside the cap in multiple segments allowed for easier machining within tight tolerances. We fabricated the cap as well as the handle parts via SLA printing for rapid prototyping and bench-top testing. Particularly for bench-top characterization, we left open two slits on the sides of the distal end of the cap to observe and characterize the axial actuation of the distal optics. For clinical use, it is necessary to fabricate the cap out of medical grade stainless steel, without these characterization slits present. Fabrication of the cap out of stainless-steel can be done via direct metal laser sintering 3D printing, with magnets added to the design for repeatable attachment and detachment of the cap during sterilization. Collection fiber bundle length was kept at 2.5 m, allowing for a realistic clinical scenario. Motivated by initial measurements with a 1 m long piece of excitation fiber verifying that pulse stretching at 775 nm was minimal, we used an excitation fiber length of 4 m in the assembled probe, increasing system layout flexibility. We used furcation tubing to protect and isolate the exposed portions of the excitation and the collection fibers. Figure 5(b) shows the fully assembled probe we use in bench-top experiments.

We mounted the assembled probe on a three-axis translation stage for bench-top characterization. A Ti:sapphire laser (MaiTai, Spectra Physics) with tunable wavelength set at 775 nm, 100 fs pulse duration, and 80 MHz repetition rate acted as the excitation source in the experiments. A programmable function generator (DG2052, Rigol Technologies) handled the fully differential driving of the piezoactuator tube, for the scanning of the fiber tip in a spiral pattern. Custom design amplifiers were used to amplify the function generator outputs to obtain the desired level deflection from the fiber cantilever. During the assembly, we determined that the length of this fiber cantilever should be around 10 mm, so that we obtain a cantilever resonance frequency around 1 kHz. This resonance frequency was targeted to satisfy a scan speed allowing image acquisition faster than 1 fps for an FOV of 120 µm, pixel size of 0.33 µm, and a data acquisition rate of 1.25 MS/s, determined by the data acquisition card we used (NI 6356, National Instruments). A low noise amplifier (Stanford Research Systems, SR570) was used for interfacing PMT module output with this data acquisition card. We set the amplifier bandwidth at 500 kHz to remain below half the sampling rate of the data acquisition card. We developed custom software in Matlab to coordinate the programming of the function generator, data acquisition, scan pattern tracking, axial actuation, and generation of images.

3. Imaging characterization of the probe

3.1 Excitation arm characteristics

We started the characterization of the excitation arm by measuring the overall transmission efficiency as 20%, with the highest loss contribution coming from laser coupling into the excitation fiber, which has a mode-field diameter of 5.5 µm. Measurements with an autocorrelator (pulseCheck, A.P.E) indicated pulse durations at laser output and excitation fiber output to be 112 fs and 97 fs, respectively. Therefore, we can say there is no significant stretching of laser pulses at excitation fiber output, which is expected since we are operating around the zero-dispersion wavelength of our excitation fiber. Next, we performed bench-top characterization with the scanning components to determine suitable scan properties of the assembled probe. We measured the deflection of the focal spot with respect to frequency and amplitude of the driving sinusoidal signals for two axes, arbitrarily labeled as Axis 1 and Axis 2, as shown in Fig. 6(a). We see that resonance frequency of either axis have a slight mismatch around a mean resonance frequency of 1.1 kHz, due to radial asymmetries in the scan system, as noted in literature [26,51]. To have high fiber deflection using low signal amplitudes, we selected the scan frequency at this mean resonance frequency in the upcoming experiments. Deflection measurements at 1.1 kHz fiber scanning frequencies shown in Fig. 6(b) indicate that the lateral scanning system can achieve an FOV of 120 µm, which is close to the maximum FOV determined by the excitation objective simulations, at voltage amplitudes of ± 30 V. For the selected scanning approach, nonlinear effects of the mechanical resonance of the fiber cantilever are known to cause deviations from the intended scan patterns. We applied methods discussed in the literature to mitigate these effects by using a position sensing detector (PSD) (PDP90A, Thorlabs) for observation of the scan pattern and mapping of data to pixels in subsequent imaging operations [23,51].

 

Fig. 6. Characterization of the excitation arm. a) Change in achieved FOV for either axis vs. actuation signal frequency (top) suggests a resonance frequency around 1.1 kHz. At this fiber scanning frequency, achieved FOV for different deflection amplitudes (bottom) attains a value close to maximum FOV determined by lens clear apertures when differential drive signal amplitudes are ± 30 V. b) Image of a set of fluorescent beads of 100 nm diameter in and out of focus, embedded within 2% agar, within FOVs of 120 µm (top) and 40 µm (bottom). Averages of 5 images are shown for this specific axial position, with pixel sizes being 0.33 µm (top) and 0.12 µm (bottom). Bottom image shows the area marked by the yellow frame in the top image. c) Lateral (top) and axial (bottom) intensity profiles of a bead in focus, marked with an orange frame in b), along with the gaussian fits to the intensity profile (R² > 0.90).

Download Full Size | PDF

For the point spread function (PSF) characterization, we imaged fluorescent beads of 100 nm diameter (F8803, ThermoFisher) embedded in a 2% agar hydrogel of 300 µm thickness, sandwiched between two cover slips and spacers for accurate thickness control. Following the solidification of hydrogel, we removed the top cover slip and imaged beads by placing the hydrogel surface directly in contact with the cap aperture. This setup follows the simulation model for excitation objective design and represents a clinical tissue imaging case accurately. Bead images from sample were collected with two scan patterns, one full-scale 120 µm FOV with 0.33 µm pixel size, and one zoomed 40 µm FOV with 0.12 µm pixel size, allowing oversampling of the PSF. Images at different axial positions were collected at 0.4 µm intervals for both FOVs. We evaluated averages of 5 frames per axial position collected at 2 fps rate to characterize the PSF. Figure 6(b) presents an image showing multiple beads within the full-scale and zoomed FOVs. Lateral and axial intensity profiles of a sample bead present within the zoomed FOV are shown in Fig. 6(c). Analyzing 15 beads located within the full-scale FOV, we measured the average lateral 2p imaging resolution as 0.66 ± 0.06 µm, which corresponds to an effective NA of 0.44. The average axial resolution was 4.65 ± 0.47 µm. Uncertainties indicate standard deviations. Increase in average lateral PSF size with consideration of beads from whole FOV confirms suggests minor alignment imperfections.

Before we moved on to characterize collection efficiency of our probe, we tested the repeatability of fiber scanning. To that end, we designed a spiral pattern with 120 µm diameter and recorded the PSD output over several hours, with pauses made in between. Pearson correlation coefficient of PSD outputs for average of initial 10 scans immediately collected at the beginning of the experiment and scans performed later were greater than 0.99 in all cases. Dividing the FOV as by pixels of size 0.33 µm, half the determined lateral resolution, and taking the average of the same 10 initial scans as a reference, we also made a comparison between data point to pixel coordinate assignments for each scan. Results show us that at least 87% of data points across FOV were assigned to the correct pixel coordinate within several hours of operation.

3.2 Collection arm characteristics

Next, we measured the collection efficiency of the probe system. To this end, we used our probe for 2p excitation of fluorescein solutions of different concentrations and recorded voltage levels acquired by the read-out block, which consists of PMT, preamplifier, the data acquisition card. For comparison, we imaged the same set of fluorescein solutions using a 0.75 NA air objective in our table-top microscope with the same read-out block interfacing it. This custom-made microscope was previously optimized for fluorescence collection [52]. We prepared the fluorescein solutions in a pH 11 buffer specifically, so that we can use the 2p action cross section data presented in [53] in our calculations. We selected the concentration levels at logarithmic intervals to cover a wide concentration range, which required us to adjust the laser power to remain within the dynamic range of the PMT module. The voltages, scaled with the square of the excitation power, demonstrate a high linearity against concentration levels both in the probe and microscope measurements as shown in Fig. 7(a), validating measurement accuracy over the wide concentration range. Slopes of the linear fits suggest that overall probe system collection efficiency is ∼ 55% of the table-top microscope collection efficiency. We note that this comparison takes into account transmission losses the fiber bundle and the imperfect coupling efficiency of the collection optics, which we measured to be 56% instead of theoretically expected 65%.

 

Fig. 7. Collection efficiency characterization of the probe. a) Scaled voltage output with respect to changing fluorescein concentrations for both our probe and the custom-built table-top microscope utilizing a 0.75 NA objective with an optimized signal collection arm. The slopes of the linear fits indicate that our probe achieves 55% of system collection efficiency that the microscope offers. b) Schematic showing the factors included in calculation of the system and probe collection efficiencies, following the work in [53]. c) Axial dependence of collection efficiency, measured by imaging of fluorescent beads embedded in 2% agar hydrogel, along with polystyrene beads introducing scattering (l_s = 32 µm) in the medium. Fit line to data indicates a drop rate consistent with simulations. d) Radial variation of collection efficiency as measured by imaging fluorescein intensity profile. The blue line presents the radial reduction in excitation power as measured at different radial positions as a result of the objective vignetting. Intensity reduction follows squared power profile closely, suggesting low radial dependence of collection efficiency, as simulations predicted.

Download Full Size | PDF

We start the calculation of our system collection efficiency, which is defined as the collection efficiency of the entire 2p-photon probe system including the probe collection efficiency and the coupling efficiency of the collection optics. For this calculation, we note that time averaged voltage acquired by the read-out block < V(t)> is determined by time averaged fluorescence photon generation < F(t)>, system collection efficiency η_s, and the read-out block sensitivity S according to:

It should be noted that, unlike in [53,54], our definition of η_s does not include PMT photocathode quantum efficiency, which is instead incorporated into S, along with PMT gain and preamplifier gain. Based on the analysis presented in [53,54] we can express < F(t)> as:

We finally analyzed axial and radial dependence of collection efficiency in two separate experiments. To measure axial dependence, we prepared a scattering tissue phantom by dissolving 100 nm diameter fluorescent beads and 1 µm diameter polystyrene beads at 3.5 × 10⁸ /ml and 3.4 × 10¹⁰ /ml concentrations within 2% agar hydrogel. Power transmission measurements after hydrogel settled revealed the phantom l_s as 32 µm. We imaged the fluorescent beads in phantom at different axial positions with our probe. We recorded the decay of average fluorescence from pixels at bead location present in a set central region of FOV. The fluorescence levels are then normalized by the average power at the focal plane considering the excitation power decay in accordance with Beer’s Law and the measured phantom scattering length of l_s = 32 µm. Figure 7(c) presents the resulting normalized fluorescence decay across several imaging depths. Fit line to data indicates a collection efficiency decrease with a slope of -0.060l_s, slightly below -0.066l_s, as predicted by the simulations for 25 µm l_s.

To measure the radial dependence of collection efficiency, we analyzed intensity changes across an image of a fluorescein solution, which exhibits a quadratic drop as shown in Fig. 7(d). To understand whether vignetting of excitation objective was responsible for this quadratic drop, we measured the output power at different radial positions as we scanned the fiber along a series of circle patterns with growing radii. The square of the output power represents the available two-photon excitation power profile. Figure 7(d) shows that normalized fluorescence intensity profile follows the square of normalized excitation power very closely, with only 3% difference at FOV edge. This result confirms our expectation that collection efficiency has minimal dependence on radial position and thus is not affected by lateral scanning.

3.3 Imaging with the probe system

Following the characterization experiments, we used our probe to image biological species. All images presented henceforth have 120 µm FOVs, in 327 × 351 pixels, and 0.33 µm pixel sizes. Frame rate being 2 fps, we show averages of 10 frames as images. Figure 8 shows z-stack images of mixed pollen gains (#304262, Carolina Biological Supply) deposited on a cover slip. This set of images allows us to test our probe with targets that are commonly used in microscopy [55–57], confirms our probe’s ability to resolve fine features, and stability of the axial scan operation.

 

Fig. 8. Two-photon images of a pollen grain cluster obtained with the 2p-probe using an average laser power of 15 mW on sample surfaces and via axial scanning of the probe against its cap. Axial depths are indicated on the upper left corner of each image.

Download Full Size | PDF

Next, we performed autofluorescence imaging of excised porcine vocal fold tissues. We chose to use this type of tissue since we had previously studied its optical properties [47] and since they offer a fair approximation to the cervical tissue as mucosal epithelium. We imaged the samples approximately 1 hour after harvesting, during which they were being transported on ice. We used the same scanning and image construction parameters as before, while adjusting the average power at sample surface at regular intervals as we increased the imaging depth up to 250 µm. Fluorescent beads were applied on tissue surface to serve as markers. Obtained images from one of the samples are shown in Fig. 9. We see the epithelial layer marked by squamous cells with bright cytoplasm and dark nucleus features giving way to fibrous collagen layers in superficial lamina propria below around 50 µm imaging depth, which is consistent with the prior studies in our group [47]. We did not observe photodamage on tissue within the applied power range of 25–100 mW, measured at sample surface.

 

Fig. 9. Autofluorescence images obtained with the probe at different imaging depths into an excised porcine vocal fold tissue sample. Imaging depths and average power at the sample surface are shown on respective images.

Download Full Size | PDF

Following the image acquisition, we performed fluorescence decay and maximum imaging depth analyses with imaged samples, following the approach presented in [58]. Normalizing pixel intensities with respect to the laser power on sample surface, in accordance with the quadratic relationship between power and fluorescence, we recorded the average of at least 50 brightest and darkest pixel intensities as MS (measured signal) and MB (measured background), respectively, at each frame. Decay trends of fluorophore signals MS-MB and MB versus imaging depth are shown in Fig. 10. We see that fluorophore and background signals follow the trends observed in our previous work [58], with a slower decay of background eventually catching up with exponential decay of fluorophore fluorescence, where image contrast drops to unity and maximum imaging depth is reached. We estimate from the decay trends the maximum imaging depth for the presented experiment as 300 µm. From the decay constant of fluorophore signal, 0.021⁻¹, it is possible to estimate a generalized scattering length value at the excitation wavelength, l_s, for the imaged tissue. There are two factors to consider for this calculation. First, the decay constant of fluorescence intensity is the decay constant of the square of incident power, which means that the decay constant for the incident power is twice the value obtained from the plot. Second, the physics of 2p imaging is such that the measured decay constant of fluorophore signal is contributed by both signal decay of excitation power and change in collection efficiency across the depth [59]. However, in our system, collection efficiency decay has a linear trend as evidenced by both simulations and characterization results shown in Fig. 3(b) and Fig. 7(c), respectively. Since the linear decay of collection efficiency is dominated by exponential decay of excitation signal in this for our probe, we can estimate l_s directly from the decay constant as 95 µm. This value is consistent with previously measured l_s of porcine vocal fold tissues obtained from the same vendor and presented in similar conditions [47]. The determined value of l_s suggests that the maximum imaging depth of 300 µm we estimated corresponds to 3.2l_s. Following the procedures in [58], we also can estimate the fluorophore inhomogeneity in tissue (χ) as 10. For the estimated values of l_s and χ, the maximum imaging depth achieved in this experiment with our probe is consistent with models and experiments defining the maximum imaging depth limits of 2p imaging [58].

 

Fig. 10. Fluorescence decay and maximum imaging depth analysis for the porcine vocal fold tissues we imaged involves working with decay trends of MB and MS-MB. Pixel intensities of each frame were scaled for its respective excitation power, squared. Maximum imaging depth is defined as the imaging depth where MB = MS-MB, which happens around 300 µm according to decay trends. Exponential decay of MS-MB suggests a tissue scattering length of 95 µm, setting the maximum imaging depth as 3.2l_s.

Download Full Size | PDF

4. Discussion and concluding remarks

A significant leap forward in early cancer detection will be possible with the clinical translation of probe systems designed for metabolic characterization. For accurate and reliable characterization of tissues with 2p imaging, it is crucial for the developed probe systems to attain high and reliable signal collection efficiencies to surmount the biggest challenge of this modality. In this work, we demonstrated a fluorescence collection method that increases both collection area and directivity by employing a multitude of 0.5 NA collection fibers arranged around excitation optics with angular termination. Using this method, we designed and assembled a probe system that achieves a system collection efficiency comparable to that of an optimized table-top microscope. We presented measured characteristics of the probe, compared them with relevant conceptual simulations, and performed autofluorescence imaging with freshly excised tissues.

Main advantages of the proposed collection method are high collection efficiencies that are similar to table-top systems and that do not decay significantly with imaging depth nor across the FOV. Zemax simulation results suggest that the level of collection efficiency does not change significantly in high scattering conditions a drastic difference compared to DCF collection, where the collection efficiency decays rapidly. Both these properties would be important in a clinical tissue characterization scenario, since the probe can extract more consistent depth-dependent information from multilayered, heterogeneous tissues. Similarly, minimal dependence on radial position change as demonstrated by simulations and measurements alike ensures consistent collection across the whole FOV. Maintenance of a collection efficiency comparable to that of a bench-top microscope at high scattering and imaging depth conditions is tied to the method’s ability to collect more of the scattering photons owing to increased collection area with engineered directivity. In contrast, DCF collection relies more on ballistic photons and is therefore affected more by multiple scattering events.

Owing to a reliable system collection efficiency, comparable to that of a table-top microscope, and the incorporated axial scanning mechanism, we were able to perform autofluorescence imaging of porcine vocal fold tissues up to 250 µm, without inducing damage to tissues. Further analysis of image data confirms that the probe system is capable of attaining maximum physically attainable 2p imaging depth.

Additional advantages of the proposed collection approach rely on rendering of collection and excitation paths physically to be independent from each other. This independence greatly alleviates the design requirements on the excitation objective as well as the optics that couple the laser into the excitation fiber. In our experience, this scheme allows more flexibility in probe optomechanical design and testing also, since either arm can be put together and tested independently with more lenient assembly requirements. Owing to increased design freedom, we were able to optimize an excitation objective, consisting of only a pair of commercially available lenses, to achieve a 2p imaging resolution equivalent to 0.45 NA.

Two design limitations that this method introduces to the system are objective working distance and probe size. We note that the proposed collection approach works better if the focal point is at a certain minimal distance away from the probe distal end, for a given collection fiber angle. The longer this minimal distance must be, the farther the collection fibers are located away from the probe axis. We can decrease this lower limit of working distance by increasing the fiber angle further to increase directivity, but this will also necessitate an increase in probe diameter. The necessity to incorporate a collection optics block with imperfect coupling efficiency in this work was specific to the particular PMT modules available to us. Different collection optics designs will be the case for different PMT modules, collection fiber numbers, and diameters.

The future direction we are taking is to extend the imaging capability of our probe to two colors with added compatibility to 860 nm of excitation and demonstrate tissue metabolic characterization in a clinical setting. Metabolic imaging of NAD(P)H and FAD in two colors allows researchers to characterize optical redox ratio, which is gaining prominence as a hallmark of cancer emergence. Our current excitation objective, while practical, can accommodate only the single wavelength of 775 nm. To that end, we will need to employ more specialized optics compatible for the 775 nm - 860 nm wavelength range, while taking additional steps to mitigate dispersion in fiber for the 860 nm wavelength. Clinical application requires an alteration in the material choice for the cap, which can be fabricated in medical grade stainless steel via DMLS printing. We believe that the demonstrated high and reliable efficiency of the collection system we present here will ensure acquisition of images reliably, providing accurate data needed for metabolic and morphological characterization of tissues and contribute to the shift in the paradigm of clinical cancer diagnosis.