A scanning system for angle-resolved low-coherence interferometry

Abstract

Angle-resolved low-coherence interferometry (a/LCI) detects precancer by enabling depth-resolved measurements of nuclear morphology in vivo. A significant limitation of a/LCI is the point-probe nature of the method, sampling <0.5 mm² before probe relocation is necessary. In this work, we demonstrate a scanning method capable of assessing an area >100 mm² without repositioning. By utilizing a reflection-only three-optic rotator (ROTOR) prism and two-axis scanning mirror, we demonstrate radial scans of a sample with a linear range of 12 mm and a full rotational range of 180°. Use of this design will improve the diagnostic utility of a/LCI for wide-area screening of tissue health.

Angle-resolved low-coherence interferometry is a coherent light scattering method which obtains depth-resolved measurements of nuclear morphology in tissue [1, 2]. a/LCI collects scattered light over a range of angles near the backscattering direction, which is analyzed to determine nuclear size and optical density, a significant biomarker of dysplasia. A substantial body of work has shown the effectiveness of a/LCI for precancer detection in the cervix [2, 3], esophagus [4, 5], and colon [6]. While these studies have demonstrated high diagnostic accuracy, the inability to scan large areas of tissue without mechanical displacement of the probe has been a limiting factor in clinical application.

The lack of scanning capability in traditional a/LCI is problematic for several reasons. The mechanical translation of an optical probe is often limited by anatomic constraints, as is the case in luminal organs such as the esophagus and colon. Even with less constrained access, manual scanning of large tissue areas such as the cervix are often cumbersome for the physician, and probe repositioning adds substantially to the procedure time. Continual relocation of the probe may also result in poor tissue apposition (flatness), leading to a corruption of the angular scattering signal. Finally, the limited sampling area (<0.5 mm²) requires a method for co-registration of the optical probe with histological biopsy. Clearly, an improved method of sampling large tissue areas is needed to improve the clinical utility of a/LCI.

The design of a scanning a/LCI system must overcome several modality-specific challenges. In most coherent imaging modalities that require scanning, such as OCT [7] or scanning confocal microscopy [8], the beam is collimated between the objective lens and the pupil plane, resulting in very little beam divergence between those elements, and enabling the use of compact scanning mirrors in the pupil plane. However, a/LCI collects scattered light across a range of angles that broadens the required aperture at the pupil. Scanning a/LCI therefore requires a larger scanning mirror than the numerical aperture alone may suggest. Furthermore, telecentricity is a paramount requirement, as a/LCI must maintain a fixed illumination angle to ensure reproducible mapping of scattering angles. This prevents the use of more compact geometries that are not telecentric. This combination of requirements prescribes a bulky optical system to achieve a useful field of view, which is a major limitation for endoscopic imaging.

However, these constraints can be overcome by realizing that the broad aperture requirement for scanning a/LCI occurs only along the axis of the collection fibers, termed the collection axis. Thus, the pupil plane contains unused space along the orthogonal axis (the scanning axis) to perform scanning in that direction, even though this would only provide 1D scanning. To achieve 2D scanning, the orientation of the collection axis and the scanning axis can be rotated together, such that they remain orthogonal and a 2D field of view can be covered in a radial pattern. In this work, we describe a method of producing radial a/LCI scans using a reflection-only three-optic rotator (ROTOR) prism and two-axis scanning mirror. The ROTOR was designed to simultaneously rotate the illumination beam and collection axis about the optic axis. A two-axis scanning mirror was then employed to scan the illumination beam perpendicularly to the collection axis, producing an “asterisk”-shaped scanning profile. Characterization of the scanning system was performed using polystyrene microsphere phantoms, as described below.

The scanning a/LCI system was constructed using a previously developed clinical a/LCI engine and probe [9]. Briefly, light from a superluminescent diode (Superlum SLD-MS, λ = 830 nm, FWHM = 20 nm) is propagated to the distal end of the probe using a polarization-maintaining (PM) fiber. Light from the PM fiber is collimated by a GRIN lens onto the sample plane, producing elastic scattering. The PM fiber is offset from the optical axis, resulting in a collimated and oblique beam at the sample. The elastically scattered light from the sample is relayed by the GRIN onto the distal end of a coherent fiber bundle (Schott Inc., Southbridge, MA), mapping angular scattering to distinct elements within the bundle. The face of the proximal end of the bundle is then relayed with a 4f imaging system to the slit of an imaging spectrograph (SP-2150i, Princeton Instruments, Acton, MA) through a beamsplitter, where it is combined with a collimated reference field and detected with a CCD camera (PIXIS: 100, Trenton, NJ) to enable coherence-gating.

The scanning a/LCI system extends the sample plane of the clinical a/LCI system using two 4f lens relays, with f₁, f₄ = 30 mm and f₂, f₃ = 100 mm, using 1” achromatic doublets [Fig. 1]. This allows for incorporation of the ROTOR prism and scanning mirror. The ROTOR prism is composed of three mirrors which rotate as a unit relative to the optic axis. Two on-axis elliptical mirrors are oriented at 55° using a custom 3D-printed mount, while a third mirror is held off-axis in a right-angle cage system. This entire structure is then mounted on two 30 mm rotation mounts, which allow rotation of the entire assembly over a full 360°. Analogous to a Dove prism or k-mirror [10], a rotation of the ROTOR over a given angle results in a rotation of the excitation beam and collection axis at twice that angle [Fig. 2]. An added benefit is that the polarization of the illumination beam will remain in-plane with the collection axis, which is necessary for repeatable scattering measurements. By employing radial scans across a 180° range (ROTOR rotation of 90°), a circular region on the image plane can be characterized.

Fig. 1.

Scanning a/LCI system schematic showing the illumination beam (red) and the beams from two collection angles (green and blue). Rotation of the ROTOR prism (red box) results in a rotation of the scanning and collection axes by twice that angle. The lens and sample planes are denoted L’ and S’, respectively. An animation demonstrating scanning motion is available online (see Visualization 1). (Inset) Pictorial illustration of the scanning axis and the collection axis.

Fig. 2.

Cross-sectional diagrams of (A–C) the lens plane L’, and (D–F) the sample plane S’, for ROTOR angles of (A, D) 0°, (B, E) 22.5°, and (C, F) 45°. This results in rotation of the scanning axis of 0°, 45°, and 90°, respectively. Colored spots correspond to beams in Fig. 1.

A gimbal-mounted two-axis scanning mirror was incorporated at the distal end of the system to scan the illumination beam perpendicular to the collection axis [Fig. 1]. Using this configuration, tilting of the scanning mirror produces a line-scan on the image plane. This line may then be rotated around the optical axis using the ROTOR prism, and the mirror can be tilted in a different direction to perform another line-scan at twice the rotation angle of the prism [Fig. 2]. Repeating this method over different positions of the ROTOR produces an asterisk-shaped scanning profile capable of reaching any point within a circle, with its diameter determined by the linear scanning range. In this way, scanning a/LCI produces a two-dimensional map of depth-resolved angular scattering profiles.

a/LCI image processing has been described previously [11]. Briefly, four exposures were taken during each scan (total, sample, reference, and dark, 100 ms each) to isolate the interferometric term in the acquisition. The wavelength axis of each interferogram was converted to wavenumber $(k=\frac{2\pi }{\lambda })$, and resampled to be linear in k. Third-order dispersion compensation was performed digitally, followed by a Fourier-transform to produce a map of intensity for each scattering angle as a function of depth. At each point, ten a/LCI scans were averaged to improve SNR, for a total acquisition time of <8 seconds per point (ten background-subtracted scans per point, four exposures per scan). Each averaged depth scan was binned over ~480 µm of depth to produce a single angular scattering profile. This depth was chosen as the range over which reliable scattering signal was present in all samples and is typical of the thickness of the epithelium in many human tissues.

a/LCI identifies the scatterer size by comparing the measured spectrum to a library of precomputed Mie scattering spectra. The library used in this study was created using MiePlot [12] for use with polystyrene microsphere phantoms in cured PDMS. Library parameters included sphere and medium refractive indices of 1.58 and 1.41, with λ = 0.83 µm and a 1% standard deviation in size distribution. Scatterer diameter was varied between 5–18 µm in increments of 0.1 µm, for a total of 131 angular spectra. Both the acquired and library spectra were normalized, and a second-order polynomial fit was subtracted to isolate the oscillatory component. A mild low-pass filter was applied to remove high-frequency noise. This spectrum was then compared to each spectrum within the library, and the spectrum with the lowest error (as determined by chi-squared analysis) was used to estimate the scatterer size. Sample spectra, along with their best fits, are shown in Figure 3.

Fig. 3.

Representative angular scattering measurements from 6 µm (left) and 10 µm (right) microsphere phantoms. The acquired profile is shown in red, while the best fit is shown in blue. In this case, both spectra were correctly identified as 6.0 and 10.0 µm, respectively.

Microsphere phantoms were constructed using standard methods. ~100 µL of polystyrene spheres in water (Thermo Fisher Scientific) were centrifuged into a pellet, dried in a vacuum chamber, and mixed with 1 mL PDMS elastomer base and curing agent (Sylgard 184, Dow Corning), pre-mixed at a ratio of 10:1. The sample was sonicated for 30 minutes to reduce sphere aggregation, placed in a vacuum to remove bubbles, and left to cure overnight in a 3D-printed mold to construct 5 mm cubes, which were used to construct heterogeneous scanning phantoms. To validate the performance of the device with scanning and relay optics, a/LCI measurements of sphere size were acquired both with and without the scanning system. The original a/LCI probe was designed to scan samples at the focal plane of the GRIN lens, with the sample and probe separated by a No.1 coverslip. Five microsphere phantoms (6, 8, 10, 12, and 15 µm spheres) were imaged (N = 6 each) and compared with the expected value. These measurements were repeated after introduction of the scanning optics.

We validated off-axis performance of the system by manually tilting the scan mirror to translate the beam perpendicularly to the collection axis. Three phantom cubes (5 mm across) were arranged linearly in the sample plane to assess the useful scanning range [Fig. 5]. This range was calculated as the linear range over which sphere size was correct to within a wavelength of the illumination light (in this case, <0.83 µm). The beam was translated in 1 mm increments over a total range of 14 mm, and the results were analyzed.

Fig. 5.

Demonstration of line-scanning capability of the scanning a/LCI system. Linear scanning was achieved by rotation of the scanning mirror in a direction perpendicular to the collection axis. Sub-wavelength accuracy (D ± λ) was achieved across a range of 12 mm. Acquisition time for each point was <8 seconds, during which each point was sampled ten times consecutively to improve SNR.

Finally, the system was validated for two-dimensional scanning using a 3×3 array of microsphere phantoms [Fig. 6]. Phantoms with diameters of 6, 10, and 15 µm were arranged in a unique pattern (10 µm on the corners, 6 µm at the edges, and 15 µm in the center) to create a sample with spatially varying scattering features. Three samples were taken along a scan line at intervals of 5 mm (−5 mm, 0 mm, and +5 mm from the optic axis) to collect one sample from each unique phantom area within the middle row of the array. The scan line was then rotated using the ROTOR, and the linear scan was repeated, pausing at each scan location. The prism was positioned at 0°, 45°, and ±22.5° to achieve a rotation of the scanning axis of 0°, 90°, and ±45°. This produces a single sample point from each phantom cube in the array. The center point was resampled in each scan, though no deviation in the size measurement was observed.

Fig. 6.

Validation of two-dimensional scanning ability of the system across a 3×3 array of scattering phantoms. Each phantom is correctly identified with sub-λ accuracy. Mean error was ~2.9%, with a maximum error of 6.7% (lower middle, 6 µm spheres were identified as 5.6 µm). Maximum absolute error was 0.8 µm, though this was on a large sphere (15 µm) resulting in a percentage error of only 5.3%.

The performance of the system both before and after introduction of the scanning optics is shown in Figure 4. In both cases, highly accurate performance was achieved, with a mean absolute error in diameter prediction of 0.15 µm without scanning optics, or 0. 23 µm with the scanning optics, relative to the NIST traceable value. This error is much smaller than the increase in nuclear diameter observed in precancer, which is typically 2–4 µm [3]. In both cases, sizing error never exceeded 0.4 µm, with excellent coefficients of determination (r²) of 0.997 and 0.993, respectively, relative to the line of perfect agreement. To compare the performance of the system pre- and post-modification, we computed standard errors of the corresponding correlation coefficients (r) and performed a t-test (df = 6). No significant difference in accuracy was found between the two cases (p>0.97).

Fig. 4.

Calibration curves for microsphere phantoms before (left) and after (right) introduction of the scanning optics. No significant loss of performance was observed by modifying the system for scanning.

The line-scanning capability of the system is demonstrated in Figure 5. Determination of the scatterer size with sub-wavelength accuracy was achieved for all sample points across a 12 mm scan range (mean error = 0.17 µm). Outside of this range, fitting errors start to exceed λ = 0.83 µm due to a number of factors. Empirically, we observed defocus (leading to blurring of Mie fringes), distortion of the angle-to-pixel mapping, and aperture limitations. Slight changes in optical pathlength were also observed, though these could be corrected in real time using a motorized reference arm, up until the point that defocus sufficiently blurred the signal.

Two-dimensional scanning is demonstrated in Figure 6. Due to the geometry of the system, the aperture of the distal scanning mirror limits the range of received scattering angles. Since the scanning mirror must be tilted to achieve the line scan, using a 1” mirror in a gimbal mount reduced the aperture in certain directions, unlike the larger elliptical mirrors within the 3D-printed mount used in the ROTOR prism. This resulted in a slight reduction in aperture when image rotation exceeded ~30°. The useable range of scattering angles was measured to be [1.15°, 24.85°] for 0° of image rotation, [1.15°, 20.35°] for 45° of image rotation, and [1.15°, 17.35°] for 90° of image rotation (the worst possible case). In each case, fitting was performed by limiting the scattering library to the angular range of available information. Fortunately, we found no perceivable loss in performance when sphere sizing was performed using the reduced range of angles. Each radial scan was performed over a 1 cm range, to fit within the measured 12 mm scan range while sampling each point within the 3×3 phantom grid. At each point, determination of the sphere size with sub-wavelength accuracy was observed. Mean absolute error was 0.26 µm, which is consistent with the 0.23 µm error observed during calibration of the scanning optics [Fig. 4].

Though this implementation of scanning a/LCI demonstrates advances that will be essential for clinical use, some limitations remain which must be addressed prior to use in vivo. A tradeoff exists in this design between the aperture and focal length of the distal optic. A shorter focal-length objective lens reduces the pupil size for a given range of angles, while the clear aperture of the optic limits the field-of-view. At a 1-inch diameter (our desired field of view), commercially available achromatic lenses are typically not offered shorter than the 30 mm lens used in this work, which resulted in the need for a 25 mm scanning mirror. To progress to a less cumbersome form factor, a customized multi-element lens design will likely be needed to improve the focal power beyond the limits of off-the-shelf components. Further, the ROTOR must be miniaturized to create a more compact, clinically viable system.

Calibrations [Fig. 4] were performed using on-axis measurements for consistency; however, we noticed a slight curvature in the focal plane, causing small amounts of defocus between on- and off-axis images. This generally manifested in the angular spectra as a slight (~3%) magnification in the angular dimension, leading to minor errors in scatterer size measurement. Because most points in a 2D image are off-axis, we chose to optimize the sample’s position and angle calibration to best fit these points. This is exemplified in Figure 6, in which the center point is characterized with modestly larger (0.8 µm, or 5.3%) but still sub-wavelength error, while off-axis points have very small errors (mean 0.19 µm, max 0.4 µm). Future work using an f-theta lens will eliminate defocus that results from off-axis aberrations.

To conclude, we have demonstrated a scanning a/LCI system capable of analyzing large sample areas without probe relocation. By utilizing the novel ROTOR prism with a scanning mirror, we have demonstrated both line-scanning and area-scanning capability. Future work will focus on precision optical design to minimize issues associated with off-axis aberration. Miniaturization and automation of the scanning components will further enable development of a clinically viable implementation.