1. Introduction

Spectral Optical Coherence Tomography (SOCT) also called spectral domain OCT is a spectrometer-based OCT modality, which enables detecting intensity of interfering light beams as a function of optical frequency. This method is well developed for cross-sectional imaging of morphological features of weakly scattering biological tissues [1–3]. It has been successfully applied as imaging and diagnostic tool in ophthalmology [4]. SOCT has been also adapted to visualize physiological parameters, since the functional studies might be important for the early detection and prevention of eye diseases [5–9].

The first attempt to provide an image of 3D vasculature in human retina and choroid was Optical Coherence Angiography (OCA) proposed by Makita et al. [10]. The non-zero phase difference between adjacent OCT signal measurements was used to contrast the blood vessels. Due to a limitation of the measurable phase difference it is difficult to detect the relatively low or high blood flow velocity as well as flows imaged by strongly attenuated OCT signals. To overcome these drawbacks an alternative method called Scattering Optical Coherence Angiography (S-OCA) was proposed to map the choroidal vasculature [11]. Here the light extinction in blood was used as a contrasting parameter. S-OCA is an en-face method with intensity threshold-based binarization used to separate a choroidal vasculature. Because the main parameter used here in the segmentation procedure is the OCT back-reflected intensity, a low scattering non-vessel region might be misinterpreted as a vasculature region. The 3D images covering 5mm × 5mm of retinal area were obtained within 5.5s examination. A slight improvement was achieved by combining S-OCA and Doppler OCA using ultra-high resolution SOCT instrumentation but still the images of choroid were difficult to interpret [11].

An and Wang proposed another approach to visualize retinal vasculature in three dimensions [12]. The method called Optical Micro-AngioGraphy (OMAG) was performed to map vascular perfusion within human retina and choroid. A constant modulation frequency was introduced to a standard SOCT measurement to separate the moving and static scattering components within a sample. In this method, Doppler shifts originating from moving constituents greater than the introduced modulation frequency can be identified and used to segment vessels. OMAG provides high quality 3D angiogram covering the area of 2.5mm × 2.5mm collected during approx. 10s of the examination. However, it needs a long and complicated data processing.

The first quantitative 3D in vivo measurement of human retinal blood flow was demonstrated by Bachmann et al. using resonant Doppler FdOCT technique [13]. The method is based on the effect of interference fringe blurring that occurs when the path difference between structure and reference changes during camera integration. An electrooptic phase modulator in the reference arm was driven with specific phase cycles locked to the Doppler frequency of the sample flow. For this reason the signals of blood flow were enhanced, whereas the signals of static structures were suppressed. The 3D image of the retinal blood flow (covering 2.4mm × 1.7mm) is measured within ~10s of examination time. Resonant Doppler FdOCT requires complex optical setup and precise synchronization, however it operates without elaborate signal processing algorithm (computational time <5min).

Recently, Tao et al. has demonstrated velocity-resolved blood flow imaging technique called single-pass flow imaging spectral domain OCT (SPFI-SDOCT) [14]. They use a modified Hilbert transform and spatial frequency analysis to obtain a stack of depth-resolved images, each representing a finite velocity range. In this way a velocity distribution in depth is reconstructed. Unfortunately, in case of 3D imaging of the retina the examination takes 25s, what makes this method rather inconvenient both for the patient and operator.

The latest volumetric tomography of human retinal blood flow was demonstrated by Schmoll et al. [15].They applied novel CMOS detector and achieved high speed 3D Doppler FdOCT. Measured velocity is proportional to the phase difference between adjacent OCT signals. In general, phase measurements are more vulnerable to low signal-to-noise ratio (SNR) than intensity-based methods thus they are considered as lower performance technique [12].

In this paper we would like to propose an alternative way of quantitative 3D blood flow velocity imaging using joint Spectral and Time domain OCT (STdOCT). We report a fully automated, straightforward and time effective, intensity-based method of extracting volumetric maps of a human retinal and choroidal blood circulation. We discuss limitations and advantages of this method and demonstrate human retinal and choroidal images of vasculature, obtained during examination lasting less than 3 second and processed within 3 minutes.

2. Theory

2.1 Joint Spectral and Time Domain OCT

The method of joint Spectral and Time domain OCT has been described in details elsewhere [16]. Therefore, we will here only summarize briefly the STdOCT method and focus on the procedure of 3D velocity imaging.

One of the main advantages of STdOCT is the fact that it does not need a significant hardware changes in the standard SOCT setup; instead it requires a high density scanning. In this technique a set of collected spectral fringe signals is arranged into subsets, each consisting of M consecutive signals. Each subset is regarded as M repetitions of standard SOCT measurement performed at the same lateral position and creates a 2D interferogram registered both in wavenumber k and in time space t (Fig. 1(a) ). Spectral fringe signals undergo a standard SOCT preprocessing consisting of background removal, rescaling to wavenumber k domain and numerical dispersion correction.

 

Fig. 1 Joint Spectral and Time domain OCT analysis of Intralipid flow in a glass capillary performed in one position of the probing beam; individual images are linked via one- (1D) and two-dimensional (2D) Fourier transformations (FT). (a) 2D interferogram consisting of M = 40 spectra recorded in time increments Δt = 40μs. (b) M = 40 reconstructions of the axial structure of the glass capillary (A-scans). (c) Retrieval of Doppler shift for each k. (d) Doppler shift distribution in depth. The flow is too high, axial velocity exceeds the velocity range and signal aliasing can be observed. (e) Single line in final structural tomogram - plot of maximal signal amplitude as a function of depth. (f) Velocity profile - plot of the points with maximal signal amplitude; displayed as intensity graph compose a single line in velocity map.

Download Full Size | PDF

Fourier transformation converts the spectra from optical wavenumber domain k into the depth position z (Fig. 1(b)) and from the time domain t into the Doppler frequency domain ω (Fig. 1(c)). In order to increase the sampling density in the Doppler frequency domain we used additional zero-padding in the time domain. Both structural and velocity tomograms are created from signal in $z-\omega$-plane. For each, in depth position z, the signal with maximal amplitude I_max is found. This amplitude defines one point of structural A-scan, while its position along the frequency axis ω(I_max) forms one point of velocity A-scan. As a result all M spectra are used in the process of creation of both structural and velocity A-scan. Applying both Fourier transformations, one after another, 2D spectral fringes ($k-t$ plane) are converted to the Doppler frequency distribution in depth ($z-\omega$ plane, Fig. 1(d)). All images obtained by Fourier transformation (Fig. 1(b-d)) are accompanied by their mirror images due to the fact that registered interferogram is a real-valued function [17]. In STdOCT as well as in standard SOCT the complex conjugation of the image is considered unwanted, thus not displayed in the final cross-sectional images. Therefore, in the practical applications only positive depths are displayed. The sign of the Doppler frequency value indicates forward or backward direction of the flow.

If any structure within the sample moves with a velocity V at an angle α to the probing beam, an axial component of the velocity value equals $v=V\cos \alpha$ and causes the Doppler frequency $\omega =2vk$ due to the movement.

In order to extract information about flow velocities we have to perform additional two steps. The first step is based on obvious statement that the structural point can be associated only with one velocity. Therefore, to get a point on v-z diagram (Fig. 1(f)), the pixel of largest intensity on each line corresponding to depth position z was chosen from the ω-z diagram. In the second step we apply a signal intensity threshold to suppress the random velocity noise. The threshold is constant (κ = 3) and chosen on the basis of SNR analysis performed previously [16]. A suppression of the velocity noise is necessary for 3-D visualization, however for 2-D cross-sectional velocity maps is unnecessary. In the case of significant drop in intensity due to the signal washout the threshold can even affect the image negatively by removing points with high velocity but low intensity from the center of a vessel.

The maximal quantifiable values of bidirectional flow axial velocity $v_{\pm \max }$ is given by the time interval $\Delta t$ between consecutive measurements of the spectral fringes, Eq. (1):

Velocity resolution depends on spatial oversampling as it was demonstrated by Tao et al. [18]. To avoid loss of lateral resolution in the velocity-resolved vessel maps, the spatial range of constructed subsets of A-scans should not exceed the scanning beam spot size.

2.2 Three-dimensional velocity imaging and segmentation of retinal and choroidal vasculature

In STdOCT each set of M spectra generates velocity profile along the direction of the probing beam. If the total B-scan is collected, both structural cross section and velocity map can be derived. In order to get a 3D distribution of flows a number of A-scans in the raster pattern has to be collected. 3-D structural image is displayed as a semi-transparent cloud of points in false color scale: blue indicates the lowest signal intensity, green the medium one and the highest signals are encoded by yellow color. Velocity values are encoded by two colors indicating direction of blood flow: blue for maximal positive velocity value + v_max and red color for negative one -v_max. Zero value is encoded by white color, thus colors’ saturation corresponds to the value of velocity. The process of 3D flow imaging is preceded by motion artifacts compensation due to involuntary eye movements. Axial displacements between adjacent A-lines are called bulk motion artifacts.

The bulk motion artifacts within single B-scan are compensated by histogram-based method analogue to that used in the phase sensitive technique presented by Makita, et al. [10]. However, in our case, it operates on Doppler frequency shifts recalculated to velocity values (Fig. 2(a) ) instead of phase changes. The first step in our algorithm for the bulk motion correction is applying an additional intensity threshold, which has been mentioned above. We assume here that the motion of the entire structure as an additive process gives much stronger effect than the blood flow. Another assumption is that signals coming from vessels always usually give smaller back reflected intensity than static structure due to interference fringe wash-out.. After eliminating signals, which are below the intensity threshold, the remaining velocity values correspond mainly to the static tissue (Fig. 2(b)). These velocity values can be plotted as a histogram. The number of bins is determined by a number of data points after Fourier transformation. The velocity of the bulk motion v_bulk is the maximum count in the histogram (Fig. 2(c)). Now it is possible to introduce an offset to the velocity profile corresponding to the value indicated by the histogram. This procedure enables centering the true zero velocity value and correcting the velocity profile by using a circular shift (Fig. 2(d)). Although the whole procedure can be performed automatically with the default threshold, it is also possible to change the threshold value manually. It is useful when signal aliasing occurs and procedure with default value cannot operate properly.

 

Fig. 2 Pictorial representation of the bulk motion correction algorithm. (a) Raw velocity profile with bulk motion artifact, complex conjugation of the image is marked by the gray background and not considered. (b) Velocity profile plotted for signals that exceed a certain intensity threshold. (c) Histogram of velocity values corresponding to (b). (d) Corrected velocity profile.

Download Full Size | PDF

Bulk motion artifacts between neighboring B-scans are eliminated by a cross-correlation method. Consecutive cross-sectional images are aligned with respect to the previous one and a shift value is calculated from the position of the maximum of the cross-correlation function between these images [19].

Except the above mentioned alignments, none of the images presented in this paper undergo any fitting, filtering, smoothing, edge detection, manual segmentation or any other advanced image processing techniques to obtain full axial velocity information and to segment out the blood vessels.

Apart from the standard structural tomogram (Fig. 3(a) ) and a velocity map (Fig. 3(b)), additional image of segmented vessels can be obtained (Fig. 3(c)). The latter can be done by another simple modification of existing numerical procedure. If we assume that all locations with velocity values above a certain threshold correspond to ocular vasculature, a structural image of moving components can be recognized as segmented vessels. To obtain such image, one can filter out the static components from the structural image. First a binary mask is created from the velocity map; a point of the mask is set to 1 for velocity higher than arbitrary value | ± 0.2 v_max|. Otherwise the pixel value is set to 0. The mask image is applied to the structural tomogram, thus only moving elements such as blood cells remain (Fig. 3(c)). Structural images of segmented vessels do not require compensation of bulk motion artifact between A-scans, thus all potential problems with this procedure can be omitted. The contrast value in this figure is adjusted to increase the visibility of weak intensities and some details are better visualized here than in velocity images. The entire segmentation process is performed automatically. The set of segmented structural tomograms forms a qualitative 3D image of vessels.

 

Fig. 3 Segmentation of blood vessels. (a) Standard structural image.(b) Velocity map used in further segmentation procedure. (c) Structural image of segmented vessels. Lower row – magnification 8.5 × .

Download Full Size | PDF

Both the set of 2D cross-sectional velocity maps and data showing segmented blood vessels are loaded to the commercially available visualization software (Amira, Visage Imaging, Inc.) and after adjusting color bars a 3D image of blood velocity in ocular vasculature and a 3D image of segmented vessels are reconstructed.

3. Methods

We use a laboratory high resolution Spectral OCT system comprising a femtosecond laser (Fusion, Femtolasers, Δλ = 160nm, repetition rate 70MHz, central wavelength 810 nm), a fiber Michelson interferometer with fixed reference mirror and a custom designed spectrometer with a volume phase holographic grating and an achromatic lens focusing spectrum on a 12-bit CMOS line-scan camera Sprint (spL4096-140k), Basler, (Fig. 4(a) ).

 

Fig. 4 (a) Experimental Spectral OCT system: PC polarization controller, DC dispersion compensator, NDF neutral density filter, XY galvo-scanners, CMOS line-scan camera. (b) Exemplary high quality and high resolution (2.3μm in tissue) cross-sectional image of human retina obtained with the described SOCT setup using high speed CMOS camera.

Download Full Size | PDF

For all retinal blood flow examination the optical power of light illuminating the cornea was 750 μW. To meet safety conditions pulses from the femtosecond laser were stretched by the fiber loop to obtain quasi continuous light illuminating the retina. CMOS camera was set to acquire 2048 pixels from 4096 available photosensitive elements. During the measurement 2200 A-scans and 100 B-scans were registered within less than 3s (Fig. 5(a) ). Axial resolution was set to 2.3μm and the lateral resolution was equal to scanning beam spot-size on the retina and assumed to be around 20μm. The velocity recovery is based on 16 spectra, taken with every 4 A-scans, thus 2200 spectra result in 546 points of velocity map in x-axis (Fig. 5(b)). The total computational time does not exceed 3 min. assuming zero padding in time domain up to 128 points.

 

Fig. 5 Scanning protocol (a) 3-D imaging - driving signals for X and Y scanners. (b) The procedure of generating 2-D velocity map from a single B-scan (c) Two types of sampling depending on the size of imaged area.

Download Full Size | PDF

For images that cover the area of 3mm x 2mm the scanning protocol enables dense sampling (Fig. 5(c)). Spatial range of constructed subsets of 16 A-scans does not exceed the scanning beam spot size and there is no loss of lateral resolution in the velocity-resolved vessel maps. There is only one volume (#1) that was sparsely scanned and does not fulfill the condition of lossless sampling.

Most of blood flow images were obtained with exposure time equal to 12 μs (repetition time $\Delta t$ = 13.3 μs) that corresponds to the velocity range of ± 15.2 mm/s. Only section 4.3. presents images taken for some other velocity ranges. Scanning protocols for measurements with repetition time longer than 30 μs was slightly modified to preserve the short examination time. In the case of Fig. 11 the number of B-scan was twofold decreased with preserving the size of imaged area. The reconstruction of vessels network in y-direction was deteriorated, however the quality in x-direction was maintained. In Fig. 12 the same sampling condition are preserved for all presented images, however in the case of measurement with repetition time $\Delta t$ = 61.3 μs the size of imaged area was decreased.

 

Fig. 11 Comparison between imaging with two different velocity ranges (volume #3). (a) 3D velocity image overlaid onto structural OCT data with delimitation between sensory retina and choroid. (b, c) En-face velocity maps corresponding to retina layer for ± 15.2mm/s and ± 5.2mm/s, respectively. Green arrows indicate visible details, asterisks mark points of size measurement: capillary diameter is ~20 μm, bigger vessel is ~50 μm. (d, e) En-face velocity maps corresponding to choroid and sclera for ± 15.2mm/s and ± 5.2mm/s, respectively. Note, that color scale was adjusted separately for each velocity range.

Download Full Size | PDF

 

Fig. 12 Velocity maps obtained from 3-D images for three different velocity ranges. Note that color scale was adjusted separately for each velocity range. (a, d, g) Velocity maps of sensory retina. (b, e, h) Velocity maps of choroid and sclera. (c, f, i) Cross-sections of vessels with velocity profiles indicated by green arrows on velocity maps.

Download Full Size | PDF

4. Results

Three-dimensional images of blood vessels and their velocity distribution were measured at several locations across the healthy human retina (Fig. 6 ). Fundus regions of 5 × 5mm and 3 × 2mm (marked in Fig. 6 as #1 and #2, respectively) were scanned in the location of optic nerve head. The other volumes were acquired from the 3 × 2mm area located in a close proximity to the optic disc and in the fovea (Fig. 6, #3 and #4, respectively). These examples are used to discuss different imaging parameters that affect quality of velocity images.

 

Fig. 6 Regions of interest in STdOCT measurements of retinal and choroidal blood flow. (a) Fundus photo with marked areas of STdOCT scanning (white rectangles #1-4), red dashed rectangle corresponds to the area covered by angiographic image (right). (b) ICG angiography (TRC-50DX Type LA, Topcon).

Download Full Size | PDF

4.1 Three-dimensional imaging with detailed velocity information

The upper part of Fig. 7 shows the retinal and choroidal vasculature in the same fundus region obtained with red free fundus photography, ICG angiography and SOCT fundus view. STdOCT data are presented in Fig. 7 (d-f) in three different ways as: a rendering of 3D distribution of blood flow velocity (d), blood flow velocity map in en-face projection (e) and segmented blood vessels also presented in en-face projection (f). A direction of blood flow is determined and coded in reference to the direction of light propagation, thus changes of a vessel orientation in z direction result in a superficial change of blood flow direction. When blood flow direction becomes perpendicular to the probe beam, a Doppler shift is zero (Fig. 9(b) ). In general, knowing the direction of blood flow and vessel orientation, a vein or artery can be unambiguously recognized. However, this is only true for vessels that can be continuously tracked starting from the optic nerve head.

 

Fig. 7 Blood vessels in the region of optic nerve head (#1, 5mm × 5mm, exposure time 12 μs, maximum value of the axial velocity ± 15.2mm/s, measurement time <3s.). (a) Red-free fundus photography. (b) ICG angiography. (c) SOCT fundus view. (d) Reconstructed 3D velocity image overlaid onto structural SOCT data (Media 1). (e) Velocity en-face map created from 3D STdOCT data. (f) En-face view of segmented vessels. None of presented images require filtering, smoothing or manual segmentation.

Download Full Size | PDF

 

Fig. 9 Different types of discontinuities of recovered blood vessels (details from Fig. 7(e)). (a) Discontinuity due to low sampling in y-direction (upper part of Fig. 7(e)). (b) Lack of signal due to vessel reorientation (in the middle of Fig. 7(e)). (c) Discontinuity due to perpendicular direction of the vessel in respect to the direction of sampling beam (bottom left hand corner of Fig. 7(e)). (d) Discontinuity due to a high value of the blood flow velocity exceeding the measurement range (top right hand corner of Fig. 7(e)).

Download Full Size | PDF

Despite STdOCT method provides distribution of true values of axial velocity component in three dimensions there are problems with visual assessment of these flows in volumetric renderings or en-face maps. In the case of color coding the gradient of velocities (usually parabolic cross-sectional distribution of velocity) causes a fading effect of displayed blood vessels. Therefore, in order to make the vessels visible and to enhance the contrast we have brought to saturation the color maps displaying the true axial velocity values in three dimensional blood flow maps (Fig. 7(d, e)). In order to get insight on velocity distribution within the vessel it is much better to look at 2D cross-sectional velocity maps (Fig. 8(a) ) – exactly the same which were used to create the 3D image presented in Fig. 7. Choosing the location of a vessel at the en-face map and decoding velocity values one can obtain velocity profiles in a desired direction (Fig. 8(b)).

 

Fig. 8 Velocity distribution within vessels. (a) 3D en-face and 2D cross-sectional STdOCT velocity maps (right panel); 2D map is extracted from 3D image with the exact location marked at the en-face map (left panel). (b) Velocity profiles in both lateral (x) and axial (z) directions for vessels visible at presented cross-sectional velocity map.

Download Full Size | PDF

4.2 Discontinuities in reconstructed images of blood vessels

Above demonstrated volumetric velocity image #1 is rather low sampled (550 × 100 pixels, 5 × 5mm). A single point of the velocity map corresponds to 36μm in x-direction, which is more than assumed 20μm of lateral resolution. Furthermore, the distance between successive B-scans is 50μm, so the light beam covers only 40% of imaged area. This low sampling density leads to discontinuities in y-direction of the image. It is clearly visible in the case of laterally tilted small vessels (Fig. 9(a)). Another reason for low visibility of reconstructed blood vessels is their almost perpendicular orientation in respect to the direction of OCT light beam. In this case the value of the axial component of the velocity is very low. Additionally due to small variations of the blood vessel topography a direction of the flow often changes generating a pattern of alternating blue and red patches distributed along the vessel (Fig. 9(c)). Finally the velocity signal can decay when flow is high and intensity signal vanishes due to fringe wash-out (Fig. 9(d)).

The influence of the low sampling density into a quality of 3-D velocity maps can be estimated and the deterioration of 3D images due to low density scanning can be assessed by comparison with more densely sampled images. As it is shown in Fig. 10 the lateral oversampling – as it is in the volume #2 (550 × 100pixels, 3 × 2mm) – is sufficient to avoid a loss of lateral resolution. Successive B-scans do not overlap, however they are taken with every 20μm to cover the entire scanned area.

 

Fig. 10 Comparison between densely-sampled volume #2 and sparsely-sampled volume #1, 3mm × 2mm, exposure time 12 μs, axial velocity range ± 15.2mm/s. (a) Movie of 3D reconstructed velocity image #2 overlaid onto structural OCT data (Media 2). (b) En-face velocity map obtained from 3D image #2. (c) En-face view of 3D image of segmented vessels #2. (d) En-face velocity map obtained from cropped 3D image #1. (e) En-face view of segmented vessels from cropped volume #1. None of the presented images require filtering, smoothing or manual segmentation.

Download Full Size | PDF

4.3 Quantifiable velocities

The maximum quantifiable velocity depends on the repetition time between consecutive A-scans (Eq. (1). In order to measure high velocities, the A-scan rate should be shortened. We also expect that increasing the maximum velocity value we should broaden the velocity range. However, this would be true if the smallest detectable velocity depends only on the random phase fluctuations and do not depend on the repetition time. To verify experimentally these thoughts we performed two volumetric measurements at the same location but for two different repetition times of 13.3μs and 36.3μs, corresponding to exposure time of 12 μs and 35 μs, respectively (Fig. 11). Although, the velocity images with the range of ± 5.5 mm/s (Fig. 11(c,e)) are sparsely scanned in y-direction (discontinuous the light beam covers 50% of the imaged area) they reveal more details. The vessel oriented horizontally in the bottom of Fig. 11(c) is visible almost through the entire volume and while it is only partially visible in Fig. 11(b). Also some capillaries (green arrows, Fig. 11(c)) are only visible when measured with repetition times of 36.3μs. This observation implies that the minimum detectable velocity is different for both repetition times. Similar results were obtain by Schmoll et al. [15].

Usually, qualitative measurements provide seemingly better reconstruction of vasculature than quantitative measurements. In qualitative measurements signal aliasing enhances visualization of vessels network. However, choosing a proper velocity range is crucial to quantify the velocity. To present that issue we have performed 3 additional measurements for 3 different velocity ranges ± 24.2 mm/s (the highest possible for presented setup), ± 9.5 mm/s and ± 3.3 mm/s (the lowest possible to keep the examination time within 4 s); all other settings remained unchanged (Fig. 12). The chosen ranges correspond to exposure times: 7 μs (repetition time 8.3 μs), 20 μs (21.3 μs) and 60 μs (61.3 μs).

Although velocity maps obtained for ± 3.3 mm/s of velocity range provide details invisible elsewhere, this measurement is useless to quantify the velocity in two bigger vessels. Even though all visible vessels are rather small and their diameters do not exceed 100 μm, the range should be matched to bigger and smaller vessels separately. Otherwise, either small vessels are distinguished or bigger are distorted.

It has to be noted that the main difficulty with imaging of small capillaries is due to their almost perfect perpendicular orientation in respect to the scanning beam. This is especially visible in close proximity to the macula. To have a closer look into this region we performed a dense STdOCT scan in the macular region (volume #4, 3mm × 2mm, exposure time 12 μs, axial velocity range ± 15.2mm/s, measurement time <3s, Fig. 13 .).

 

Fig. 13 Reconstructed 3D velocity image at the region of fovea overlaid onto structural OCT data (Media 3); volume #4, 3mm × 2mm, exposure time 12 μs, axial velocity range ± 15.2 mm/s, measurement time < 3 s.

Download Full Size | PDF

Beyond the foveal avascular zone in the retina layer a few vessels are visible. The vascular structure also can be distinguished in the choroidal layer.

5. Discussion and conclusion

Quantitative retinal blood flow imaging raises an issue of providing true values of blood velocity in the retinal and choroidal vasculature. STdOCT has been already proved as a sensitive and reliable method for flow velocity estimation [16]. However, similarly to the phase sensitive techniques it has a finite measurement window of velocities, which can be reliably found. Because the most of Doppler OCT techniques are able to measure only the axial component of the velocity the entire span of different blood flow velocities in a real retina is large. It depends on the sizes of blood vessels, general blood circulation and additionally on the retinal topography. Therefore, even in a small area (2 × 3mm) it is difficult to have the measurement range which will cover blood flow velocities in all imaged vessels, especially if we would like to measure simultaneously both vessels and capillaries.

A closer look at the Fig. 11 reveals that quantitative measurements of blood flow beneath the retina provides additional random Doppler signals that form well-defined vasculature pattern. The velocity images of volume #3 reveals that this vascular pattern considered as a choroidal vessel, surprisingly is observed at the depth of sclera (Fig. 14(d) ). Since observed vessels partially correlate with the vasculature visible in the retina and/or choroid, the origin of Doppler shift in signal corresponding to the sclera is unclear. It seems that some residual signals appears beneath the vessel and create “virtual” vasculature in deeper layers. In contrast to the real vasculature the velocity values in the virtual vessels are random.

 

Fig. 14 Three-dimensional structural OCT image (volume #3) with indicated three main layers: retina, choroid and sclera and corresponding velocity maps obtained by STdOCT (a) Delimitation between sensory retina, choroid and sclera. (b-d) En-face velocity maps corresponding to retinal, choroidal and scleral layer, respectively.

Download Full Size | PDF

Signals detected at the depth of sclera originate either from scattering on scleral tissue or from the multiple scattering in choroid. The question is how a movement that occurs on the path of light beam affects the optical frequency and phase of penetrating light. This phenomenon has not been reported yet and requires a systematic study.

In conclusion, we demonstrate the capability of joint Spectral and Time domain Optical Coherence Tomography (STdOCT) to assess human ocular blood axial velocity in-vivo with high sensitivity in three dimensions. Straightforward segmentation of vessels is simultaneously performed with velocity estimation resulting in two sets of 3D data, one quantitative and the other qualitative. The data are acquired within regular measurement time. Time requirements of both measurement (3s) and post-processing (3min) renders STdOCT the fastest three-dimensional OCT technique to image blood vessels and blood velocity in retina and choroid. For the first time the quantitative method is applied to image choroidal vasculature.

Additionally, we report an observation of vascular pattern in OCT signals at the depth of sclera. Further systematic investigation of observed Doppler random signals in light backscattered from sclera is however beyond the scope of the current study.

We also discuss the question of unambiguous velocity estimation for ocular vessels in terms of chosen scan density and velocity range.

Future work on this technique should include accurate analysis of the retinal and choroidal blood flows, which requires extended experiments with additional heartbeat control and statistical analysis performed for many subjects.