Motivation and Background
The potential utility of a miniature optical-sectioning microscope for biological investigation, preclinical animal studies, and clinical use has been well appreciated and described in the literature. 1, 2, 3, 4, 5, 6, 7, 8, 9 Our group is developing a miniature dual-axis confocal microscope (DACM) that can be deployed through the instrument channel of an endoscope to detect precancerous lesions in the human gastrointestinal tract. In particular, a near-IR fluorescence device is being constructed to enable speckle-free imaging of molecularly targeted reagents deep within the mucosal layer of the esophagus, stomach, and colon. In the development of this instrument, we seek to obtain a spatial resolution of a few micrometers in all three directions, and deep tissue penetration, which is desirable for the precise delineation of tumor margins through full-thickness epithelium. In this paper, we address a void in the literature by experimentally characterizing the ability of a DACM to image deeply within tissue. A tabletop DACM is used to investigate the deleterious effects of tissue scattering by imaging reflective targets in a reproducible tissue phantom. We also provide examples of deep 3-D fluorescence image sections of ex vivo gastrointestinal mucosa.
The inspiration for our DACM design was provided by a method to achieve improved resolution through off-axis illumination and collection of light with high-numerical-aperture (NA) objectives.10, 11, 12 Later, low-NA optics were proposed as a method to achieve a long working distance and a large field of view.13 More recently, we have shown that a dual-axis architecture may be fiber coupled and combined with postobjective scanning to provide scalability of the design to millimeter dimensions.9, 14 Because the vertical response of the DACM is determined by the overlap of the input and output beams, this configuration exhibits an improved axial (vertical) sectioning response that is comparable to the transverse response.15 Furthermore, in the dual-axis configuration, it is more difficult for light randomly scattered from the path of the input beam to be collected from the output beam, especially if the NA is small. This was investigated previously through Monte Carlo scattering simulations, indicating superior rejection of scattered light using the DACM as compared to a conventional single-axis confocal microscope.16
Until now, there have been no quantitative experimental studies of the optical-sectioning performance of the dual-axis confocal architecture in turbid media. While our final goal is to develop a 3-D fluorescence device for speckle-free molecularly targeted disease detection, a mirror-reflectance model is chosen as a well-established and repeatable method for quantifying the sectioning performance of the dual-axis confocal architecture in scattering media. Reflectance and fluorescence imaging are physically different in a number of ways. However, one can assume that the fundamental spatial-filtering mechanism of the DACM, for rejecting out-of-focus scattered light, is similar in both cases. For fluorescence microscopy, an analogous mirror model is difficult to realize in practice, and computationally challenging to simulate. Furthermore, such results would not enable relevant comparison with the many published studies of optical-sectioning technologies that explore reflectance mirror models.17, 18, 19
In this paper, we experimentally investigate the ability of dual-axis confocal microscopy to reject background scattering by measuring the axial response to a plane mirror embedded in a scattering phantom. Diffraction theory calculations (no scattering) and Monte Carlo scattering simulations (ignoring diffraction) are used to validate these experimental results. Transverse line scans of a chrome-on-glass knife-edge in turbid media are also presented to study how image contrast deteriorates due to scattering. Finally, as a practical demonstration of fluorescence imaging in tissue with a DACM, we provide -deep fluorescence images of thick gastrointestinal mucosal specimens. The images corroborate the observations from our reflectance studies in turbid media that FWHM spatial resolution is preserved deep into tissue and that image contrast deteriorates quickly at the point where background scattering overwhelms the ballistic signal from unscattered photons.
The DACM utilizes low-NA (0.21) optics and an inexpensive low-power diode laser source for imaging deep within biological tissues (Fig. 1 ). Near-IR excitation of fluorescence at is used to minimize tissue absorption and scattering for deep penetration, as well as to minimize the autofluorescence background. Aberration-free postobjective scanning of the illumination and collection beams is possible due to the long working distance afforded by the low-NA lenses, as well as the use of a hemispherical index-matching sample holder (Sec. 2.1), enabling a large field of view to be scanned. Unlike single-axis confocal microscopes, in which the diffraction-limited axial resolution is generally much inferior to transverse resolution for a given objective-lens NA, the DACM produces a diffraction-limited focal volume that is relatively balanced in all three spatial dimensions.
Dual-Axis Optical Setup
A schematic of the DACM is shown in Fig. 1. Corning HI 780 fiber is utilized for single-mode (SM) transmission of laser and fluorescence radiation. The fiber-coupled laser is a SM diode source from Micro Laser Systems, Inc. (SRT-F785S-36) with a maximum output power of . The laser radiation passes through a fiber-coupled optical isolator from Optics for Research to suppress back-reflections. The maximum laser power transmitted through the isolator is . For imaging, the laser power is reduced to a maximum incident power of on tissue samples.
Two low-NA Gaussian beams are configured to intersect at their focus with a crossing half-angle of (Fig. 1). The beams are focused into the sample medium with a half-angle of , where the angle is defined with respect to the lens aperture and in the sample medium. Lenses and have a focal length of and a clear aperture of . Lenses and have a focal length of and a clear aperture of . The intensity NA of our SM fiber (Corning HI 780) at is 0.11 (mode field ).
A solid immersion lens20 consisting of a fused silica hemisphere is used for index matching of the illumination and collection beams into tissue . The wave front curvature of the focused beams is matched to the curved surface of the hemisphere as it crosses the air-glass interface, thus minimizing the spherical aberration, coma, and astigmatism that would otherwise be introduced by a flat optical interface at the tissue surface15 (i.e., flat glass window). The hemisphere also preserves the focusing half-angle and the incident angle of the beams, thereby increasing the effective NA in tissue by a factor of (from 0.21 to 0.29). The flat surface of the hemisphere acts as a -diam circular sample holder that is translated in for vertical sectioning and translated in for volume scans. An analysis of wave front aberrations due to dual-axis beam scans in and , through a similar index-matching optic, was performed previously,21 indicating a maximum difference in wavefront aberration of only over a horizontal field of view.
For the reflectance experiments in turbid media, the collection fiber is connected directly into a powermeter (EXFO PM-1100). The filter, photomultiplier tube (PMT), amplifier, and frame grabber are only used for raster-scanned fluorescence imaging of tissues.
Measuring the Axial and Transverse Response in Turbid Media
Intralipid scattering phantom
Among the most widely used and well-studied tissue phantoms for optical imaging are lipid emulsions similar to milk. In these studies, 20% Intralipid is used (Baxter Healthcare Corp). Approximate values for the optical properties of Intralipid can be found in the literature.22 For our studies, the scattering coefficient is measured directly during each experiment. The absorption coefficient is negligible, and the anisotropy (defined in Sec. 2.2.3), is assumed to be 0.75 in our simulations, based on results reported in the literature. Vigorous stirring of the Intralipid solution prior to use results in consistent measurements of .
Axial mirror and transverse knife edge scans
A piezoelectric stage (Physik Instrumente P-783.ZL) translates the dual-axis microscope’s hemispherical index-matching sample holder in the (axial) direction. A plane mirror is held at calibrated distances above the sample holder and is translated with a high-resolution piezoelectric scanner (Physik Instrumente P-762.ZL).
For the axial mirror scans, an Intralipid scattering phantom, or water (for zero-scatter measurements), is injected on top of the sample holder, as shown in Fig. 2 . The sample holder is lowered such that the dual-axis focus is located at a precalibrated distance above the flat surface of the sample holder. Then, the silvered mirror is lowered into the Intralipid media until a peak signal is found, indicating the position at which the mirror is at the focal plane of the dual-axis microscope. Axial scans are performed by scanning the silvered mirror in the vertical direction. Control signals are generated in LabVIEW using a National Instruments waveform generation board (PCI-MIO-16E-1). The axial scans are performed with a linear ramp waveform at . Laser intensity signals are recorded at onto a memory buffer in a powermeter (EXFO PM-1100). The memory buffer is then transferred via a general purpose interface bus (GPIB, IEEE 488.2) onto a PC through a National Instruments GPIB board (PCI-GPIB) programmed with LabVIEW.
Transverse knife-edge scans are accomplished by using a piezoelectric stage (Physik Instrumente P-783.ZL) to horizontally translate a chrome knife-edge target imprinted on a glass substrate (Applied Image Group, Rochester, New York). The surface of the knife-edge target is positioned axially in the focal plane of the dual-axis microscope to achieve optimal signal strength and contrast in the line scans. Intralipid, 20%, is injected in the gap (precalibrated distance) between the hemispherical sample holder and the knife-edge target for measurements through scattering media.
Diffraction theory calculations, described previously,15 can be used to calculate the theoretical FWHM resolution of the DACM, in response to both point and plane reflectors. Numerical integration routines are performed in MATLAB. In the dual-axis system studied here, the ratio of the aperture diameter to the Gaussian beam diameter is 1.3. Experimental parameters are , , , and (approximate index of tissue). Calculated and measured resolutions are listed in Table 1 .
|TheoreticalResolution,FWHM (PointReflector)||Theoretical AxialResolution,FWHM (PlaneMirror Reflector)||Theoretical FocalVolume ∼ (4∕3)π[(Δx∕2)] (Δy∕2)[(Δz∕2)]||MeasuredResolution,FWHM|
We performed Monte Carlo simulations on an optical model that approximates our experimental setup. As described previously,16 a nonsequential ray optics program (ASAP® 2006, Breault Research Org.) is used to perform scattering simulations. The angular distribution of scattered light, for nonabsorbing spherical scatterers, is given by the Henyey-Greenstein phase scattering function, which is an approximation to Mie scattering theory:23, the anisotropy factor, is defined as
In these simulations, a detection “pinhole” diameter of is chosen, which approximates the “top-hat equivalent” diameter24 that would yield the same throughput as a Gaussian-weighted pinhole with a diameter of (the mode field diameter of our SM fiber). The illumination source in the simulations is a collimated beam with a Gaussian intensity distribution that matches the beam in our experimental DACM. However, diffraction effects are not modeled in the Monte Carlo simulations.
As described in Sec. 3.1, the scattering coefficient of our Intralipid scattering phantom is measured experimentally and is the value used in our Monte Carlo simulations. The anisotropy is estimated based on published studies.22 Our measured values for are also consistent with published values in the literature.
Dual-Axis Imaging of Gastrointestinal Mucosa
Vertical sectioning is the primary mode of imaging for our device. Postobjective horizontal (fast- -axis) scanning is accomplished sinusoidally at with a galvanometric scan mirror from GSI Lumonics (VM500), and vertical (slow- -axis) depth scanning is performed at with a piezotranslation stage from Physik Instrumente (P-290). The vertical waveform is a sawtooth modified with smooth turn-arounds to avoid mechanical ringing. Waveforms are produced with National Instruments waveform generators programmed in LabVIEW®. Signals are displayed and stored with an frame grabber from Data Translation (DT3152). The frame grabber is synchronized with horizontal and vertical trigger pulses supplied by the same National Instruments boards used to generate the scanning waveforms. The imaged field of view is 800 (horizontal) by (vertical), acquired at a pixel rate of for images that are 1000 (horizontal) by (vertical) in size. For volumetric imaging, the sample stage is translated in the third dimension with a computer-controlled motorized actuator from Newport Corporation (LTA-HL). Vertical images are acquired and stored at , while the sample is translated in at a constant velocity to obtain serial sections separated by . The 2-D and 3-D images and animations are rendered with the Amira 3.0 software package.
Depth-weighted fluorescence detection
The fluorescence microscope utilizes a Hamamatsu PMT (H7422-50) and PMT controller (C8137-02) for the detection of incoherent single-photon fluorescence. A Semrock RazorEdge® long-pass filter (LP02-785RU-25) is used to reject laser radiation prior to PMT detection. The PMT output is converted to a voltage signal with a current amplifier from FEMTO Messtechnik GmbH (DHPCA-100). The transimpedence gain of the amplifier is set to with a corresponding bandwidth of .
A large dynamic range of fluorescence signal ) is encountered when imaging vertical sections with deep tissue penetration. To prevent PMT detector saturation, as well as to optimize the signal levels for acquisition with an frame grabber, the PMT gain is modulated with a sawtooth waveform synchronized to the slow-axis waveform (vertical scan). By adjusting the offset and amplitude of this gain-control waveform, it is possible to compensate for an exponential loss of fluorescence signal with imaging depth due to absorption and scattering losses, as well as a depth-dependent dye distribution. Due to the slow gain-settling time of the PMT , it is necessary to introduce a slight delay (phase shift) between the sawtooth waveforms for imaging at .
Fresh tissues were collected via pinch biopsy during standard endoscopy at the Palo Alto Veterans Administration Hospital. Patient informed consent was obtained and approved by the Institutional Review Board of the Stanford University School of Medicine. Fresh biopsy specimens were soaked for in a near-IR dye from LI-COR Biosciences (IRDye® 800CW). The dye was dissolved, at a concentration of , in water that contained 5% DMSO to facilitate tissue penetration. Prior to imaging, excess dye was removed by soaking and irrigating the tissues with water. After imaging with our dual-axis fluorescence confocal microscope, biopsy samples were fixed in 10% formalin and submitted for routine histologic processing [hematoxylin and eosin (H&E) staining].
Axial Response in Turbid Media
As described in Sec. 2.2.2 and depicted in Fig. 2, the DACM was used to image reflected light from a plane mirror located at various depths within an Intralipid scattering phantom. Our axial mirror response experiments were all performed in a regime where the peak signal, from a mirror placed at the focus of our microscope, is dominated by ballistic (unscattered) photons. Therefore, the decay in the peak mirror signal, as a function of depth, follows Beer’s law: . Here we assume that absorption effects are negligible, as compared to scattering losses. The depth of the mirror in the scattering media is given by and the factor of 2 accounts for both the illumination and collection path lengths. For the dual-axis configuration, a factor of is introduced due to the off-axis geometry of the dual-axis beams.
Figure 3 is a normalized plot of the measured fall-off in the ballistic signal (mirror at the focus) and the signal due to background scatter (no mirror) as a function of depth. As we can see from the plot, the peak signal is well above the background scattering level, suggesting that it is dominated by ballistic photons reflected at the mirror surface. The scattering coefficient may be inferred from a curve fit to the peak-signal decay as a function of depth. The background signal due to scattering is measured by removing the mirror from the Intralipid phantom. Monte Carlo simulations, based on the measured value for , yield good agreement with experimental values: the ballistic signal falls off exactly according to Beer’s law and the scattering background decays at the same rate as the experiments. The Monte Carlo simulation of the background scattering level is approximately below the experimental values. This disparity is due to the fact that a SM fiber was used as the collection pinhole in our experimental DACM. Therefore, diffraction, slight optical misalignments and/or aberrations introduce losses in the peak (ballistic) signal that were not captured in the ideal diffraction-free Monte Carlo simulations.
Axial mirror scans, from the focal plane at to a defocus of , were performed at various depths within the scattering phantom, as displayed in Fig. 4a . Note that the near-focus axial response, primarily due to ballistic photons reflected from the mirror, remains diffraction limited until the background scattering level is approached and begins to dominate. The Monte Carlo simulations in Fig. 4b show good agreement with the measured results. Figure 4b also contains results from two diffraction-theory calculations. The first diffraction-theory calculation is of the axial response of our DACM to a plane mirror in the absence of scattering. This calculation matches the experimental curve, shown in Fig. 4a, for an axial mirror scan in water (no scattering). The second diffraction-theory calculation in Fig. 4b is of the axial response of a conventional single-axis confocal microscope with the same FWHM axial resolution as our DACM. The single-axis geometry yields a relatively slow fall-off in its diffraction-limited axial response compared with the dual-axis . More implications of these results are discussed in Sec. 4.
Transverse Response in Turbid Media
As described in Sec. 2.2.2, a transverse scan, of a chrome knife-edge on glass, was performed in scattering media. The reflectivity of chrome at is approximately 40%, and the Fresnel reflection at a water-glass interface is . This difference in reflectivity is seen in Fig. 5a for low levels of scattering . As the optical length increases, the contrast in signal between chrome and glass decreases. The high reflectivity of the chrome (40%) ensures that ballistic photons reflected from that surface dominate over the scattering background. Therefore, the signal from the chrome surface obeys a Beer’s law decay with respect to depth, even at . The weaker signal from the glass surface, however, is increasingly dominated by the scattering background as extends above 7.5.
In Fig. 5a, the transverse resolution of the knife-edge line scan remains relatively constant as a function of scattering length, whereas contrast deteriorates quickly at the most extreme scattering lengths . A measurement of the 10 to 90% intensity points of the knife-edge response confirms that this quantity is constant as a function of imaging depth [Fig. 5b]. This is consistent with the axial mirror scan results: the FWHM spatial resolution is clearly preserved regardless of imaging depth in turbid media [Fig. 4a]. However, contrast degrades quickly at the point where background signal due to scattering overwhelms the ballistic signal. Figure 6 shows the effect of the scattering background on the contrast between the chrome and glass surfaces of the knife-edge, where contrast is defined as4.
Deep Dual-Axis Fluorescence Imaging of Gastrointestinal Mucosa
Clearly, to demonstrate optical-sectioning performance in real tissues, axial and transverse mirror models, though quantitatively informative, are poor substitutes for images obtained in fresh biological tissues. Therefore, we present volumetric fluorescence images of gastrointestinal mucosa, as described in Sec. 2.3.1. While there is much variability in biological tissues, reported values for , for esophagus and colon at , are approximately 7 and , respectively.25, 26
Diagnostic imaging of esophageal mucosa
Mucosal biopsies from normal esophagus were imaged following dye application. A vertical section is shown in Fig. 7a , demonstrating the ability to visualize squamous cells to a depth of . Note that the PMT detector gain is varied during each vertical image, at , to compensate for an exponential decrease in fluorescence signal with depth. A horizontal rendering at a depth of , of 300 vertical sections spaced by , is shown in Fig. 7b.
Barrett’s esophagus is a condition in which the normal squamous epithelium of the esophagus has been replaced by an abnormal columnar epithelium termed specialized intestinal metaplasia. Since Barrett’s esophagus is often indistinguishable from normal stomach (gastric) mucosa under white-light endoscopy, biopsy and histological examination are necessary for the accurate diagnosis of this precancerous condition. Vertical and horizontal images of Barrett’s esophagus, respectively, are shown in Figs. 7c and 7d. Figure 7e shows a vertical section of a transition junction between squamous and Barrett’s esophagus. A horizontal rendering, at a depth of , is shown in Fig. 7f along with the corresponding histology (H&E staining) in Fig. 7g. The presence of mucus-secreting goblet cells, which appear as brightly stained vacuoles among the columnar cells lining the glands [Fig. 7f], confirms the specialized intestinal metaplasia of Barrett’s esophagus rather than gastric mucosa.
Diagnostic imaging of colonic mucosa
Colonic polyps are routinely biopsied and diagnosed histologically. Adenomatous polyps are a dysplastic transformation of colonic mucosa with precancerous potential. Figure 8a shows a vertical section of a colonic polyp biopsy, following dye application. A horizontal rendering of this tissue, at a depth of , is shown in Fig. 8b, along with corresponding histology (H&E staining) in Fig. 8c. This example of an adenoma, showing low-grade dysplasia, is easily contrasted with and distinguished from optical sections of normal colonic mucosa [Fig. 8d] as well as optical sections of adenoma exhibiting high-grade dysplasia [Fig. 8f]. The corresponding histology (H&E staining) is shown in respective order [Fig. 8e and Fig. 8g]. In normal mucosa, the crypts (glands) are circular and evenly spaced. In adenomatous mucosa, the crypts are ellipsoidal with high eccentricity and are lined with hyperproliferative and stratified colonocytes.
The dual-axis confocal architecture has advantages over conventional single-axis confocal microscope designs, particularly for miniaturization and in vivo use. The DACM utilizes low-NA objectives, which provide long working distances and enable aberration-free postobjective scanning over a large field of view. Deep tissue microscopy requires efficient rejection of background scattering, which has been accomplished in a variety of ways and described in the literature. Among popular methods are time- and coherence-gating methods, such as optical coherence tomography, as well as nonlinear microscopy techniques, accomplished with high-power lasers and high-NA focusing. In this paper, we demonstrate that simple low-NA optics and low-power diode lasers can be utilized in a dual-axis confocal architecture to image deeply within turbid media. The DACM does not require coherent detection, thereby making it compatible with molecularly targeted fluorescence imaging that is free from speckle noise. For example, we plan to utilize a miniature endoscopic DACM to image fluorescent peptides, discovered by other researchers within our group, that preferentially bind to cell-surface molecular biomarkers of disease in the colon.27
To quantify the optical-sectioning performance of a tabletop DACM, simple reflectance experiments were performed that may be verified with Monte Carlo scattering simulations. We previously performed diffraction-theory calculations to show that the axial response of the DACM falls off quicker, and has a larger dynamic range, than the response of a single-axis confocal microscope with an equivalent FWHM axial resolution.15 In this study, we show that in the vicinity of the focal plane , the axial response of our DACM to a plane mirror, through up to mean free paths of scattering media (round-trip optical length), is superior to the diffraction-limited (scatter-free) performance of a single-axis confocal with an equivalent FWHM axial resolution. We also demonstrate excellent contrast in transverse line scans of a knife-edge target, through round-trip optical lengths of up to in turbid media. Our results, presented in Secs. 3.1, 3.2, indicate that the dual-axis confocal architecture exhibits excellent optical sectioning that is balanced in performance along both the axial and transverse dimensions.
As reported in the literature, transverse knife-edge experiments similar to ours have been carried out to study single-axis confocal microscopes. The most impressive results were reported by Kempe 19 In that study, chrome-on-glass reflectance gratings were imaged, yielding results that are comparable to our own results shown in Fig. 6. However, those results were obtained from a confocal microscope that had an effective NA of only 0.28, which is similar to the effective NA of our DACM (Sec. 2.1). Such a confocal, in a single-axis arrangement, would exhibit an FWHM axial resolution of the order of , not to mention an extremely slow axial response, as discussed in Sec. 3.1. It has been determined that there is a reduction in the amount of background scattered light collected by a single-axis confocal as NA is reduced, at the cost of a degradation in resolution and axial-sectioning response.28 Therefore, while a low-NA single-axis confocal microscope may perform well for imaging a flat “presectioned” reflective target in scattering media, the slow axial response and poor axial resolution would not enable effective optical sectioning in 3-D biological tissues. Note that our study is by no means intended as a comprehensive comparison between single-axis and dual-axis confocal microscopies. These two confocal architectures represent distinct modalities that possess strengths and weaknesses for different applications. Our primary aim was to perform experimental studies to characterize the DACM in turbid media, as has been done with single-axis confocal microscopes in the past.
Of the numerous studies of optical-sectioning technologies,17, 18, 19, 28 few studies investigate both axial and transverse performance in scattering media. Of those studies, still fewer, if any, provide a demonstration of imaging performance in actual biological tissues as well. Since mirror and knife-edge models may be misleading, we demonstrate deep 3-D fluorescence microscopy in biological tissues. Collectively, these results consistently demonstrate the deep optical-sectioning abilities of the DACM. In particular, for the reflectance studies described in Secs. 3.1, 3.2, we observed that FWHM spatial resolution is not significantly affected by imaging depth (optical length), but that image contrast quickly degrades as the scattering background overwhelms the ballistic signal. This is also apparent in many of the images of biological tissues. For example, in Figs. 7a and 7e, the membranes of the individual squamous esophagus cells remain well defined and resolvable deep into the sample, but quickly lose visibility at a certain point as the scattering background overwhelms the ballistic signal from the cell walls.
In summary, the dual-axis confocal architecture exhibits efficient rejection of scattered light, as demonstrated through quantitative mirror reflectance studies in a scattering phantom, and through imaging exogenous single-photon fluorescence contrast in human gastrointestinal biopsy samples. The results indicate the utility of this technology to distinguish cellular and morphological signatures of various pathologies such as Barrett’s esophagus and colonic adenomas, as well as to distinguish between grades of colonic dysplasia. The 3-D optical-sectioning ability of the DACM enables arbitrary orientations and locations to be viewed. Thus, a miniature endoscope-compatible version of this device would provide a valuable tool for rapidly diagnosing and staging diseases, as well as for guiding surgical resection.
This work was funded in part by grants from the National Institutes of Health, including K08 DK067618 (NIDDK), U54 CA105296 (NCI), and R33 CA109988 (NCI). This work was also supported by funding through the Center for Biophotonics, a National Science Foundation Center managed by the University of California, Davis (PHY 0120999). Jonathan Liu is supported by a Canary Foundation / American Cancer Society postdoctoral fellowship for early cancer detection. We thank Shai Friedland, Roy Soetikno, Peyman Sahbaie, and Larry Wong for technical support.