Dynamic contrast in scanning microscopic OCT

Michael Münter; Malte vom Endt; Mario Pieper; Mario Pieper; Malte Casper; Martin Ahrens; Martin Ahrens; Tabea Kohlfaerber; Ramtin Rahmanzadeh; Peter König; Peter König; Gereon Hüttmann; Gereon Hüttmann; Gereon Hüttmann; Hinnerk Schulz-Hildebrandt; Hinnerk Schulz-Hildebrandt; Hinnerk Schulz-Hildebrandt

doi:10.1364/OL.396134

Optical coherence tomography (OCT) is an imaging technique that is based on the interferometric measurement of backscattered light. Providing fast, high-resolution sectional images of biological tissue, OCT has become a valuable tool in different clinical fields such as ophthalmology and dermatology [1]. One especially useful characteristic of OCT is the decoupling of the axial resolution and lateral resolution [2]. For resolving cells and even subcellular structures, it is necessary to increase the resolution of OCT from typically 10 to 1 µm. Optical coherence microscopy [3], micro-OCT [4], and microscopic OCT (mOCT) [5] typically reach a resolution comparable to other microscopic techniques such as confocal or nonlinear microscopy [6]. The use of ultra-broadband light from coupled SLDs, Ti:sapphire lasers, or supercontinuum light sources increases systems axial resolution to 1 µm. Using a high-numerical-aperture (NA) microscope objective, a lateral resolution of 1 µm can be achieved. In OCT, coherent noise (speckle) reduces contrast [7] and often makes it impossible to differentiate individual cells based on their scattering properties. Recently, a novel approach to analyze data of full-field OCT (FF-OCT) was demonstrated and termed dynamic OCT [8]. This method exploits metabolically driven dynamic scattering changes of cellular structures to enhance contrast. FF-OCT illuminates the object over a large area with spatially and temporally incoherent light. A horizontal sectional (en-face) image is created by interference between the light scattered by the sample and the light reflected from the reference mirror. To create cross-sectional images, it is necessary to move the microscope or the sample in axial direction. Dynamic contrast is obtained by directly analyzing temporal fluctuation of the intensity of the interference in each pixel of the imaging camera. Correlation functions or power spectra of the signal fluctuations are evaluated over a few seconds at a typical sampling frequency of 100 Hz with ${{1}}\;{\rm{nW}}/{{\rm{\unicode{x00B5}} m}^2}$ [9]. Dynamic FF-OCT detects motion of cellular structures with nanometer sensitivity and 1 µm spatial resolution delivering images of unfixed biological tissue which appear to be similar to conventional histological techniques up to a depth of 100 µm. The disadvantages of this technique are a low penetration depth in scattering tissues, the limited use of scattered photons from only a 1 µm thick sample plane, the high sensitivity to axial movements of the sample and the incapability to quickly acquire large volumes. In contrast, scanned frequency-domain OCT (FD-OCT) provides tomographic imaging simultaneous over a certain depth range and uses all scattered photons for one A-scan. Additionally, it has more degrees of freedom in defining the imaged region.

Fig. 1. (a) Schematics of the mOCT setup. LS, light source; FB, filter box; FC, 50/50 fiber coupler; C, collimators; G, 2-axis galvanometer mirror scanners; TS, telescope system; O, microscope objective; DC, dispersion compensation; A, aperture; RR, retroreflector; DAQ, data acquisition device; S, spectrometer; PC, computer for data acquisition and scanning control. (b) dOCT image is generated by color-coding of three frequency bands after Fourier transforming the temporal signal fluctuations in each pixel of the B-scan image stack.

Download Full Size | PDF

However, scanned OCT typically lacks phase stability between adjacent A- and B-scans and suffers from a shallow depth of focus, when using high-NA objective lenses, e.g., at a 1 µm focus diameter, the confocal parameter is only 17 µm. Outside this range, lateral resolution decreases, and dynamic contrast may be annihilated. This Letter demonstrates that scanned FD-OCT produces surprisingly rich dynamic contrasted B-scans and volumetric images of freshly excised murine tissue samples of tongue and liver, despite the high imaging NA. The mOCT setup illustrated in Fig. 1 uses a supercontinuum light source (SuperK EXTREME–EXW-OCT, NKT Holding, Denmark) in combination with the spectrometer covering a spectral range from 550–950 nm and achieves 1 µm axial resolution. Emission of the supercontinuum light source was bandpass filtered (SuperK SPLIT, NKT Holding, Denmark) and injected into a fiber-based Michelson interferometer. A broadband 50/50 fiber coupler (TW630R5A2, Thorlabs Inc., U.S.) splits the light into the sample and reference arm. Light in the sample arm is collimated (60FC-L-4-M25-02, Schäfter ${+}$ Kirchhoff GmbH, Germany) and scanned via a pair of galvanometer scanners (6210 H, Cambridge Technology, U.S.) and an achromatic telescope (SL50-CLS2 and TL200-CLS2, Thorlabs Inc., U.S.) in the back focal plane of a ${10} \times /{0.3}$ NA microscope objective (HCX APO L ${10} \times /{0.3}$ WUVI, Leica Microsystems, Germany). Light in the reference arm was collimated (60FC-L-4-M25-02, Schäfter + Kirchhoff GmbH, Germany), attenuated by a variable neutral density filter (NDC-50C-2M, Thorlabs Inc., U.S.), and matched in dispersion to the light of the sample arm by a 14 mm thick piece of SF57 glass substrate (Casix Inc., China). Reference light from a retroreflector (PS975M-B, Thorlabs Inc., U.S.) and sample light were brought to interference in the custom-designed spectrometer (Thorlabs GmbH, Germany). Two different cameras (SprintSpL4096 140 km, Basler, Germany and OctoPlus CL, Teledyne, e2v, Canada) were used for maximum A-scan rates of 127 and 248 kHz, respectively. Synchronization of the scanner, spectrometer, and data acquisition software was realized using a USB data acquisition device (NI USB-6251, National Instruments, U.S.). For volumetric imaging, at each line, 100 B-scans with 500 A-scans each was taken at a 30 kHz A-scan rate with the Sprint SpL4096 camera, which results in an effective B-scan rate of 46 Hz. The average irradiance at the sample was ${{40}}\;{\rm{\unicode{x00B5} W\!/\!}}{{\rm {\unicode{x00B5}}{m}}^2}$. Adjacent A-Scans were axial aligned using the reflection of the glass plate’s surface to compensate for sample motion and image field curvature. Volumes were recorded by stacking a series of B-scans in y direction. With the OctoPlus CL 150 B-scans with 500 A-scans, each was sampled at a 100 kHz A-scan rate. With an effective B-scan rate of 108 Hz and a total recording time of 1.39 s, the frequency of signal fluctuations was measured between 0 to 54 Hz. The raw data from the spectrometer were Hann windowed and Fourier transformed to obtain the OCT images. Residual dispersion was numerically compensated for using a fifth-order polynomial for the spectral phase error correction. Polynomial coefficients were determined by optimizing image quality, which was measured by the Shannon entropy [10]. Spectral processing and dynamic contrasting of the data were performed in MATLAB (MATLAB R2019b, The MathWorks, Inc., U.S.). At each voxel, the temporal variations of the absolute value of the OCT signal were evaluated. The time series was Fourier transformed, and the integral amplitude was calculated in three frequency bands [Fig. 1(b)]. The pixels were color-coded in an RGB image, representing different time scales of motion activity [11,12]. Blue represents slow motion frequencies (0–0.5 Hz), green represents medium motion (0.5–5 Hz), and red represents fast motion (5–25 Hz). Using the phase, we were also able to reconstruct individual structures, but the achieved image quality was significantly worse. It may be connected with speckle noise or phase noise from the scanning. For display, the color channels were scaled logarithmically, and image contrast was enhanced using contrast-limited adaptive histogram equalization [13]. Brightness outside the focus was adjusted to the peak intensity in the focus. Finally, a ${{3}} \times {{3}}$ median filter was applied to each color channel. Reference measurements were acquired with a commercially available FF-OCT device (LightCT, LLTech Inc., France), which was designed for real-time optical biopsy using dynamic OCT contrast. The device uses an LED light source with a spectral width of 100 nm at a central wavelength of 565 nm. Together with a 0.3 NA water immersion objective, an isotropic resolution of 1 µm is achieved. In contrast to FD-OCT which first has to reconstruct A-scan, before signal fluctuation can be evaluated, the FF-OCT directly analyzes the interference pattern on the camera. The low axial and lateral coherence of the illumination assures a full correspondence of each pixel to one location in the sample. The same frequency bands were used by the LightCT for calculating RGB images from the signal fluctuations. For all measurements, the sample, either freshly excised tongue or liver tissue of C57BL/6 mice, was placed in Ringer’s solution in the specially designed sample holder which was supplied with LightCT. Mouse tongue was selected as an example for a stratified epithelium due to its good accessibility. With inoculated cells from various tumor cell lines, murine tongue is a well-accepted model for ENT tumors. Tissues were imaged by slightly pressing the tissue surface against the quartz cover glass plate to which the objective was coupled using silicon immersion oil. Despite the sufficient resolution, mOCT and LightCT OCT images of liver tissue showed only weak contrast. The murine liver shown in Fig. 2 was imaged over a field of view (FOV) of ${{270}}\;{\rm{\unicode{x00B5}{\rm m}}} \times {{285}}\;{\rm{\unicode{x00B5}{\rm m}}} \times {{550}}\;{\rm{\unicode{x00B5}{\rm m}}}\;({\rm{xyz}})$. Fine cellular details of the tissue are difficult to identify without dynamic contrasting in averaged en-face OCT image planes shown in Figs. 2(a) and 2(c). Brighter details might indicate extracellular matrix structures and spherical void cell nuclei. However, images with the dynamic contrast clearly show cellular structures [Figs. 2(b) and 2(d)]. Cell nuclei and cell borders can be distinguished with high contrast. In FF-OCT, cell nuclei appear to be mostly dark, while mOCT shows mostly red nuclei with some dark substructure. The extracellular matrix is a bluish color, indicating slow movement. Resolution and sensitivity of FF-OCT do not seem sufficient enough to catch the dynamic changes in the nuclei. They appear with red color and a distinct substructure in FF-OCT only at higher NAs [9]. Besides, mOCT and LightCT images show the tissue structures in similar colors. For example, the cytoplasm of the cells colored in the LightCT image in light green can be seen in a similar color in the dynamic mOCT image. The intensity of color components differs slightly, since the measurements were taken 2 h apart, and post-processing is not exactly identical. A volumetric dmOCT

Fig. 2. (a) LightCT en-face OCT image of murine liver. (b) LightCT dynamic OCT en-face image with corresponding regions. (c) Averaged mOCT en-face image, which was acquired at the same region. (d) In the corresponding dynamic mOCT image, hepatocytes become visible with nuclei. (e) Cropped volume representation; size: ${{270}} \times {{285}} \times {{135}}\;{\rm{\unicode{x00B5}{\rm m}}}\;({\rm{xyz}})$ (see Visualization 1); scale bar, 100 µm.

Download Full Size | PDF

image cropped in z direction is shown in Fig. 2(e) and Visualization 1. Cellular details are also visible in cross-sectional imaging of the bottom side of murine tongue by dynamic contrast mOCT (Fig. 3) imaged over a FOV of ${{500}}\;{\rm{\unicode{x00B5}{\rm m}}} \times {{380}}\;{\rm{\unicode{x00B5}{\rm m}}}\;({\rm{xz}})$. This part of the tongue consists of five layers, a cornified stratified squamous epithelium, the basal layer, and two subepithelial layers of connective tissue and skeletal muscle cells [Fig. 3(a)]. The average of 150 mOCT B-Scans only shows a faint image of the tissue structures [Fig. 3(b)]. On

Fig. 3. (a) HE stained histology of the imaged sample at a different location: (I) cornified layer, (II) granular & spinous layers, (III) basal layer, (IV) lamina propria, (V) muscle, and (VI) glass plate. (b) OCT image of mouse tongue; lamina propria (IV) can be identified by brighter contrast. (c) Corresponding dynamic contrast mOCT image with a focus in the basal cell layer; (I–V) and even cell nuclei (*) are visible. (d) Dynamic contrast mOCT image with a focus in the lamina propria; the image size is ${{380}} \times {{500}}\;{\rm{\unicode{x00B5}{\rm m}}}\;({\rm{zx}})$; scale bar, 100 µm.

Download Full Size | PDF

the right side of the OCT image, only four tissue layers can be guessed. On the left side, a layered structure is not visible at all. Using dynamic contrast, the five histologically relevant layers become clearly visible in the whole FOV [Fig. 3(c)]. Below the glass plate, the cornified layer, the layer of the squamous epithelium with the transition to the granular and spinous layer, is contrasted as a purple and a green-purple layer, respectively. Deeper, the basal layer, which is characterized by yellow-colored cell nuclei, is displayed. Since the optical focus was placed in this region, the basal layer is imaged at the highest resolution, and structures even within the nuclei became discernible. Below, the lamina propria is visible as a purple band again followed by the green-colored skeletal muscle fibers. Shifting the focus below the basal layer increases the visibility of the muscle layer [Fig. 3(d)]. Dynamic contrast mOCT images show an excellent match to structures in HE stained histological sections [Figs. 3(a), 3(c), and 3(d)]. The images are completely free of speckle noise and have remarkably high contrast. Cellular structures are discernible, even well beyond the focal plane, which are typically invisible in averaged OCT images, even when exactly focused. The high contrast permits a reliable segmentation of layers and even individual basal cells. These results demonstrate that implementing dynamic contrast in a scanning mOCT is possible and yields dramatically increased contrast which allows identification of individual epithelial cells. While the concept of dynamic contrast was developed for FF-OCT, here we show that it can also be used with scanning mOCT. This is somewhat surprising as dynamic contrast is inferred from motion on nanometer scale, and scanning can lead to additional phase noise potentially overcasting phase fluctuation caused by motion in the sample. However, FD-OCT has ${\sim}{{15}}\;{\rm{dB}}$ sensitivity advantage compared to FF-OCT; therefore, it is more suited to image low reflective structures [14,15]. A direct comparison of sensitivity and the signal-to-noise is challenging, since the general image forming process between scanned and FF-OCT is different. FF-OCT generated dynamic contrast based on the interferometric raw data, whereas FD-OCT requires more complex numerical reconstructions. mOCT works confocally and detects mainly ballistic photons. In general, photon shot noise performance can be achieved. Interference is used to tag photons coming from different depth scattering. FF-OCT relies on coherence gating for axially and laterally restricting the origin of the photon scattering. The detected signal on the camera has a huge incoherent background, but massive parallel detection compensated for this. It is expected that the different noise sources (i.e., photon shot noise, detector noise, and mechanical phase noise) will influence image quality of FF-OCT and mOCT differently and in a complex way. Since the biological structures observed with mOCT are similar to FF-OCT, as well as to histology, and we still achieve at least comparable image quality, we hypothesize that scanning has no adverse effect on dynamic contrast OCT. Instead, FD-OCT offers several advantages. First, it provides inherently sectional images across the tissue layers, which show several tissue layers at a glance as known from conventional histological sections. Reconstruction of a B-scan from a series of en-face FF-OCT images is time-consuming, and image quality can easily be degraded by small sample motions. Secondly, acquiring a sectional image plane reduces the sensitivity to axial sample motion, which has the most severe influence on dynamic contrast. While axial sample motion is difficult to detect and correct in FF-OCT systems [16], in sectional scans, each A-scan can be individually corrected by multiplying an appropriate phase factor. Thirdly, the parallel measurement of the complete depth information in FD-OCT potentially increases the speed for volumetric imaging. Dynamic contrast OCT needs to measure each pixel over a few seconds with a temporal resolution of about 10 ms. This limits the imaging speed. FF-OCT, which currently uses a 2 megapixel camera, is able to image about 1 million voxels/s in the dynamic mode. At 1 µm resolution, this converts into an imaging time of ${{1}}\;{\rm{s}}/{{\rm{mm}}^2}$ for one en-face dOCT image. For volumetric imaging over 300 µm depth, time is increased to ${{300}}\;{\rm{s}}/{{\rm{mm}}^2}$. Scanned OCT with B-scan-based evaluation of the signal fluctuation does not increase imaging speed for volume acquisition. However, FD-OCT with a megahertz A-scan rate could acquire a whole volume within 10 ms [17] and complete volumetric dynamic contrast OCT imaging within a few seconds. Currently, our imaging speed is limited by three components–the galvo scanner, line scan camera, and relative intensity noise (RIN) of supercontinuum light sources. No negative effects of RIN on image quality could be found yet and, recently, we were able to demonstrate mOCT imaging at 1 µm isotropic resolution with a 600 kHz A-scan rate and nearly 12 volumes per second [18]. Although there was higher irradiance, thermal damage was not observed. So far, no significant influence of thermal changes on the dynamic signal has been shown [19]. Forth, dynamic contrast with scanning OCT is easier to implement in an endoscopic system, since scanning OCT has already been developed for rigid and flexible endoscopes [20]. Dynamic contrasting could provide endoscopes with the additional contrast needed for non-invasive, in vivo optical histology. In conclusion, the analysis of the signal dynamics of cellular resolution scanned FD-OCT visualizes epithelial tissues comparable to conventional histology without the need for tissue processing. Loss of lateral resolution outside the focal plane can be compensated for by Bessel beams [21,22], introducing aberrations [23] or a numerical correction of the defocus in post-processing [24]. By evaluating the fluctuation of the OCT signal, speckle-free B-scan and volumetric imaging with high contrast and resolution are achieved. Spectral analysis discriminates and visualizes important cellular components such as nuclei and cytoplasm and allows the identification of different tissue components such as epithelium, connective tissue, and muscle in cross-sectional images which are acquired in seconds. We expect that scanning FD-OCT is more robust against sample motion than FF-OCT and can increase the imaging speed by two orders of magnitude. Given these assets, dynamic contrast OCT might even fulfill the 20-year-old promise of optical biopsies by OCT [2].

During the review process, a related paper on dynamic OCT for scanning OCT was published [25].

Funding

European Union project within Interreg Deutschland-Denmark from the European Regional Development Fund (ERDF) (CELLTOM); German Ministry of Research, Innovation and Science, Helmholtz Center Munich of Health and Environment DZL-ARCN (82DZL001A2); Deutsche Forschungsgemeinschaft (RA 1771/4-1, EXC 2167).

Disclosures

The authors declare no conflicts of interest.

REFERENCES

1. W. Drexler, U. Morgner, R. K. Ghanta, F. Kärtner, J. Schuman, and J. Fujimoto, Nat. Med. 7, 502 (2001). [CrossRef]

2. J. Fujimoto, C. Pitris, S. Boppart, and M. Brezinski, Neoplasia 2, 9 (2000). [CrossRef]

3. J. Izatt, E. Swanson, J. Fujimoto, M. Hee, and G. Owen, Opt. Lett. 19, 590 (1994). [CrossRef]

4. L. Liu, J. Gardecki, S. Nadkarni, J. Toussaint, Y. Yagi, B. Bouma, and G. Tearney, Nat. Med. 17, 1010 (2011). [CrossRef]

5. M. Pieper, H. Schulz-Hildebrandt, M. Mall, G. Hüttmann, and P. König, Am. J. Physiol. Cell. Mol. Physiol. 318, L518 (2020). [CrossRef]

6. G. Cox and C. Sheppard, Microsc. Res. Tech. 63, 18 (2004). [CrossRef]

7. O. Liba, M. Lew, E. Sorelle, R. Dutta, D. Sen, D. Moshfeghi, S. Chu, and A. De La Zerda, Nat. Commun. 8, 15845 (2017). [CrossRef]

8. C. Apelian, F. Harms, O. Thouvenin, and C. Boccara, Bio. Opt. Express 7, 1511 (2016). [CrossRef]

9. O. Thouvenin, C. Apelian, A. Nahas, M. Fink, and C. Boccara, Appl. Sci. 7, 236 (2017). [CrossRef]

10. H. Schulz-Hildebrandt, M. Münter, M. Ahrens, H. Spahr, D. Hillmann, P. König, and G. Hüttmann, Proc. SPIE 10591, 105910O (2018). [CrossRef]

11. O. Thouvenin, C. Boccara, M. Fink, J. Sahe, M. Păques, and K. Grieve, Invest. Ophthalmol. Visual Sci. 58, 4605 (2017). [CrossRef]

12. C. Apelian, C. Gastaud, and C. Boccara, Proc. SPIE 10053, 100531H (2017). [CrossRef]

13. K. Zuiderveld, in Graphics Gems, P. S. Heckbert, ed. (Academic, 1994), p. 474.

14. O. Thouvenin, K. Grieve, P. Xiao, C. Apelian, and C. Boccara, Biomed. Opt. Express 8, 622 (2017). [CrossRef]

15. R. Leitgeb, C. Hitzenberger, and A. Fercher, Opt. Express 11, 889 (2003). [CrossRef]

16. J. Scholler, Opt. Express 27, 19562 (2019). [CrossRef]

17. W. Wieser, B. Biedermann, T. Klein, C. Eigenwillig, and R. Huber, Opt. Express 18, 14685 (2010). [CrossRef]

18. M. Münter, H. Schulz-Hildebrandt, M. Pieper, P. König, and G. Hüttmann, Proc. SPIE 11078, 1107811 (2019). [CrossRef]

19. O. Thouvenin, “Optical 3D imaging of subcellular dynamics in biological cultures and tissues: applications to ophthalmology and neuroscience,” Ph.D. thesis (Université Sorbonne Paris, 2017).

20. H. Schulz-Hildebrandt, T. Pfeiffer, T. Eixmann, S. Lohmann, M. Ahrens, J. Rehra, W. Draxinger, P. König, R. Huber, and G. Hüttmann, Opt. Lett. 43, 4386 (2018). [CrossRef]

21. A. Curatolo, P. Munro, D. Lorenser, P. Sreekumar, C. Singe, B. F. Kennedy, and D. Sampson, Sci. Rep. 6, 23483 (2016). [CrossRef]

22. R. Leitgeb, M. Villiger, A. Bachmann, L. Steinmann, and T. Lasser, Opt. Lett. 31, 2450 (2006). [CrossRef]

23. H. Schulz-Hildebrandt, M. Pieper, C. Stehmar, M. Ahrens, C. Idel, B. Wollenberg, P. König, and G. Hüttmann, Biomed. Opt. Express 9, 636 (2018). [CrossRef]

24. T. Ralston, D. Marks, P. Carney, and S. Boppart, Nat. Phys. 3, 129 (2007). [CrossRef]

25. H. Leung, M. Wang, H. Osman, E. Abouei, C. MacAulay, M. Follen, J. Gardecki, and G. Tearney, Biomed. Opt. Express 11, 2768 (2020). [CrossRef]

Dynamic contrast in scanning microscopic OCT

Abstract

Funding

Disclosures

REFERENCES

Supplementary Material (1)

Cited By

Figures (3)

Optics Letters

Tabea Kohlfaerber	https://orcid.org/0000-0003-3789-539X
Hinnerk Schulz-Hildebrandt	https://orcid.org/0000-0003-4389-9151