We have built a fibre optic confocal reflectance microscope capable of imaging biological tissue in near real time. The measured lateral resolution is 3 µm and axial resolution is 6 µm. Images of epithelial cells, excised tissue biopsies, and the human lip in vivo have been obtained at 15 frames s−1. Both cell morphology and tissue architecture can be appreciated from images obtained with this microscope. This device has the potential to enable reflected light confocal imaging of internal organs for in situ detection of pathology.
Optical technologies are being increasingly used to perform real time assessment of tissue pathology in vivo. One particularly promising new technology is confocal microscopy, which samples small volumes of tissue, producing images with micrometre resolution at depths up to several hundred micrometres within tissue. A number of groups have used non-fibre optic confocal microscopes to obtain reflected light images of accessible tissues. In skin, confocal imaging can provide detailed morphological images of epithelial cell morphology and tissue architecture throughout the entire epithelial thickness, as well as detailed imaging of subepithelial blood flow in capillaries (Betrand & Corcuff, 1994; Rajadhyaksha et al., 1995; Corcuff et al., 1996; Masters et al., 1997; Rajadhyaksha et al., 1999b). Recently, confocal imaging has been used to image neoplastic skin lesions in vivo (Busam et al., 2001; Langley et al., 2001). However, the application of confocal imaging to in vivo detection of pathology in epithelial tissues other than the skin has been limited by the difficulty in access to these organ sites.
A number of groups have attempted to develop flexible endoscopes to record confocal images in vivo based on fibre optics. Gmitro & Aziz (1993) first proposed the use of a fibre optic bundle to transfer the scanned image plane and raster scanning the proximal end of the bundle to produce en face images of the sample at the focal plane. Each fibre within the bundle serves as the point source as well as the detection pinhole for confocal imaging. The Gmitro group has recently developed a slit-scan confocal fluorescence microendoscope with a miniature objective and a hydraulic focusing mechanism (Sabharwal et al., 1999). Fluorescence images of biological samples have been obtained from cultured human prostate cells and tissue sections, as well as mouse peritoneum in vivo. All samples were stained with vital fluorescent dyes to provide image contrast. In vivo fluorescence imaging has two important limitations. First, fluorescent stains must typically be used to yield sufficient signal for imaging. These dyes must be non-toxic to the tissue and able to penetrate to deeper layers within the tissue. Second, the excitation wavelength is usually in the UV or blue range, which has a short penetration depth within tissue because of scattering.
Fibre optic confocal microscopes imaging reflected light from tissue could be used with longer wavelength sources to obtain deeper penetration for in vivo imaging without the need for fluorescent stains. Fluctuations in refractive indices of tissue provide contrast that allows cellular and subcellular structures to be detected (Dunn et al., 1996). Refractive indices have been reported to be n= 1.4–1.45 for the nuclei and n = 1.35–1.37 for the cytoplasm (Schmitt & Kumar, 1996). Therefore, for biological imaging the system needs to be able to detect reflected light from Δn < 0.1 and resolve the size of cell nuclei. One fibre-based confocal system utilized a single- mode optical fibre to deliver illumination light to the sample to be imaged and to collect the backscattered light from the illuminated focal volume (Dickensheets & Kino, 1998). En face images were obtained by microfabricated scanning mirrors and focusing optics at the distal fibre end. Juškaitis et al. (1997) developed a reflection confocal microscope using a fibre optic bundle and a white light source. A tandem scanning microscope was used to obtain real time images. A primary limitation for fibre-based confocal reflectance microscopes is that the specular reflections generated at the end faces of the fibre are many orders of magnitude stronger than the tissue backscattering (Yang et al., 1999). Angle polishing the fibre faces and index matching has been used to reduce the unwanted reflections. However, no images of biological samples have been reported because of the low light efficiency and sensitivity of these devices.
Recently we have developed a fibre optic confocal reflectance microscope (FCRM) capable of imaging biological tissues with subcellular resolution. In this paper we present the design, performance and biological image results of the first prototype of this imaging system, which uses a fibre bundle coupled to a conventional microscope objective. The development of the focusing optics for this system has been described in detail (Liang et al., 2001). The basic concept of the FCRM is similar to that described by Gmitro. However, our system detects reflected light produced without the need for exogenous stains, and operates at longer wavelength for deeper penetration. Two galvanometer mirrors are used to generate a raster scan of the illumination spot at the proximal end of the fibre bundle. The problem of low efficiency in Juškaitis’ approach has been solved by using a monochromatic laser source and focusing the illumination beam onto individual fibres for higher illumination intensity. The scanning mirrors also have much higher efficiency than the tandem scanning method. The specular reflections from fibre end faces are reduced by index matching liquid at both fibre ends. In vitro and in vivo images of biological specimens including epithelial cells, biopsies from human uterine cervix, and human lip have been obtained with this prototype and are presented in this paper. The goal of our research is to develop a flexible, fibre optic confocal microscope to image pre-cancer in internal organs such as the cervix and oral cavity. The curable precursor of cancers, epithelial pre-cancer, is characterized by increased nuclear size, increased nuclear to cytoplasmic ratio, hyperchromasia and pleomorphism, which can only be assessed through invasive biopsy. Screening and detection could be vastly improved by technologies that image subcellular structure in vivo without need for expensive and painful tissue removal and examination.
2.1.Fibre optic confocal reflectance microscope
The system diagram of the FCRM is shown in Fig. 1. A coherent image guide from Sumitomo (IGN-15/30, Sumitomo, Electric USA Inc.) is located between the focusing lens L1 and the objective optics. This image guide has 30 000 fibres, an overall outer diameter of 2.5 mm, a length of 5 m, and a nominal numerical aperture (NA) of 0.3. The fibres inside this bundle have an average core diameter of 4 µm and average centre-to-centre spacing of 7 µm. The laser beam from the illumination optics is focused onto the proximal end face of the fibre bundle such that only one fibre is illuminated at one time. At the distal end of the bundle, the illumination fibre is imaged to the sample by the objective optics. The illumination fibre also serves as the detection pinhole in a conventional confocal microscope (Juškaitis & Wilson, 1992). Backscattered light from the sample goes back through the fibre bundle and emerges from the proximal fibre end. A secondary pinhole, located in the detection path, is adjusted to a position that is conjugate to the illumination/detection fibre. Backscattered light from locations other than the illumination/detection fibre will be mostly rejected by the pinhole. Therefore, the condition for confocal imaging is achieved.
The light source is a continuous wave Nd:YAG laser (Optomech Ltd, Torrance, CA, USA) that emits near-infrared light at 1064 nm. This particular laser is chosen for its short coherence length so that the effect of interference is minimized. The laser beam passes through a spatial filter, a collimation lens and a variable iris that limits the beam diameter. A beam splitter partially transmits the incident beam to the first scanning mirror and partially reflects backscattered light toward the detection arm. The scanning system consists of two scanning mirrors, two lenses located between the mirrors, and the electronics to drive and monitor the mirrors. In order to achieve a frame rate of 15 frames s−1 and an adequate number of pixels in the digitized images, the line-scan device must operate at several kHz. One scanner that meets our requirements is a resonant scanner (SC-30 with driver PLD-XYG, Electro-Optical Products Corp., New York, USA) that oscillates sinusoidally at a pre-set resonant frequency of 7.68 kHz. The second scanner is a magnetically driven mirror (6800HP, Cambridge Technology, Inc., Cambridge, MA, USA) operating at 15 Hz to provide a linear frame-scan. The two lenses between the mirrors form a telescope configuration with two functions: to expand the incident beam and to make the exit pupil of the first scanning mirror coincide with the second mirror so that the design of the following optics is simplified.
The first custom-made lens system, L1, is designed to: (1) place a diffraction-limited spot at the image plane over the active region of the fibre bundle; (2) be used with immersion oil to reduce the specular reflection from the fibre surface; (3) be telecentric so that the chief ray is parallel to the optical axis and normal incidence is obtained over the radial position on the bundle surface; (4) have a NA of 0.3 to match the NA of the image guide. Given the NA of L1 and the wavelength of the source, the diffraction-limited spot size is calculated to be 4.3 µm, which is close to the core diameter of the fibres. The optical design is aided by the software package ZEMAX and the details have been reported previously (Liang et al., 2001).
At distal end of the bundle, the other lens system, L2, relays emerging light from the fibre to the back aperture of a 40×, 1.15 NA, water-immersion objective (OApo 40× w/340, Olympus, Melville, NY, USA). The design goals of L2 are the same as those of L1 except for a larger back aperture.
2.3.Detection optics and signal to background
Backscattered light from the object follows the same path as illumination light in reverse direction through the objective lens, L2, the fibre bundle, and L1. The returned light is descanned by the two scanning mirrors and is deflected to the detection path by the beam splitter. The lens in front of the pinhole forms an image of the proximal end of fibre bundle at the plane containing the pinhole. The pinhole diameter is selected to be slightly larger than the size of the image of a fibre core. Backscattered light coming from an out-of-focus spot in the tissue will be distributed over a number of fibres within a certain distance to the illumination fibre. Only light travelling back through the illumination fibre will pass through the pinhole, ensuring optical sectioning. The detector is a high-speed avalanche photodiode (APD) with a preamplifier module from Hamamatsu (C5460, Hamamatsu, Bridgewater, NJ, USA). This APD module has an APD gain of 10–300, and a responsivity of 1 × 105 V W−1 at 1064 nm when the gain is 30.
One of the biggest challenges for the FCRM is the overwhelming specular reflection from the surfaces of the illumination fibre. This background light is intrinsic to the system because the fibre end faces are conjugate to the detection pinhole. Cleaving the fibre end faces at an oblique angle will not work well because the fibres within the bundle will be at different locations other than the desired focal plane of either lens system L1 or L2. An alternative, pursued here, is to use index-matching oil to reduce the background so that the extremely faint signal from the tissue can be detected. Complete elimination of the background cannot be achieved because the fibres consist of two materials, the core and cladding. Immersion oil with a refractive index halfway between the core and cladding indices has been used at the proximal (L1) end. The index difference between the oil and the fibres is 0.015. At the distal (L2) end the index of immersion oil matches the index of the fibre core so specular reflection from the distal fibre surface can be neglected.
The signal-to-background ratio of the FCRM is predicted to be 0.4 assuming that the signal is produced by Δn = 0.1 interface and a total attenuation of e−2 in the tissue. The signal-to-noise ratio (S/N) is predicted to be 500, assuming a maximum power of 0.15 W at the input of L1. These calculations take into account the transmission of L1, the fibres, L2, and the objective lens. The main sources of noise are shot noise of the photodiode and thermal noise of the preamplifier electronics.
The FCRM is a point-scanning device. In other words, at any instant, only one (or none) small region in the tissue is illuminated and the backscattered light from that region is detected. In order to form a 2D image, we need to scan the focus over a plane that is perpendicular to the optical axis, record the signal from each sampled region, and reconstruct the image based on the collected data. In this case the signal is a time-varying voltage from the APD preamplifier, which corresponds to intensity of backscattering from the sample. We use a frame grabber (MV-1000, MuTech, Billerica, MA, USA) to digitize the analogue signal and convert it to VGA-compatible format so that we can view the images on a computer monitor. Because the fibres in the bundle are not regularly packed, and the digitizer only works at a fixed sampling rate, it is not possible to selectively save signal that is from the object through the fibres and discard signal that is from the cladding or interfibre filling areas. Therefore, we oversample the APD output signal at a very high frequency of 10.4 MHz so that at least five points are sampled within each fibre.
Image formation is accomplished by synchronizing the operations of the frame grabber and the scanning mirrors. The controller of the scanning mirrors provides two outputs that correspond to the positions of the line-scan and frame-scan mirrors. The zero crossings of these two position outputs are detected and used to generate two TTL-compatible pulses that are sent into the frame grabber to provide the H and V synchronization pulses, respectively. Each H sync pulse triggers the frame grabber to start a new line and each V sync pulse triggers a new frame. The frame grabber can be programmed to have the proper delay time between the triggering signals and the actual start of a line or a frame so that images corresponding to the raster scan are reconstructed. Images are updated at a rate of 15 frame s−1 on the computer monitor and no drift is observed between individual frames.
The lateral resolution of the FCRM cannot be described as 0.46λ/NA as in a conventional confocal microscope because the image of the object is pixelated by the fibres. Following the definition of resolution as the smallest separation between two object points that can be visually resolved, the lateral resolution of the FCRM cannot be better than the separation between two spots that are illuminated by two adjacent fibres. The transverse magnification from the tissue to the distal fibre end is 3.8, given that the NAs of the objective lens and L2 are 1.15 and 0.3, respectively. The distance between two adjacent illumination spots in the tissue can be calculated as 7 µm/3.8 = 1.8 µm. Because the diffraction-limited spot diameter in the tissue is only 1.1 µm, the illumination spots do not overlap in the tissue. Therefore, the lateral resolution of the FCRM is limited by sampling rather than by diffraction. A lateral resolution of 3 µm is estimated and is sufficient to resolve the size of a cell nucleus.
The axial resolution of the FCRM can be calculated following the method reported by Gu et al. (1991). In their model a single-mode fibre with the parameter A = (2πa0r0/λd)2 is used for both illumination and detection. The field distribution of a single-mode fibre with is approximated by a Gaussian profile with a radius r0, at which the intensity is e−1 of the maximum; a0 is the aperture radius and d is the focal length of lens system L2. The axial resolution can be calculated from the axial optical coordinate u1/2, which is related to parameter A. The parameter A is calculated to be 6 using the specifications of the system set-up, predicting a full width at half maximum (FWHM) axial resolution of 2 µm.
The performance of the FCRM was tested by imaging standard samples including a mirror, a glass Ronchi grating and polystyrene microspheres. Then the FCRM was used to image various biological samples including rat breast cancer cells in vitro, human epithelial cells scraped from oral mucosa, ex vivo biopsies taken from human cervix, and the human lip in vivo. Biopsy specimens were excised from patients undergoing colposcopic examination at the M.D. Anderson Cancer Center in Houston, Texas. Informed consent was obtained from each patient, and the study was reviewed and approved by the Internal Review Boards at the University of Texas M.D. Anderson Cancer Center and the University of Texas at Austin. These biopsies were put in tissue growth media (DMEM without phenol red) for 6–8 h before being imaged at the University of Texas at Austin. For all biological samples both still images in bitmap format and videos in AVI format were saved. Background images were recorded along with regular images, and the corresponding background images were subtracted from the regular images in order to reduce the effects of residual specular reflection. In raw images, fibres at the left and right edges appear to be extended horizontally because the line scan pattern of the resonant scanner is sinusoidal instead of linear. The image distortion was corrected by image processing and the contrast and brightness of the resultant images were adjusted for better presentation.
Figure 2 shows two images of a glass Ronchi grating orientated both horizontally and vertically. The target has widths of 25 µm for both the bright and dark regions. The field of view can be found to be about 180 µm × 160 µm from these images. The intensity is not uniform for all fibres over the bundle, which is mainly because of variations in coupling efficiency of fibres.
The lateral resolution was measured by imaging a Ronchi grating and calculating the distance marked by 10% and 90% intensity across the edge. A line profile from a grating image is shown in Fig. 3(a). The line profile is averaged over a line width of 100 µm to smooth the intensity variations resulting from the pattern of fibres. The 10–90% distance is found to be 1.8 µm, which is the same as the calculated distance between adjacent illumination spots in the object.
The axial resolution was measured by moving a planar mirror in 2 µm steps through the focus of the system. The average intensity versus the axial position is shown in Fig. 3(b) and the FWHM is 6 µm. The predicted FWHM axial resolution is 2 µm using Gaussian approximation for single-mode fibres. The discrepancy between the measured and predicted values is partly attributed to aberrations in the optics and multimode propagation of light in the fibres. Higher order modes can also propagate in the fibres because the fibre parameter V of the fibres that we used is 3.5, given NA = 0.3 and core radius = 2 µm. The second order modes have been shown to have larger axial response than the fundamental mode (Juškaitis & Wilson, 1992).
The spatial resolution measured from planar objects (mirror and grating) is sufficient to resolve cellular and subcellular structures in epithelial tissue. In order to test the system performance on small objects, polystyrene microspheres with 4.3 µm diameter were imaged (Fig. 4). These microspheres provide a good sample for testing the system because the nuclei of human epithelial cells are usually 5–15 µm in diameter. The image shows that the microspheres are well resolved by the FCRM.
Epithelial cells in suspension were imaged by the FCRM. Figure 5(a) shows an image of cultured MTC cells (rat breast cancer cells) that were washed of growth media, suspended in phosphate buffered saline (PBS), placed on top of a thick layer of gelatin and then covered by a coverslip. The gelatin layer provides a substrate for the cell suspension and avoids any stray background from the saline/gelatin interface because the refractive index of gelatin is approximately matched to saline. Figure 5(b) shows an image of human epithelial cells taken from the oral mucosa. The cells were scraped from the inside of the cheek of one of the authors (R.R.K.) and immediately smeared onto the gelatin substrate. PBS and 6% acetic acid solution were added and a coverslip was placed. The bright features in the images are identified as cell nuclei based on the size and the contrast.
The optical sectioning ability of the FCRM was tested by imaging ex vivo biopsy specimens. Immediately before imaging, 6% acetic acid solution was added to enhance the contrast of the cell nuclei (Drezek et al., 2000). The image plane was parallel to the tissue surface and the specimen was imaged at different depths from the surface to about 150 µm below the surface. The maximum depth of imaging is currently limited by working distance of the objective lens. Figure 6(a)–(d) show images of a colposcopically normal cervical biopsy with the image plane located at about 40, 80, 100 and 150 µm below the tissue surface, respectively. Images of an abnormal cervical biopsy from the same patient are shown in Fig. 6(e)–(h), with the image plane 40, 60, 90 and 120 µm beneath the tissue surface, respectively. The cell nuclei are clearly resolved and useful information such as nuclear density, nuclear area and nuclear/cytoplasmic ratio can be extracted from the images. In general, a decrease in signal level is observed when the image plane is moved deeper into the tissue, which is expected because light is attenuated (mostly scattered) in tissue. Variations in signal level and contrast, however, exist from sample to sample. The source of these variations may be intrinsic to tissues or dependent on the process of sample preparation, or a combination of the two.
Tissue sections 200 µm thick were cut from cervical biopsies in the orientation perpendicular to the surface of the epithelium. Six per cent acetic acid solution was added prior to imaging. Figure 7(a)–(c) show images of a tissue section from a normal biopsy and Fig. 7(d)–(f) show images of an abnormal tissue section from the same patient. For both tissue sections the images are arranged such that the stroma region is on the left and top of the epithelium is on the right. The basement membrane of epithelium is shown clearly in Fig. 7(a) and (d). The nuclear density in the images taken from the abnormal biopsy is clearly increased as compared to the images taken from the normal biopsy. The area of individual nuclei in the images of the abnormal tissue is also greater than that of the normal tissue.
The FCRM was used to image human epithelial cells in the lip in order to assess the feasibility of in vivo imaging by this approach. Figure 8 shows an image of epithelial cells in the lip of one of the authors (K.B.S.). The maximum output power that entered the tissue was less than 40 mW, which was comparable to that used by Rajadhyaksha et al. (1999a). Six per cent acetic acid solution was added to enhance the contrast of the cell nuclei. Images were easily blurred by motion of the volunteer, so videos were saved into AVI files and individual frames of still images were extracted from the video files. The same procedure of background subtraction and contrast enhancement were applied to the images afterwards. The cell nuclei were clearly resolved in the images and showed images with higher contrast than those of the ex vivo biopsies.
The results shown here provide evidence that backscattered light confocal imaging of biological tissues can be carried out in near real time through flexible fibre optic bundles. The spatial resolution, sensitivity and signal-to-background of the FCRM are sufficient to permit imaging of subcellular structures in epithelial tissues. Images of cervical biopsy specimens (Figs 6 and 7) show significant differences in nuclear density and area between the colposcopically normal and abnormal biopsy. Morphological information such as nuclear to cytoplasmic ratio can be extracted from confocal images (Collier et al., 2000) and used to distinguish between pre-cancerous and normal tissue (Collier et al., 2002). The results suggest that such a device would be useful for in situ detection of pathology. The outer diameter of this prototype is 3.8 cm; this is small enough so that it could be used to image the cervix during colposcopy (Utzinger et al., 2001). We are currently developing a second prototype which incorporates a miniature objective lens, along with an appropriate mechanism to hold the objective lens stably against the tissue and provide a controllable axial scanning. This device should enable imaging of the oral cavity and other internal organ sites.
The lateral resolution of the FCRM is limited by the pixelated nature of the fibre bundle. According to the Nyquist theorem, for fully reconstructing the image we need to sample the image at a spatial frequency at least two times that corresponding to the diffraction limit of the objective optics (Webb & Dorey, 1995). Here the sampling frequency posed by the fibres is much lower than the required Nyquist frequency. The incidence of undersampling may cause aliasing in the images. However, for our application we are interested in the area, density and distribution of cell nuclei. In epithelial tissue the cell nuclei are 5–15 µm in diameter and about 10–20 µm apart. The sampling pitch (1.8 µm) of the FCRM is sufficient to locate cell nuclei and resolve the areaof a nucleus to certain accuracy.
The use of a fibre bundle with multimode fibres in a laser scanning confocal reflectance microscope may result in a high degree of speckle (Juškaitis et al., 1997). Our approach is to use a laser source with a short coherence length to reduce interference between reflections from various surfaces in the light path and backscattered signal from tissue. For epithelial tissue which the FCRM is designed to image, there is no solid, highly reflective planar surface to generate considerable level of speckle. This is confirmed by the images of biological samples in which no prominent speckle is observed.
A resonant galvanometer is selected to achieve the high-speed line scan because of its optical and mechanical simplicity, fairly low cost, high reflection efficiency, adjustable scan angles and negligible variations from one scan line to the next. The major disadvantage of using a resonant galvanometer is the sinusoidal nature of the scan. In the FCRM, only the central part of the scan in one direction is used as the active line, resulting in a low duty cycle of about 40%. Non-linearity is noticeable at the left and right edges of raw images. The image distortion was corrected by image processing for all images presented in the Results section. A software subroutine was made to correct the distortion in the image of the grating (Fig. 2a). The same subroutine was applied to all images of biological samples. Another solution is to acquire the pixels at intervals that are not uniform in time but correspond to equal intervals in image space along the lines (Montagu, 1991).
Index matching at the proximal and distal surfaces of the fibre bundle sufficiently reduces the specular reflections produced there to enable acquisition of high contrast images. Image contrast is degraded somewhat by the residual specular reflection from the proximal fibre surface. This background can be further reduced by polishing the fibre bundle end face at a small angle instead of perpendicular to the optical axis, provided that the angle is small enough so that the focal spot of L1 remains nearly diffraction-limited over the active areas of the bundle. A constant background level can be subtracted from the video signal by blocking the dc component of the APD output. However, noise generated by the electronic components needs to be carefully controlled because the electronics work in the radio frequency range. In practice, moving the pinhole axially off its optimal position can also reduce the background. The pinhole will no longer be conjugate to the illumination/detection fibre so part of the specular reflection from the fibre surface will be blocked. The trade-off is, of course, worse spatial resolution.
The images presented here have a pixelated appearance because of the fibre bundle. Gaps between fibres within features can be filled by image processing techniques such as low pass filtering, median filtering or morphological operations (Thiran & Macq, 1996). Interpolation can also be utilized to produce regular, smooth-looking images. Because of the rapid acquisition speed, these are difficult to implement in real time. The pixelation is perceived most strongly in the still images presented here. The raw videos displayed on the computer screen are perceived to be much less pixelated. Persistence of vision provides a time-averaging effect on the image sequence so the effects of noise and pixelation are reduced. Small movements of the object across the field of view greatly facilitate perception and recognition of small or dim features over a constant background in the videos. Therefore, the current system set-up is preferable for a real time diagnostic imaging device.
We have built a FCRM that has sufficient resolution and sensitivity to image biological samples in cellular and subcellular level at half video rate. No fluorescent stain is needed because backscattered light from the tissue is detected. Images of epithelial cells, excised tissue biopsies and the human lip in vivo have been obtained. The evident differences in nuclear morphology between normal and abnormal biopsy images suggest that the FCRM has diagnostic value for detection of cervical pre-cancers. We believe that FCRM will be highly useful for the recognition and monitoring of pathology in epithelial tissues in vivo.
This project is supported by an NIH grant (R01 CA82880). The authors thank Ina Pavlova for sample preparation.