1.

Introduction

Optical coherence tomography (OCT) is a recently developed optical technique which has the ability to noninvasively visualize three-dimensional (3-D) structural information of a biological sample on the micron scale.¹ Generally, the method is based on a Michelson-type interferometric set-up which utilizes a broadband light source to coherence-gate the light returning from a sample and hence provide high axial resolution. OCT has been utilized in many biological applications including ophthalmology, cardiology, oncology, neurology, dermatology, etc. Full-field OCT is an alternative coherence domain imaging modality which, unlike conventional OCT, produces tomographic images in the en face orientation (i.e., orthogonal to the optical axis).² This modality utilizes a charge-coupled device (CCD) or complementary metal-oxide-semiconductor (CMOS) camera to capture the full-field at a particular depth, negating the necessity for lateral scanning.³ The axial resolution is set by the bandwidth of the light source used (as is the case with all modalities of OCT), and the transverse resolution is set by the choice of optical components on the reference and sample arms. This means that transverse resolutions similar to that of conventional microscopy ($\sim 1\mu \text{m}$) are easily achievable, which has led to numerous studies on cellular-level tomographical imaging in both ex vivo^4,5 and in vivo^6,7 imaging applications.

Various methods of image reconstruction have been reported with, generally, a trade-off of speed of image acquisition and system sensitivity.⁸ In this work, the aim is to utilize spatial and temporal correlations in successive frames in order to extract flow information. For this reason, speed of acquisition is a priority and hence a two-step method is employed in conjunction with the mathematical Hilbert transform to reconstruct each en face image.⁹

The correlation mapping optical coherence tomography (cmOCT) processing algorithm was developed by our group^10,11 and utilizes the time-varying speckle effect produced from moving scatterers to provide contrast against static tissue. The technique has previously been applied to images acquired with frequency-domain OCT systems and hence in a cross-sectional geometry.^10–12 To our knowledge, this is the first instance of the application of cmOCT to the full-field modality of OCT. In the present study, full-field cmOCT is utilized to image a phantom in order to illustrate its ability to separate static structures from dynamic structures moving under Brownian motion. In addition, in vivo blood flow imaging of a Xenopus laevis tadpole is demonstrated.

2.

Experimental Configuration and System Performance

A diagram of the experimental set-up is shown in Fig. 1(a). The light source for the set-up consists of a near-infrared super luminescent diode centered at 840 nm with a bandwidth of 50 nm (Superlum S-840B-I-20, Carrigtwohill, Co. Cork, Ireland). This light source has an output power of 15 mW with a double-Gaussian spectral profile. The incident light is input into the interferometer via a single mode optical fibre and is collimated into a beam with diameter of $\sim 10\mathrm{mm}$ using a doublet of focal length 50 mm. The purpose of collimation was to ensure uniform illumination of the sample. The beam splitter has a splitting ratio of $50:50$ which splits the amplitude into the reference and sample signals. Two identical variable neutral density filters are placed in the paths of each of the beams in order to provide dispersion compensation and control the visibility of the interference pattern. Two identical microscope objectives [$4\times$ numerical aperture $(\mathrm{NA})=0.1$ Olympus, UK, working $\text{distance}=18.5\mathrm{mm}$] are placed on each arm to focus the beam onto the reference mirror and sample. All components are coated with the appropriate anti-reflection coating in order to reduce reflections coming from within the system.

Fig. 1

(a) Schematic representation of the experimental set up. The light source is a super-luminescent diode. PC, polarization control; BS, beam splitter ($50:50$); MO, microscope objective ($4\times \mathrm{NA}=0.1$); L1 and L2, achromatic doublets (50 and 200 mm focal length, respectively); M, mirror for upright sample arm; ND, variable neutral density filters (optical density 0 to 2); S, sample and stage; PZT, piezoelectric actuator (oscillation of 10 Hz, peak-to-peak $\text{amplitude}={\lambda }_0/4=0.21\mu \text{m}$). (b) The axial response of a single pixel of each en face image measured by scanning a mirror in the axial direction. The axial resolution of the $\text{system}=8\mu \text{m}$.

The process of reconstructing the interferogram images involves use of the mathematical Hilbert transform, details of which can be found in Na et al.⁹ A phase shift of $\pi$ was provided by a piezoelectric transducer (PZT) which oscillates the reference mirror. The PZT system employed is a closed-loop system using a Physik Instrumente, Karlsruhe, Germany, piezo actuator (P-841.1) and amplifier (E-665). The camera model used is a Hamamatsu, Hertfordshire, UK, Orca Flash 2.8 (C11440), which incorporates a CMOS sensor consisting of $1920\times 1440$ square pixels of size 3.63 μm and a 12-bit analogue-to-digital converter. The camera is triggered externally with a maximum frame-rate of 20 Hz. (Higher frame rates are possible at the expense of the amount of pixels utilized). Since a two-step image reconstruction algorithm is utilized, the PZT must operate at half the frequency of the camera meaning a maximum OCT frame-rate of 10 Hz. The signals to drive the PZT and the camera are provided by a National Instruments, Berkshire, UK, data acquisition system, which is controlled by LabVIEW allowing accurate synchronization.

The detection sensitivity and dynamic range of the system was measured experimentally by scanning a mirror through the focal plane of the sample microscope objective. A neutral density filter of optical density, $D=1.1$, was placed in the path of the sample beam. The measured fringe intensity of a single pixel was plotted as a function of position and the dynamic range and sensitivity were measured from this graph. The detection sensitivity represents the noise floor of the system and was measured as the intensity of a single pixel, averaged over 100 points, at a position far from the zero-position and was found to be 61 dB. It is important to note that this test was done without averaging multiple frames at each position. The dynamic range was measured as the ratio between the highest and lowest signals measureable by the camera and was found to be 39 dB.

The axial resolution of the system, $\mathrm{\Delta }z$, is equal to the “round-trip” coherence length of the light source. This was measured experimentally by using a mirror as the sample and scanning in the axial direction using a step-size of 0.1 μm. The signal from a single pixel of each en face (demodulated) image was plotted as a function of position and the full-width at half maximum was measured to be 8 μm [Fig. 1(b)]. The transverse resolution is determined by the NA of the microscope objectives and was measured experimentally by imaging a standard target (1951 U.S. Air Force). The smallest resolvable feature at this magnification was found to be Group 6, Element 6 (i.e., 4.39 μm). The overall magnification of the system was measured to be 5.65 resulting in a field of view of $1260\times 945\mu \text{m}$ covered by the $1920\times 1440\text{pixels}$ of the camera.

2.1.

Correlation Mapping

The correlation mapping algorithm is applied to pairs of consecutive en face images. In order to suppress the decorrelation of noisy background pixels between consecutive frames, each image is first subject to a threshold algorithm which equates any pixel below the threshold value to zero. Finding the correlation map between frame A and frame B involves calculation of the two-dimensional (2-D) cross-correlation of a grid of size $M\times N$ from frame A with the corresponding grid in frame B using the equation:

Eq. (1)

$$\mathrm{cmOCT}(x,y)=\sum _{p=0}^{M-1}\sum _{q=0}^{N-1}\frac{\lfloor I_{\mathrm{A}}(x+p,y+q)-{\overline{I}}_{\mathrm{A}}(x,y)\rfloor \lfloor I_{\mathrm{B}}(x+p,y+q)-{\overline{I}}_{\mathrm{B}}(x,y)\rfloor }{\sqrt{{[I_{\mathrm{A}}(x+p,y+q)-{\overline{I}}_{\mathrm{A}}(x,y)]}^2{[I_{\mathrm{B}}(x+p,y+q)-{\overline{I}}_{\mathrm{B}}(x,y)]}^2}},$$

where $\overline{I}$ is the grid’s mean value. The grid is stepped across the whole image resulting in a 2-D cross-correlation map having values on the range of 1 to $-1$, representing correlated to anti-correlated. This process is shown diagrammatically in Fig. 2.

Fig. 2

Flow chart of the procedure of applying the correlation mapping algorithm to pairs of consecutive en face images. The two-dimensional cross-correlation is found between a grid (illustrated as $3\times 3$) in frame A and frame B and repeated by stepping the grid across the whole range of the image. The process is repeated for all consecutive pairs of en face images.

3.

Results and Discussions

A glass capillary with an internal diameter of 300 μm (outer $\text{diameter}=400\mu \text{m}$) was filled with 5% intralipid solution and embedded into a synthetic clay [Blu-Tac (R), Bostic Wauwatosa, Wisconsin] in order to simulate the background heterogeneity of tissue. The solution was allowed to move under Brownian motion and the beam was focused to an approximate depth of 80 μm into the capillary. A total of 100 images were acquired at a frame rate of 10 Hz over a time period of 10 s with an exposure time of 10 ms. The exposure time was maximized for each scan so that the maximum intensity was just below the saturation point of the camera-sensor. Figure 3(a) shows the microscope image (taken with the reference arm closed) of the glass capillary. The saturated region in the center is caused by the Fresnel reflection from the top surface of the capillary. Figure 3(b) shows a demodulated en face image of the phantom. The reflected intensity from the clay is approximately equal to that of the solution; however, over time, the time-varying speckle is clearly visible. The correlation mapping algorithm is applied to the en face images and the corresponding cmOCT image is shown in Fig. 3(c). In this image, it can clearly be seen that the light reflections from the static region [marked as “S” in Fig. 3(b)] remains correlated. The dynamic scatterers [region marked as “D” in Fig. 3(b)] cause a reduction in the correlation between consecutive frames and hence a significant contrast between the two regions is observed. Movie clips are provided for Figs. 3(a)–3(c).

Fig. 3

(a) Microscope image of a phantom consisting of a glass capillary containing 5% intralipid solution. The internal diameter (marked “ID”) of the capillary is 300 μm. (b) An en face image of the capillary. The intralipid solution moves under Brownian motion (region marked “D”) and the static regions are marked “S.” (c) The corresponding cmOCT image of the phantom. Videos 1 (MOV, 10.0 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.12.126008.1], 2 (MOV, 10.1 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.12.126008.2], and 3 (MOV, 9.99 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.12.126008.3] corresponding to (a), (b), and (c) are also provided.

In vivo imaging of a Xenopus laevis tadpole was also performed with the tail region placed on a glass slide while the head of the tadpole was kept in water (water anaesthetic solution). The sample beam was focussed on the tail region to a depth of $\sim 100\mu \text{m}$. A series of images were acquired at a rate of 10 Hz over a period of ten seconds giving a total of 100 frames with an exposure time of 15 ms. A microscope image (image taken with the reference arm closed) is shown in Fig. 4(a).

Fig. 4

(a) Microscope image of the tail region of a stage 42 (6 days postfertilization) Xenopus tadpole. (b) The mean en face image showing the dorsal longitudinal anastomosing vessel (DLAV), the dorsal aorta (DA) and the posterior cardinal vein (PCV). (c) The corresponding cmOCT image. Videos 4 (MOV, 10.0 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.12.126008.4], 5 (MOV, 10.0 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.12.126008.5], and 6 (MOV, 9.88 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.12.126008.6] corresponding to (a), (b), and (c) are also provided. (d) Three-dimensional (3-D) scan of the tail region reconstructed from en face images. (e) The corresponding 3-D scan reconstructed from cmOCT images. (f) Y–Z slice of the en face volume taken at the position of the red plane. (g) Y–Z slice of the cmOCT volume taken at the position of the red plane.

The mean en face image is shown in Fig. 4(b) clearly showing the dorsal longitudinal anastomosing vessel (DLAV), the dorsal aorta (DA), and the posterior cardinal vein (PCV).¹³ Again, over time, the time-varying speckle caused by the blood moving through the blood vessels is clearly observable. An example of a single cmOCT image is shown in Fig. 4(c). The PCV and the DA are clearly distinguished. Movie clips are provided for Figs. 4(a)–4(c). Figure 4(d) shows a 3-D rendering at a similar location of the tail region. The volume is reconstructed from of a series of en face images taken at 2-μm intervals over a depth of 100 μm. Figure 4(e) shows the corresponding volume reconstructed from the cmOCT images. Figure 4(f) shows a Y–Z slice of the en face volume taken at the position marked by the red plane. Similarly, Fig. 4(g) shows a Y–Z slice of the cmOCT volume taken at the position marked by the red plane.

4.

Summary

This work has illustrated the feasibility of applying the cmOCT algorithm to the full-field modality of OCT for the purpose of utilizing the time-varying speckle effect as a contrast mechanism between static and dynamic scatterers. A tissue-mimicking phantom was created and it was shown that Brownian motion in intralipid solution caused a reduction in correlation of the signal-intensity in successive frames at 10 Hz. Hence, a significant contrast between dynamic and static regions was provided, highlighting the system’s sensitivity to flow. In vivo imaging of the microvasculature of a tadpole at development-stage 46 was performed showing good contrast between regions of blood flow and static tissue.