1. Introduction

Resolution, contrast, imaging depth and imaging speed determine the clinical value of medical imaging technologies. While computer tomography (CT), magnet resonance imaging (MRI) and ultrasound (US) offer excellent visualization of structures and dynamic processes in the whole body, resolution is limited to a few hundred micrometers. Hence, imaging of dynamic processes at cellular and sub-cellular level is still a challenge. For example, in the field of lung research, the impairment of mucociliary clearance in diseases such as asthma, cystic fibrosis (CF) or primary ciliary dyskinesia (PDC) is of great interest, but investigating the basic component of mucus transport, the ciliated airway cells, in vivo is currently very limited. Also, distinguishing in vivo between benign and malignant changes during cancer surgery is challenging [1]. In vivo, not only a lack of resolution but also a lack of contrast is problematic. Currently, there are no sufficiently reliable non-invasive methods for visualizing malignant cellular changes and biopsy has been for more than hundred years still the standard procedure for tumor diagnosis.

One successful technique which may overcome this limitation is optical coherence tomography (OCT). Since its invention [2] in 1991, OCT has emerged as an important imaging modality in medical diagnostics, especially in ophthalmology [3]. Further clinical applications are found in dermatology, cardiology and gastroenterology [4,5]. OCT uses interference of radiation scattered in the tissue with a reference radiation to retrieve an independent depth information. There are two fundamentally different detection schemes in OCT. One is time-domain OCT (TD-OCT) and the other is frequency-domain OCT (FD-OCT). While in TD-OCT the interference of low coherence radiation is measured directly and the depth information is obtained by moving the sample or changing the path length in one interferometer arm, FD-OCT measures the interference spectrally resolved. FD-OCT measures each A-scan without changing the path length and uses photons more efficiently than TD-OCT [6–8]. Both properties give FD-OCT an advantage over TD-OCT for imaging at NAs below 0.2, especially in retinal imaging. Using increasingly faster line cameras or fast wavelength swept laser an imaging speed of several million A-scans per second was achieved [9].

Another way to increase imaging speed is lateral parallelization. With the introduction of full field OCT (FF-OCT), imaging time of TD-OCT was reduced by several orders of magnitude. TD-FF-OCT uses a conventional broadband light source and achieves an axial resolution of one micrometer. Combined with high NA optics, axially and laterally subcellular resolution is achieved. The drastically reduced Rayleigh length at high NAs make focus tracking necessary for achieving optimal resolution and substantially revokes the advantages of FD-OCT. Parallelization of FD-OCT was demonstrated using ultra-high-speed cameras [10]. Due to these cameras and the complex optics, full field systems were so far only used successfully for ex vivo applications or in vivo applications at easily accessible tissues, such as the eye or skin. Internal organs can only be reached with endoscopes. Although rigid endoscopes were demonstrated with FF-OCT ten years ago [11], it seems to be difficult to design clinically useful systems, especially flexible endoscopes. Currently, endoscopically use of OCT is mostly based on scanning systems.

Fastest imaging rates with scanning FD-OCT are currently achieved with tunable lasers. These devices are called swept source (SS-) OCT or optical frequency-domain imaging (OFDI) [12–14]. A-scan rates of up to 20 MHz were demonstrated in experimental systems [9]. Due to the limited bandwidth of semiconductor optical amplifiers used in these light sources, axial resolution of SS-OCT is limited to above 5 µm [15]. This resolution is not sufficient to visualize cellular and intercellular morphology.

In contrast, an axial resolution comparable to FF-OCT can be achieved with spectral-domain (SD-)OCT. Compounding multiple superluminescence diodes (SLDs) with different emission wavelengths or using ultrabroad band lasers enhances the axial resolution to below 1 µm [16–18]. However, imaging speed is limited by the frame rate of the line scan camera which measures the spectra. Figure 1 provides an overview of resolution and voxel rates which were achieved with flying-spot SS-OCT and SD-OCT devices. It clearly shows the different domains of SD- and SS-OCT. Combined with high-NA focusing, only SD-OCT achieves a more or less isotropic subcellular resolution of nearly 1 µm³ [19,20], which is comparable to confocal or non-linear microscopy. Using cameras with line rates of 40,000 and 250,000 Hz an imaging speed of up to 250 Mega voxels per second was demonstrated [21].

 

Fig. 1. Imaging speed and axial resolution of experimental flying spot (fs-) OCT setups for research SD- (yellow) and SS-OCT systems (blue) cover different parameters regions. SD-OCT achieves higher axial resolutions, SS-OCT higher voxel rates, which makes it particularly suitable for in vivo volumetric imaging.

Download Full Size | PDF

For many ex vivo or in vivo imaging applications, speed is not very important. However, for endoscopic imaging, it becomes a very critical factor, especially at high resolution. On the one hand, the beam deflection in endoscopes is often based on micromechanical actuators, which are difficult to design with slow scanning speeds. On the other hand, due to tissue movements, it is challenging to keep the scanning position with micrometer accuracy, which is especially important when studying dynamic processes at micrometer resolution. Examples of highly dynamic processes in living organisms are mucociliary clearance, which has been studied intensively ex vivo and in vivo in recent years [20,22] and movements of intracellular structures [23,24]. Membranes, vesicles or other organelles are not static [25,26], and their motion randomly modulates the OCT signal. Apelian et al. showed with TD-FF-OCT [23] that a classification of the OCT signal fluctuations in a frequency range from 0.5 to 50 Hz is possible. The biological micromotions characterize different tissue features like nuclei, cells bodies, or connective tissue and constitute a new contrast mechanism termed dynamic OCT (dOCT). Recently, dOCT was also demonstrated with SD-OCT [27,28]. Instead of series of en face images, SD-OCT evaluates series of a few hundred B-scans over 1 to 2 s acquisition time to analyze the tissue micromovements [28]. SD-OCT is expected to be more suitable for endoscopic in vivo applications. In addition to a better compatibility with endoscopy, the parallel imaging of A-scans can use the average phase of structures at different depths to efficiently compensate axial tissue movements, for which dOCT is especially sensitive.

Here, we describe an FD-OCT with microscopic resolution that uses a new 600 kHz line-scan camera, which is 2.4 times faster than other available scanning SD-OCT systems. The potential of the camera for future application in OCT endoscopes is demonstrated by volumetric imaging of dynamic processes. First, we show that complex particle transport mediated by cilia on the surface of murine trachea can be recorded and analyzed in three dimensions. Furthermore, we demonstrate that the computation of dynamic OCT based on volume time series instead of time series of B-scans or en face planes is possible and can image tissue volumes with dynamic OCT in a few seconds.

2. Material and methods

2.1 mOCT setup

The setup was essentially described in detail in [28] and is shown in Fig. 2(a). Main difference is the use of a faster camera. In short, a fiber-based Michelson interferometer with a supercontinuum light source is used and lateral scanning was achieved by a pair of galvanometer mirrors (6210 H, Cambridge Technology, U.S.). A custom designed spectrometer (Thorlabs GmbH, Germany), which covers a spectral range from 550 to 950 nm, and a 10×/0.3 NA microscope objective (HCX APO L 10×/0.3 WUVI, Leica Microsystems, Germany) were used to achieve nearly 1 µm³ lateral resolution. Radiant flux in front of the objective was measured to be 60 mW. A novel CMOS line scan camera (xposure monochrome, AIT Austrian Institute of Technology, Austria) with 2016 active pixels enables a maximum A-scan rate of 600 kHz [29]. Pixel size and fill factor are 9 × 9 µm² and 90%, respectively. Quantum efficiency of the camera varies between 0.42 and 0.24 in the wavelength range from $463\; \textrm{nm}$ to $860\; \textrm{nm}$ (Table 1). The dynamic range, defined by the maximum attainable SNR, is usually limited by the full well capacity (FWC) of the spectrometer camera and the system noise. Maximum SNR is achieved at full modulation of the camera, i.e. reference and sample radiation each cover 25% FWC. For this camera with a full-well capacity (FWC) of 13,000 electrons and a camera noise of 25 e^- a dynamic range of 65 dB can be expected [6]. Accounting for a realistic bell-shaped spectrum, which contains 50% energy of a rectangular spectrum with the same peak intensity, reduces the dynamic range to 61 dB. The camera data were transmitted via an optical GigE interface (Generic Compatible 40G QSFP+ to 4 × 10G SFP+ Breakout Active Optical Cable, Fiberstore, U.S.) to the computer. A self-written C# library was used for inter-process communication between camera SDK and online OCT processing software. Real-time B-scan preview and C-scan acquisition were accomplished by a self-written LabVIEW (LabVIEW 2017, National Instruments, U.S.) program. Camera data were streamed to an M.2 solid state drive (Samsung 960 Pro M.2 2280, Korea), which is capable to receive and store camera data at a rate of 1.6 GB/s. For OCT reconstruction, data were processed by a MATLAB program using the parallel computing toolbox (MATLAB R2020b, The MathWorks, Inc., U.S.). Residual dispersion was numerically compensated using a 5^th order polynomial representation of the spectral phase error [16,30]. Polynomial coefficients were determined by optimizing image quality, which was quantified by the Shannon entropy [30]. For fast volumetric data acquisition, bidirectional scanning was used.

 

Fig. 2. (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: personal computer. (b) Evaluation of individual B-scans and volume scans of the dynamic mOCT (dmOCT) at time stepst₁, t₂,…, t_N for calculating the dynamic contrast in a frequency range from 0 to 25 Hz.

Download Full Size | PDF

2.2 System characterization

The dynamic range was determined by calculation of the SNR at the maximum non-saturated signal after placing a silver mirror in the sample arm. Measurements were done at different exposure times to study the influence of spectral noise of the supercontinuum light source. By changing the length of the reference arm, the SNR was determined at different positions in the axial field of view. The axial resolution was measured as the full width at half maximum (FWHM) of the linear OCT signal of a silver mirror. Lateral resolution was determined by scanning the edge of precision optical grating having 5 cy/mm (RR5-25-CG, Applied Image Inc., U.S.).

2.3 Animals

Fresh tissue samples were obtained from male C57BL/6 6- to 25-weeks old mice (Charles River Laboratories, Sulzfeld, Germany). All animals were kept with a 12-hour day-night cycle and food and drink ad libitum at the animal facility of the University of Lübeck. We confirm that all methods were performed in accordance with the relevant guidelines and regulations.

2.4 Cilia-driven transport

Cilia-driven transport was studied ex vivo in trachea tissue. The trachea was carefully cut open and tissue samples were immobilized, epithelium facing upwards, with needles in Sylgard polymer (Dow Corning, Wiesbaden, Germany), which coated the surface in a petri dish. Tissue samples were fully immersed in culture medium (HEPES Buffered Saline, pH 7.4). To visualize ciliary transport, 4.5 µm sized polymer particles (Dynabeads – CELLection Biotin-Binder Kit, Dynal Biotech ASA, Norway) were added to the culture medium. Ciliary beating and particle motion were imaged at 12 volumes per second with the focus adjusted to the surface of the ciliated epithelium. Volumes were recorded with 1024 voxels in the axial direction (z), 500 voxels on the fast axis (x) and 101 voxels along the slow axis (y) and afterwards cropped to 334 × 100 × 1024 (xyz) voxels to remove scanning artifacts at the beginning and end of the fast axis. Shifts between adjacent B-scans due to bidirectional scanning were corrected manually. No further motion correction was needed. For a visualization with a correct aspect ratio, y-data were interpolated to twice the voxel number. The imaged volume had a size of 250 µm x 150 µm x 840 µm (xyz). Volumetric data was analyzed and visualized with Imaris 9.2 software (Oxford Instruments, Abingdon, United Kingdom) using the tracking toolbox. For quantification of transport 25 particles were segmented, tracked and their speed and acceleration were automatically calculated.

2.5 Volumetric dynamic OCT

To increase the imaging speed of dynamic contrast OCT, 10 volumes of 500 × 50 × 1024 voxels (xyz) were sequentially acquired with a 24 Hz volume rate. Shifts between adjacent dynamic B-scans due to bidirectional scanning were corrected with a deinterlacing algorithm (Matlab Computer Vision Toolbox, Mathworks, U.S.). The evaluation of the recorded data for dynamic OCT was similar as described in [28]. The main difference in this study is the measurement of volume series instead of B-scan series at a rate of 24 Hz for calculating the dynamic OCT volumes [Fig. 2(b)]. After Fourier transform of the amplitude of the OCT signal in each voxel of the 100 consecutively recorded volumes, the resulting spectra were divided into three frequency bands and the amplitude in each band was integrated. Logarithmizing and scaling the amplitude of each channel between 0 and 1 resulted in the RGB color values. The blue color channel represents slow signal fluctuations frequencies (0–0.5 Hz), the green medium frequencies (0.5–10 Hz) and the red channel fast fluctuations (10–12 Hz). At 10 adjacent sites, sequences of 100 OCT volumes were consecutively recorded and stitched together after calculating the dynamic contrast. In total, we acquired a volume of 300 × 300 × 840 µm³ with 500 × 500 × 1024 voxels within 41.7 seconds. Processing took approximately 10 minutes. For a demonstration of volume based dynamic contrast OCT, we used freshly excised liver tissue of C57BL/6 mice, placed in Ringer’s solution (HEPES-buffered). Dynamic contrast OCT imaging of murine liver is well characterized and allows to visualize cells and subcellular structures of hepatocytes with good contrast.

3. Results

3.1 System characterization

SNR and roll-off are crucial parameters that determine the image quality. At full A-scan rate (600,000 A-scans/s, 1.1 µs exposure time), the maximum achievable SNR ranged from 51 dB at 80 µm image depth to 38 dB at a depth of 800 µm near the Nyquist sampling frequency [Fig. 3(a)]. The decrease of SNR by 13 dB over the entire imaging depth is significantly larger than the 7 dB roll-off which is expected due to averaging over the pixel area [31]. Additional losses result from the limited spectrometer resolution and pixel crosstalk of the camera. In contrast to a photon-noise-limited OCT device, in which the noise level is flat, a 5 dB increased noise is observed at small imaging depth, which is caused by pulse-to-pulse modulations of the spectra of the supercontinuum light sources. Also, the dynamic range is considerably lower than the expected 61 dB. The additional noise caused by the supercontinuum light source is most prominent at small depth. It increases at higher A-scan rates, since fewer spectra contribute to the measured interference pattern. For example, at 600 kHz A-scan rate, only 352 spectra are averaged, whereby at 50 kHz more than 4,000 spectra of the pulsed light source contribute to one A-scan. Averaging more spectra (i.e., decreasing the A-scan rate) increases SNR, as shown in Fig. 3(b). Reducing the A-scan rate from 600 kHz to 50 kHz improves the SNR by 3-4 dB.

 

Fig. 3. (a) A-scans of a reflecting surface at 600,000 A-scans/s and 1.1 µs exposure time. The mirror was adjusted to maximum non-saturated OCT-signal. The length of the reference arm was changed to measure maximum SNR for different imaging depths. The roll-off was 13 dB over the full depth range of 840 nm. Artifacts at higher imaging depth are caused by undersampling in parts of the spectrum which is considerably chirped due to the broad spectral range. (b) Measured maximum SNR at 10% of the measurement range for different camera exposure times. SNR drops with decreasing exposure time to below 52 dB at 1.1 $\mathrm{\mu}$s (600 kHz A-scan rate). An exponential function with constant offset was fitted to the experimental data to guide the eye (dashed curve).

Download Full Size | PDF

Table 1. Comparison between the specifications of the 600 kHz camera used here with the so far fastest camera Teledyne Octoplus used for high resolution OCT imaging.

View Table

Due to their high power, broad emission spectrum and compact and robust construction, supercontinuum light source are frequently used for OCT with microscopic resolution. However, while increasing imaging speed, spectral noise increasingly sacrifices image quality. Increasing the repetition rate of the supercontinuum laser or using Ti:Saphir femtosecond lasers [32] are ways to revert this effect.

After spectral shaping, apodization with a Hann window and numerical dispersion correction, an axial resolution of 1.8 µm full width at half maximum (FWHM) was experimentally determined [Fig. 4(a)]. Since the spectrometer is not wavelength calibrated, the achievable resolution was determined from the maximum measurement depth, which is directly related to the spectral difference between adjacent pixels, and the number of spectrometer pixels [33]. After full dispersion correction, the FWHM only depends on the apodization function. For the Hann window exactly two pixels are expected [33]. Therefore, with an experimentally determined full imaging range from $- 840\,\mathrm{\mu}\textrm{m}$ to $+ 840\,\mathrm{\mu}\textrm{m}\; $ and 2016 camera pixels the theoretical limit for the axial resolution is $1.7\,\mathrm{\mu}\textrm{m}$. From the measurement of the edge function, the lateral point spread function (PSF) was calculated by differentiation. A FWHM of 1.1 µm was determined [Fig. 4(b)]. For confocal imaging, the FWHM of the diffraction limited point spread function is given by 0.37 λ/NA [34], which is 0.93 µm for the NA of 0.3 and the center wavelength $\lambda = 750 {\textrm{ nm}}$ used here. The confocal parameter for a 0.3 NA is approximately 20 µm.

 

Fig. 4. (a) Axial resolution was determined from the full width at half maximum (FWHM) of the linear OCT signal of the reflection of the silver mirror. (b) Point spread function (PSF) was calculated by differentiation of the edge-spread function measured along fast scanning axis. A FWHM of 1.1 μm was measured for the PSF.

Download Full Size | PDF

3.2 4D ex vivo imaging of particle transport

The unique combination of high resolution and high imaging speed was used to image cilia-driven transport in airways. Collective motion of the epithelial cilia propels mucus and fluid on top of the cells. This active process is crucial for airway cleaning and impaired clearing function is a hallmark of several diseases. The fluid is not necessarily transported as a laminar homogeneous flow over the tissue surface. Varying thickness, speed and directions are observed [22]. A non-uniform distribution of ciliated cells may introduce complex motion already on a millimeter scale. Breathing, coughing and ciliary malefunction additionally influence the clearing process.

To demonstrate 3-dimensional visualization of ciliary driven transport, microparticles were added to a short time tissue culture of murine airway. Particle movement was imaged in time sequences containing 119 volumes over 10 seconds [Fig. 5, Visualization 1]. Particles are readily visible in the fluid above the tissue and directly on the surface. In addition, the maximum intensity projection of the volumes shows the borders of individual epithelial cells. Cilia were not directly visible in the volume representation, but ciliary beating causes fast signal fluctuations in the extracted B-scans (see Visualization 1). The SNR was sufficient for full thickness imaging of the murine airway [Fig. 5(a)]. Epithelial cell layer on the inside of the trachea, glands and vessels were visible. Particles appear with high contrast as small spheres in the focus plane and disk-like structures outside the focus [Fig. 5(a) and (b)]. The influence of defocus is directly visible as the particles move towards the epithelium [Fig. 5(a) and Visualization 1]. Particles move in a diagonal direction from the upper right to the lower left in varying distances up to 150 µm from the surface. An accumulation of particles [Fig. 5(a), arrow 5] blocks transport in the upper left corner of the volume. Particle speed obviously depends on their distance from the epithelium. More remote to the surface (> 100 µm) particles move only slowly, while, in general, near the surface lateral velocities are highest. In addition, also stationary particles were observed near the tissue surface. Reasons could be either a lack of ciliary motion of the epithelial cells or by an excessive accumulation of particles, which adhered to the epithelial layer.

 

Fig. 5. (a) Average of five adjacent B-scans from one recorded volume. Different tracheal tissue structures like gland duct (1), epithelium, and subepithelial tissue are visible. Single particles appear as small points (2) near the focal plane and above the focal plane laterally extended due to the defocus (3). A larger number of deposited particles (4) are visible at the left side of the B-scan. The particles shadow the lower tissue structures (5). The outer surface of the trachea is visible at the bottom of the image. (b) Maximum intensity projection of one mOCT volume. The overlay shows the complex 3-dimensional motion for 25 particles, which was calculated from the sequence of imaged volumes. Below the trajectories, individual cells of the epithelial cell layer are visible with clear demarcation for their borders (6). (c) Side view of the tracked particles showing color-coded particle speed. Particles are accelerated near the epithelium, rise into an upward motion and fall back to the epithelium (7) where they are accelerated again by ciliary action. The size of the cropped volume was 250 × 150 × 270 µm³ (xyz). Scalebar (white): 50 µm. See Visualization 1.

Download Full Size | PDF

For quantitative evaluation of individual particle motion, automated particle tracking was used. In total, 25 particles were evaluated, and their pathways are displayed color-coded in Fig. 5(b) and (c). The maximum particle speed in the volume was 91 μm/s with an average speed of 26 µm/s and a standard deviation 13 µm/s.

For each particle, the speed was not constant, but rapid acceleration to maximum speed near the epithelium or slowing down and even stopping in larger distance of the tissue surface were observed. The correlation between speed and distance from the particles to the ciliary cells is best seen in Fig. 5(c). Our model system for ciliated transport shows an astonishing heterogeneity of motion speed and motion direction. Thanks to the high spatial and temporal resolution, mOCT is able to follow the particle transport with high accuracy. This would not be possible using only a series of B-scans. Particles would move in and out of the imaged plane [Fig. 5(a)] and precise characterization of their trajectories would be very difficult. In contrast, the sequence of 3-dimensional volumes shows the complete motion trajectory of the particle together with a full thickness volumetric image of trachea [Fig. 5(b)]. Investigations of mucus transport would profit from video-rate volumetric imaging. Transport could be measured in all three dimensions and global motion due to heartbeat and breathing, which disturb a quantitative evaluation of the images, could be corrected in all three dimensions.

3.3 High speed volumetric dynamic OCT

High imaging speed is also crucial for implementing dynamic contrast in mOCT. Currently, analysis of the signal fluctuations for dynamic OCT imaging is based on en face images [23,24] or B-scan sequences [27,28]. For the interesting frequency range from 0.5 Hz to 25 Hz, A-scan rates of 25 to 50 kHz are sufficient, and a cross-sectional dynamic B-Scan can be obtained in a few seconds. However, stacking these dynamic B-scans to volumes, the total acquisition time results from the sum of the acquisition times of the individual dynamic B-scans. This can add up to a quarter of an hour. Increasing the A-scan rate to 600 kHz provides the opportunity to calculate the dynamic contrast on volume series instead of B-scans. Figure 6 shows a dynamic contrast mOCT volume of liver tissue, which was calculated from 10 stitched subvolumes with 500 × 50x 1024 voxel (xyz). Total acquisition time was only 41.6 seconds. After cropping in z-direction for better representation, the size of the stitched dynamic volume is 300 × 300 × 160 µm³ (xyz). Using the dynamic contrast, individual hepatocytes are visible with high quality. In the visualization of the volume [Fig. 6(a)] and in cross-sectional [Fig. 6(b)] images, a layer of connective tissue (Glisson's capsule) surrounding the liver can be identified colored in blue. In the cross-sectional [Fig. 6(b)] and en face [Fig. 6(c)] image cell borders can be clearly distinguished from the interstitial volumes. These cellular structures are barely visible in the averaged OCT image [Fig. 6(d)]. The 10 µm to 30 µm sized hepatocytes, which take up about 80 percent of the liver volume, are highly metabolically active and contain numerous cell organelles. The cell nucleus can be identified in the center as a darker spot. As expected, outside of the focal region in 30 µm depth, cells become blurred due to the influence of the defocus [Fig. 6(b) and Visualization 2]. This is best seen in the B-scan in Fig. 6(b). Right, the cells directly under the tissue surface are in focus. They are visible with excellent contrast and resolution. In contrast, left, the upper cell layer, which is 50 µm above focus appears blurred with considerably lower contrast.

 

Fig. 6. (a) Volume based dynamic OCT imaging of murine liver. The size of the measured volume was 300 × 300 × 800 µm³ (xyz). Here the volume is cropped in z for better representation [Visualization 2]. Scanning artifacts are visible at the back of the volume (*). (b) B-scan extracted from the dynamic OCT volume. The B-scan was reconstructed perpendicular to the fast scanning axis. An artifact due to tissue motion is visible by a color change of the tissue surface (**), (c). En face dynamic OCT at 60 µm depth below the surface. Hepatocytes become visible due to dynamic contrast and can be identified by cell plasma (green) and cell nuclei (red). The layer of connective tissue (Glisson's capsule) is blue (d) In the corresponding averaged OCT image cellular structures are barely visible.

Download Full Size | PDF

Artifacts due to the fast scanning and tissue motion are minimal due to the favorable ex vivo imaging conditions. Only in the 3D representation of the volume a slight artifact is visible at the upper right rim (*). In the reconstructed B-scan the color of the tissue surface differs in a localized area in the left corner of Fig. 6(b). The B-scan was reconstructed perpendicular to the fast-scanning axis, i.e. Figure 6(b) cuts through all 10 subvolumes. During the measurement of the volume sequence, the leftmost volume moved slightly downwards, as can be seen in sequence of the raw OCT images [Visualization 3]. The series of en face images corresponding to Fig. 6(c) and (d) does not show any macroscopic motion [Visualization 4]. These sequences of OCT images demonstrate that, in general, in the ex vivo imaging situation global tissue motion does not interfere with the local micromotion which causes a visible speckle fluctuation. Therefore, advanced motion correction was not necessary.

4. Discussion

Since the development of OCT in the early nineties, imaging speed has been steadily increased [35]. The benefits of faster OCT imaging were often controversially discussed, since in general higher imaging speed is related to a more complex instrumentation and reduced SNR and sensitivity [36]. However, it always became apparent that higher imaging speed led to new opportunities. While in the beginning speed was needed to reduce artifacts of the inherent movements of the patient during the measurement, later, generating more information in less time became important. This development was seen in ophthalmology and dermatology. B-scan imaging was replaced by volumetric imaging for the diagnosis with increasingly larger imaged areas. This was made possible by a change in technology from TD-OCT to FD-OCT. Over the years, the speed was continuously increased by technological advances of line scan cameras, tunable laser sources, DA-converters and processing hardware. Besides increasing the imaged tissue area, the increased speed allowed the visualization of dynamic processes, such as blood flow, which is an important diagnostic parameter for various pathologies in ophthalmology and dermatology [37–40]. Visualizing capillary structures by OCT angiography was only possible by volumetric imaging, because the vessel structure is not seen in B or C-scan alone [41]. Current high-speed OCT systems used in research, achieve a few MHz A-scan rate [15]. Increasing the A-scan rate beyond this range is not trivial as many components reach their technological limits [36].

An alternative way to increases imaging speed is parallelization [42,43]. Instead of scanning a focused beam, a line (line field or LF-OCT) or an area (full-field or FF-OCT) is imaged in parallel. Retinal imaging with more than 40 MHz A-scan rate was demonstrated using FD-FF-OCT [44,45]. By evaluating phase changes in series of OCT volumes it was possible to observe nanometer length changes of human photoreceptors, during stimulation by light, though the axial resolution was only 14 µm [44]. Thereby, not the time constant of the photoreceptor response itself was the reason for the fast-imaging speed, but the necessity to compensate motion in all three dimensions and the correct sampling of the phase changes. Volumetric imaging is crucial for complete compensation of global tissue motion.

A lateral parallelization of the image acquisition has also led to a renaissance of TD-OCT [46]. With the TD-FF-OCT as well as the TD-LF-OCT, the acquisition speed can compete with FD-OCT and for certain applications parallelized TD-OCT seem to be the better choice. Since TD-FF-OCT can be realized with inexpensive components, it is well suited for point-of-care and home-care applications [47–53]. Another application well suited for TD-FF-OCT is optical coherence microscopy. Since in TD-OCT the axial information is acquired sequentially, the focus position can be adjusted to any depth plane. Spatially incoherent light sources can be used which offer a broad spectral range at low complexity and price and additionally suppress multiply scattered light [54]. An axial resolution of one micrometer is achieved with simple LED light sources. With the introduction of dynamic OCT, TD-FF-OCT could introduce motion contrast not only for vessels but for every living cell [23]. Dynamic OCT provides tissue specific contrast and can also overcomes one of the major drawbacks of OCT, the speckle noise, because similar to OCT angiography [41], it effectively averages over uncorrelated speckle patterns. Dynamic OCT with microscopic resolution may in the future replace conventional histology in certain applications [1,55].

However, parallelization of OCT imaging with TD-FF-OCT holds also some drawbacks. On the one hand, the confocal aperture is missing, which means that multiply scattered photons form a strong background signal on the detector, when imaging in scattering tissues. Though they do not contribute to the OCT signal, they increase detection noise. On the other hand, the endoscopic implementation for flexible endoscopes is currently not possible. The development of rigid endoscopes is very challenging, since special cameras are needed, which are quite bulky and energy consuming [11].

For endoscopic imaging, scanned FD-OCT is therefore currently a technology of choice. Here, only spectrometer-based systems support the 1-2 µm resolution, that is needed for imaging subcellular structures. The necessary spectral bandwidth of a few hundred nanometers is achieved by combining multiple SLDs, using supercontinuum light sources [56,57] or ultra-broadband femtosecond lasers [58]. Tunable light sources suitable for SS-OCT are currently limited to the spectral width of their semiconductor gain media.

In this study, we demonstrate a spectrometer-based OCT that is 2.4 times faster than the current state of the art and delivers sufficient image quality for 4D-OCT imaging with cellular resolution. By using a novel 600 kHz line-scan camera, volumes with a spatial resolution in the focus region of 1.1 µm lateral and 1.8 µm axial were acquired at 0.6 Giga voxels/s over several seconds, faster than with any other flying spot FD-OCT device at that resolution. Despite the reduced dynamic range due to a lower full-well capacity and an increase of spectral noise of the supercontinuum light source, we were able to image simultaneously microparticles transport in 3D and the full thickness of the murine trachea wall tissue at 12 Hz volume rate. For dynamic OCT of liver tissue, we reached even 24 Hz.

Volumetric imaging at that speed and resolution allows the correlation of transport with structures and function of the airway tissue at a reasonable field-of-view. The model system for ciliated transport presented here does not fully represent the in vivo situation. Instead of coverage with a thin mucus layer, the tissue is fully immersed in fluid. In addition, there is no contribution of air flow caused by breathing or coughing to the transport. However, even without these effects our model shows astonishingly complex motion patterns, that can only be measured in all details with the resolution and speed of the new OCT device. Imaging mucus transport in vivo by transtracheal mOCT was demonstrated earlier in mice [20,22] using series of B-scans. However, evaluation of the mucus transport was severely affected by motion perpendicular to the B-scan.

The calculation of dynamic OCT has so far only been demonstrated on B-scans in FD-OCT systems or on en face planes in TD-FF-OCT. In our experiments, a volume of 0.07 mm³ was measured in less than a minute. The same volume size would have resulted in more than 10 minutes of imaging with B-scan or en face plane based dynamic contrast.

Individual cellular components such as the hepatocytes and their cell borders were visualized with a similar quality as in our previous work which used a 100 kHz A-scan rate and evaluated the dynamic signal on a B-scan basis [28].

Comparing the images in Fig. 6 with the sequence of one hundred volumes (Visualization 3 and Visualization 4), shows the strength of dynamic contrast imaging. Firstly, evaluating a large number of volumes increases SNR by an expected factor of ten, secondly the tissue fluctuations average out the speckle. Both is visible in the increased image quality of Fig. 6(d) compared to the raw images. Thirdly, the spectral distribution of the speckle fluctuations provides a cell and tissue specific contrast. Consequently, fast measurement of multiple volumes with low SNR provides more information than a slow measurement of one volume with high quality unless the volume sequence is not corrupted by sample motion. This is a strong argument for faster OCT imaging.

The work presented here is an important step towards endoscopic OCT imaging at subcellular resolution. In addition to the increased imaging speed, which provides more information in less time, registration of individual volumes allows compensation of tissue motion in all three orthogonal directions, which will be the main problem for in vivo application of dynamic OCT. Motion correction of OCT volumes was successfully used for functional retinal imaging [59]. For phase sensitive measurement in different retinal layers, volumes of 640 × 368 × 256 voxels were recorded every 125 ms. Translational and rotational motion caused by the pulsation of the choroid and larger retinal vessels was sufficiently corrected by a rigid registration and realignment of subvolumes to achieve a phase sensitivity below 10 nm. In contrast, series of B-scans provide not enough information on perpendicular motion for a successful correction. Here only motion parallel to the plane of the B-scan could be corrected. Although tissue motion in endoscopic OCT will differ from motion during retinal imaging and dOCT at 1 µm resolution reacts different than functional retinal imaging at 10 µm resolution, we anticipate that recording volumes at 25 Hz or faster is the key to a successful motion correction in in vivo dynamic OCT imaging.

The NA used here leads to a confocal parameter of 20 µm. Even though this is only a small part of our measurement depth we think it is a good compromise between the axial resolution on the one hand and the signal roll-off caused by the confocal detection and the defocus on the other hand. Unlike other optical microscopy techniques, the out-of-focus signal can be used due to the high dynamic range of OCT. It carries significant information on coarser structures and the position of the imaged field in the tissue and is especially helpful for interactive imaging in patients. E.g., in the trachea it was possible to visualize individual epithelial cells and microparticles located a hundred micrometers above the tissue surface. Especially the position of isolated particles can be determined quite accurately over the entire image field of view by calculating the center of gravity or fitting the point spread function (PSF). Cellular structures of the liver could also be detected over a certain depth beyond the focal region in dynamic OCT, even if they become blurred with increasing depth.

A fascinating perspective for microscopic FD-OCT is to apply refocusing algorithms in post-processing. Further increasing the A-scan speed into the MHz range may provide sufficient phase stability, to correct individual volumes for aberrations and defocus by manipulating the lateral phase in the Fourier space of the en face images [60].

5. Conclusion

By using a novel 600 kHz line-scan camera, we demonstrated fast 4D OCT imaging with cellular resolution at an imaging speed of up to 24 volumes per second. Despite the reduced dynamic range due to a lower full-well capacity, an increase of spectral noise of the supercontinuum light source and the reduced lateral resolution outside a comparable thin focus layer, we were able to image simultaneously microparticles transport and the full thickness of the murine trachea wall at 12 Hz volume rate and with up to 24 Hz for the computation of dynamic OCT of liver tissue. Volumes were acquired, faster than with any other flying spot FD-OCT device at that resolution. The maximum imaging time was only limited by the storage capacity of the computer. With improved hardware real-time visualization of 4D OCT volumes would be feasible [61]. Quantitative parameters of mucosal transport, which are essential to understand the involved biological processes, like velocity or acceleration can be measured. We also could demonstrate dynamic contrast OCT imaging with subcellular resolution, based on the evaluation of volumes rather than B-scan. This new concept for dynamic OCT allows to fully exploit the increased A-scan rate and is especially attractive for future in vivo applications. Fast volumetric mOCT may also enable high-resolution volumetric OCT endoscopy with cell-specific contrast [62]. Pathologies often manifest in changing cell morphology, cell density or cellular distribution in the tissue. Fast 3-dimensional imaging and a marker-free optical contrast based on dynamic cell processes could be an important tool for clinical diagnosis. Although we demonstrated the ability to study real-time dynamic processes in three dimensions with cellular resolution, there is still a need for even higher camera speed in SD-OCT. Exploiting advances in data processing and increasing volume size will be crucial for applying dynamic contrast OCT for in vivo imaging.