High-resolution imaging of the human retina has always been a challenge due to imperfect optical properties of the human cornea and lens, which limit the achievable resolution. We present a noniterative digital aberration correction (DAC) to achieve aberration-free cellular-level resolution in optical coherence tomography (OCT) images of the human retina in vivo. The system used is a line-field spectral-domain OCT system with a high tomogram rate, reaching 2.5 kHz. Such a high speed enables us to successfully apply digital aberration correction for not only imaging of human cone photoreceptors but also to obtain an aberration- and defocus-corrected 3D volume. Additionally, we apply DAC on functional OCT angiography data to improve lateral resolution and compensate for defocus. The speed necessities for the use of DAC in patient imaging are quantified by measuring the axial motion of 36 subjects. The first demonstration of DAC on OCT angiography as well as the motion analysis is important for future work dealing with DAC.
© 2017 Optical Society of America
Optical coherence tomography (OCT) is a noninvasive imaging technique, which has become a standard medical imaging tool in ophthalmology. With the rise of cell and gene therapies, the quest for cellular resolution during in vivo tissue assessment has become an important research focus in biomedical imaging. The investigation of major eye diseases ultimately calls for tools that give access to cellular details, both anatomic and physiologic. Improved structural sensitivity and resolution leads furthermore to enhanced diagnostic capabilities to spot diseases early on, when treatment may be most successful.
Higher resolution for retinal imaging is in theory achieved by adjusting the probing beam size to the pupil size of the human eye. For a fully dilated pupil, the diameter can reach 7 mm, which at a wavelength of 840 nm, leads to a diffraction-limited spot size of 2.5 μm at the retina. Unfortunately, the optics of the human eye are far from being perfect, introducing significant aberrations. The aberrations become dominant for larger beam diameters, leaving it practically impossible to obtain cellular resolution in vivo. Only the application of adaptive optics to retinal imaging—a technique adapted from astronomy—brought an important breakthrough. Such systems measure the wavefront distortions introduced by the human eye with a wavefront sensor and correct the distortions by a programmable deformable mirror or other spatial phase modulator. The combination of hardware adaptive optics (AO) and scanning laser ophthalmoscopy (SLO) or fundus photography has already been successfully demonstrated [1,2]. Still, even AO SLO with its tight confocal gating is limited in its depth sectioning capabilities. Only the combination of AO with OCT allows for high isotropic resolution in all three dimensions . The reason is the decoupling in OCT of lateral resolution from the axial resolution. The latter is only dependent on the supported optical bandwidth and center wavelength of the light source. This unique strength enables fascinating new insight into retinal anatomy and physiology on the level of single photoreceptors, such as the recent study of disk shedding . However, the complexities of adaptive optics together with its costs make the commercialization of such technology difficult.
Computational correction of wavefront distortions has been demonstrated first in the framework of holography . OCT senses the full complex backscattered sample field similar to holography. First attempts to use this information for defocus correction have been based on simple geometric optics modeling to determine the defocus phase . More advanced approaches employed inverse modeling to determine the scattering potential and thereby the backscattered field . Both models yield similar performance for defocus correction. For higher order aberration correction, iterative approaches have been applied using optimization of image sharpness metrics , which is computationally expensive. This method has been used to determine the distorted wavefront and to feed this information into a hardware deformable mirror device for adaptive optics correction . Such a sensorless adaptive optics approach is advantageous in situations when a wavefront sensor cannot easily be applied, for example, for small animal imaging.
Computational aberration correction with OCT is, however, challenging for in vivo situations. A requirement for the phase correction is a proper lateral relation of the signal phase. During in vivo assessment, however, significant involuntary axial as well as lateral motions occur. Beam scanning geometries therefore require high recording speeds to maintain lateral and axial phase correlation. In a recent approach, en face OCT has been employed, which records single en face planes with high line rates of 4 kHz to maintain phase correlation . With this system, first, to the best of our knowledge, in vivo aberration-corrected photoreceptor images have been demonstrated without hardware adaptive optics. The system has, however, several drawbacks. First, as a time-domain system, the sensitivity is significantly smaller than for Fourier-domain OCT. Second, it is challenging to axially adjust the en face plane to structures at a depth of interest given the short coherence length of less than 10 μm together with the presence of axial motion. En face time-domain OCT has only been applied in a stable manner in connection with axial tracking, enhancing again the system complexity significantly . Another approach based on full-field swept-source OCT employed high-speed camera technology to record a full tissue volume with a single wavelength sweep . The method is attractive as it intrinsically achieves high lateral phase correlation. The phase stability allowed for iterative aberration correction in postprocessing as well as for phase-sensitive recording of intrinsic photoreceptor signals. The disadvantages of the approach are the missing confocal gate leading to contrast loss due to multiple scattering of light as well as the low detection sensitivity due to the ultrahigh speed recording array sensor.
The approach that we are following is based on parallel OCT, using line-field (LF) illumination instead and thereby maintaining half of the confocal gating . We introduce a high-resolution spectrometer-based or spectral LF OCT reaching B-scan rates of more than 2 kHz with high sensitivity of single B-scans. The high speed together with the parallel acquisition of a full tomogram in a single camera exposure ensures excellent phase stability. We demonstrate digital wavefront sensing based on a noniterative split aperture approach for in vivo retinal imaging and achieve aberration-free cellular resolution by correcting the wavefront in the pupil plane through phase conjugation . The B-scan-based imaging mode allows further for label-free aberration-corrected volumetric imaging of retinal capillary beds based on motion-induced signal decorrelation between tomograms.
A. Experimental Setup
A LF system based on spectral-domain OCT (SDOCT)  was adapted for high-resolution retinal imaging; see Fig. 1 LF SDOCT obtains a full tomogram within a single recording using a 2D array sensor: the horizontal coordinate of the sensor samples the lateral sample structure, whereas the orthogonal coordinate samples the spectrum of the light at each lateral position. The spectral interference pattern yields the depth structure after remapping to wavenumber and Fourier transformation. The B-scan rate is thereby defined by the frame rate of the employed sensor. The light source is a broadband superluminescence diode (Superlum BLM-S-840) with a central wavelength of 840 nm and a full width at half-maximum (FWHM) bandwidth of 50 nm, resulting in an axial resolution of 6.2 μm in air. The power incident on the sample is 4.4 mW, which is well below the maximum permissible exposure for an extended light source, according to the European laser safety standards . An achromatic cylindrical lens CL1 at the entrance of the interferometer is used to produce a line focus at the scanner, which is relayed via an achromatic telescope (L1, L2) into the ocular pupil. In the orthogonal plane of such an anamorphic optical scheme, the retina is finally illuminated by a focused line (blue dotted line). The pupil size of the beam is 3 mm, yielding a theoretical beam waist at the retina of 5.9 μm based on the le Grand eye model . The spot size has been confirmed experimentally using an eye phantom with a resolution test target at the retina plane. The achromatic cylindrical lens CL2 in the reference arm recollimates the beam and passes it to a dispersion-compensating glass prism (DP) and the reference mirror, which is placed on a linear translation stage. The backscattered light from the sample interferes with the reference beam at the beam splitter and is propagated via the telescope (L3, L4) to a transmission diffraction grating and focused vertically via objective L5 onto the 2D array sensor, which is placed in a conjugated plane to the retina. The sensor is a 2D complementary metal-oxide semiconductor (CMOS) camera (Mikrotron EoSens CL MC1362), with a square pixel size of 14 μm. The camera is connected to a National Instrument frame grabber (NI PCIe-1433) and set to record 1280 horizontal or spectral () and 200 vertical pixels (), respectively, with 8 bit information depth, and an exposure time of 250 μs. The maximum camera frame rate at this setting is 2500 frames per second (fps), leading to an equivalent B-scan rate of maximum 2.5 kHz. The volume rate is 5 Hz for 500 pixels in the (slow) scanning direction (). The measured sensitivity is 91.1 dB for 250 μs exposure time. The overall magnification of the system from the retina plane to the sensor is 10, yielding a lateral tomogram size of 3 mm at the detector plane. The lateral extension corresponds to approximately 300 μm or 1° field-of-view (FOV) on the retina, with a pixel pitch of 1.5 μm along the tomogram. The same pixel pitch has been maintained for the scanning direction. The spectral interference pattern is recorded across the whole 1280 pixels, leading to a nominal imaging depth of 4.5 mm in air. For stable positioning, a fixation target at infinity was presented to the partner eye of the volunteer.
During lateral scanning along , the camera records a full 3D stack of pixels. In the first step, background subtraction is performed for the optimal image signal-to-noise ratio (SNR). To determine the background, the average tomogram is calculated along the slow scanning direction (), across the full volume. This accounts for the Gaussian-shaped lateral illumination across the tomogram. The background is then subtracted according to
B. Axial Motion Correction
In the following, we assume to be the background-corrected and remapped signal according to Eq. (1). By performing the Fourier transformation with respect to , one obtains the following complex-valued SDOCT signal as a function of :1, involuntary axial motion during in vivo assessment causes signal phase distortions. Let us assume that the axial displacement between successive B-scans does not lead to a phase change exceeding the unambiguous range of . To perform the phase correction, we determine first the inter-B-scan phase difference between consecutive B-scans. This is calculated by multiplying the complex signal with the complex conjugate of each A-scan pair of adjacent B-scans and taking the argument of the result [Eq. (3)]. We have 18]:
C. Digital Aberration Correction
After the bulk phase correction, we apply a split aperture method  to determine the wavefront aberration. The method operates on each en face plane by first splitting the Fourier or pupil plane of the image into equally sized subapertures and by then calculating the inverse Fourier transform for each subpupil. The shift of the low-resolution image copy of each subaperture with respect to the central image copy is then proportional to the local slope of the wavefront. Hence, the wavefront can be reconstructed similar to a Shack–Hartman wavefront sensor but based on the full image scene. By conjugating the phase of the reconstructed wavefront and applying it to the original full-complex-valued image in the Fourier plane, an aberration-corrected image can be retrieved after inverse Fourier transform. For expressing the wavefront, Zernike polynomials are used. The number of Zernike coefficients that can be calculated is dependent on the number of split apertures. Using numbers of subapertures means that a maximum of Zernike coefficients can be calculated since the central subaperture is used as reference for calculating relative shifts. The process of multiple splitting, however, leads to loss in resolution and sampling, which may result in inaccurate wavefront error calculation. Therefore, first the amount of defocus is determined, which needs only two subapertures at the pupil plane. The local slopes allow us to calculate the defocus phase . Applying the inverse of the defocus phase to the Fourier or pupil plane of the respective depth slice and calculating the inverse Fourier transform corrects for the defocus error in a single step, without the need to know physical parameters, such as the defocus distance, refractive index, or sensor pixel size.
For correcting higher order aberrations, we make use of the fact that the boundary between inner and outer segments of retinal cone photoreceptors acts as a specular reflector. In OCT, we are able to extract the particular depth at which the reflex appears. This reflex can be used as a guide star for determining the aberrated wavefront at the respective depth. In particular, after the defocus correction explained above, those reflexes of the cone mosaic become already visible. We determine then a region-of-interest (ROI) centered at the maximum intensity value of one of these cone photoreceptors reflexes and zero out the remaining image pixels. The selection of the reflex is uncritical and can be first limited to the central image region. The size of the quadratic ROI is chosen to include only a single reflex (typically ). After masking of the ROI, the spatial 2D Fourier transform is taken to get to the pupil plane, and a multiple split aperture method is applied as explained by Kumar et al. . The reconstructed wavefront is then represented in terms of Zernike polynomials. We used subapertures, allowing us to determine 24 Zernike coefficients.
3. RESULTS AND DISCUSSION
A. Phase Correlation Analysis
Proper lateral phase correlation is essential for manipulating the wavefront in postprocessing. Lateral motion distortions due to saccades happen at typical time intervals of 0.5–2 s. They appear in the worst case only once during a full volume acquisition and will therefore be neglected. Axial motion, on the other hand, will continuously be present due to the pulsation of the eye fundus with the heart rate as well as pulsating axial motion of the head itself [21,22]. Even compared to continuous lateral motion, for example, due to drift or tremor, axial motion can be identified as the dominant source of phase decorrelation during scanning. The quantification of axial motion has been extensively studied in the field of Doppler OCT for blood flow velocity determination. Assuming with good approximation linear axial bulk motion during B-scan recording time , the axial velocity can be expressed by the accumulated phase shift between successively recorded complex-valued signals [Eq. (2)] as [23,24], where is the central wavelength, and is the refractive index of the medium. If the phase shift exceeds , phase wrapping occurs and the velocity reading becomes ambiguous. In our case, where the bulk motion needs to be quantified for proper phase correction, we impose the condition that the motion-induced phase shift should not exceed . This leads to the following expression for the critical axial speed:
If the motion-induced phase difference exceeds a value of , we are not able to correct for it and loose phase correlation along the scanning direction. To determine the average maximum axial bulk motion, we retrospectively analyzed in vivo retinal OCT data acquired at the ONH region of 36 different individuals and determined the maximum present axial bulk motion speed. The bulk motion was calculated by quantifying the axial shift between successive B-scans. A median maximum axial speed of was calculated over the population of 36 subjects. Note that this value is the maximum velocity, over a period of 8 s, in the case of dominantly pulse-induced axial motion. This speed is assumed to be related to the systolic phase of the heart pulse. We included a box plot of the maximum bulk motion analysis in relation to the critical axial speed [Eq. (6)] in Fig. 2(a). Taking into account the large intersubject variation of the maximum axial motion, we need a B-scan rate of larger than 3.2 kHz to obtain stable aberration correction over a large population. The current limitation is set by the frame rate of the employed camera. The exposure time setting of 250 μs would actually support a 4 kHz B-scan rate with the same sensitivity. Due to the limitation by the data transfer rate of the camera, the B-scan rate was reduced to 2.5 kHz.
In order to exemplify the impact of proper phase correlation, we analyzed datasets acquired from a healthy volunteer at varying B-scan rate, ranging from 500 to 2.5 kHz. The maximal axial bulk motion of the volunteer was measured to be with a standard deviation of . Following Eq. (6), the B-scan rate should be larger than 1.9 kHz for this subject for stable wavefront reconstruction and correction. For each recorded volume, we extracted the phase difference between successive B-scans and analyzed the fluctuations in the scanning direction (). Figure 2(b) shows a phase difference tomogram for the 500 Hz B-scan rate in the scanning direction at an arbitrarily selected position before bulk motion correction. Figure 2(c) plots the corresponding difference phase average (along the parallel direction) across the full volume with 500 Hz B-scan rate (yellow), 1.5 kHz B-scan rate (blue), and with 2.5 kHz B-scan rate (red). As expected, at the low B-scan rate of 500 Hz, strong fluctuations are visible due to multiple phase wrapping hindering proper phase correction. The rate at 1.5 kHz still shows several phase wrappings, whereas at the rate of 2.5 kHz, the phase difference fluctuations stay within the unambiguous range of . Single phase wraps, as visible in Fig. 2(c), can still be handled by the correction without additional effort.
B. Resolution Performance
As a next step, the performance of the digital aberration correction for high-resolution in vivo imaging was tested by applying the steps outlined in Sections 2.A–2.C. We acquired a volume of the same healthy volunteer of the previous analysis at an eccentricity of 5.5° nasally from the fovea, with a B-scan rate of 2.5 kHz. Figure 3(a) shows a maximum intensity projection over seven depth pixels centered at the layer of the junction between inner and outer segments of cone photoreceptors of the original data. In the projection image, one recognizes already the cone photoreceptor mosaic, although the image suffers from aberrations. As a first step, we use a vertically split aperture over the full FOV to calculate the defocus phase and then correct the image as outlined in Section 2.C. The correction phase is plotted to the right of Fig. 3(a), with the amount of defocus given relative to the aperture diameter of 1.6 mm. After correction, the image improves in terms of resolution, as seen in Fig. 3(b). Only a slight improvement in image contrast is seen as the image does not suffer from much defocus, but still strong aberration is visually present. For higher order aberration correction, a prominent reflex from the photoreceptor inner outer segment is used as a natural guide star, as explained in Section 2.C . We then apply the split aperture method that yields the wavefront distortion in a single step. For this task, we choose a ROI centered at the photoreceptor reflex indicated in Fig. 3(b). After reconstructing the aberrated wavefront from the masked ROI (red box), the image is corrected by employing phase conjugation in the pupil plane . To the right of Fig. 3(b), the higher order correction phase is plotted. By using a photoreceptor as a guide star, we have a well-defined landmark, which is advantageous for the subaperture method, as it becomes less sensitive to speckle noise and residual motion distortions in some image areas. In Fig. 3(c), once photoreceptor reflexes have been selected, the aberration is calculated and afterwards corrected. A clear improvement in reflex shape and SNR is obtained. The SNR was determined using the maximum value of a cone photoreceptor reflex and the standard deviation within a region, with a shadow artifact from superficial vessels. The SNR improved from the original image by 0.58 dB to the refocused and by 4.5 dB to the higher order aberration-corrected image. The root-mean-square (RMS) value for the defocus correction is 0.09 μm and 0.38 μm for the higher order aberration correction.
To further determine the performance for in vivo imaging, we acquired a series of measurements of the healthy volunteer, at decreasing eccentricity from the fovea. As the photoreceptor density toward the fovea increases, the size of each receptor decreases and will at some point cross the diffraction limit of the system. Figure 4(a) shows the projection views of the imaged OCT volumes registered to an OCT angiography (OCTA) fundus image from the same subject. The OCTA image has been obtained with an independent point scanning system  prior to LF OCT assessment and is used to properly locate the volumes with respect to the fovea. The respective eccentricities are plotted as dashed circles starting at 7–3°. Figures 4(b), 4(d), 4(f), and 4(h) are the original projection views at the photoreceptor layer. Figures 4(c), 4(e), 4(g), and 4(i) are the higher order aberration-corrected projection views. The RMS values for the correction are 0.24 μm for Fig. 4(c), 0.44 μm for Fig. 4(e), and 0.09 for Fig. 4(g). A measure for resolving the cone mosaic is in the presence of the Yellot’s ring after a 2D Fourier transform. Figures 4(j)–4(m) show the corresponding spatial frequencies of the corrected fundus views. We could resolve single cone photoreceptors down to 3.5°–4° nasally, which corresponds to a resolution of 7–8 μm . Closer than 3°, no cone pattern and associated Yellot’s ring can be seen as the photoreceptor size is smaller than the achieved resolution of our setting.
In order to demonstrate the broader applicability of the system, we included the results for a second subject as supplementary material (Supplement 1, Fig. S1).
For the investigation of digital aberration correction performance and dynamic range of the split aperture method, we reduced the amount of defocus stepwise from 3.5 to 0 diopters and recorded in vivo volumes for each setting at an eccentricity of 6°. The amount of defocus was controlled by adjusting the position of the ophthalmic lens L2 with respect to the scanning lens L1. The accommodation of the volunteer remained unaffected as the fixation at infinity was presented to the partner eye. For each position, the defocus was determined with the single-split aperture method described above . As a quality measure, we take the SNR from the signal of the most prominent cone photoreceptor reflex to the background noise. The background noise is taken as the mean signal within a vessel shadow area. Although this metric is a relative measure, it is particularly apt for the well-defined cone mosaic pattern and serves our aim to relatively quantify the trend of defocus and higher order aberration correction. Figures 5(a), 5(d), and 5(g) show the original fundus projections for the defocus positions of 3.5, 2.4, and 0 diopters. The corresponding SNR for the different settings is plotted in Fig. 5(j) (blue line), showing the expected increase toward the physical focus. The projections in Figs. 5(b), 5(e), and 5(h) are the result of digital defocus correction. We observe in Fig. 5(j) (red line) an approximately constant SNR, meeting as expected the SNR of the uncorrected image at the physical focus position. In the last step, we applied the guide-star-based digital aberration correction to the defocus-corrected images Figs. 5(c), 5(f), and 5(i). The RMS values are 1.1 μm for Fig. 5(c), 0.62 μm for Fig. 5(f), and 0.4 μm for Fig. 5(i). The SNR curve in Fig. 5(j) shows a continuous improvement toward the physical focus position up to about 3 dB in our chosen metric. The plot already indicates that the digital correction is not yet fully recovering from the physical diffraction-limited performance. The final improvement depends strongly on the prior defocus correction.
Our choice of subapertures for calculating high-order aberrations is an empirical value that gave the most stable results with the given image SNR and sampling. A limitation on the maximum number of split apertures is set by the fact that the local wavefront slope is found by correlating structural details in each subaperture. Further splitting will lead to loss in structural detail and SNR, leading ultimately to inaccurate wavefront reconstruction as discussed by Kumar et al. Similarly, strong defocus also might cause loss of structural detail in the subapertures. Based on the above results, we set the dynamic range of defocus correction with the split aperture method to 3.5 diopters.
In the case of significant defocus error, such that the defocus correction using a single-split aperture fails, iterative methods, using, for example, the Shannon entropy  as a quality measure, might still help to correct for defocus prior to employing the split aperture correction. Figure 6 presents the result of iterative defocus correction with a defocus of 5.7 diopters. Figure 6(a) is the original image, showing no visible structural details. Figure 6(b) is the digitally refocused image already with visible cone mosaic features. Figure 6(c) is the result, exhibiting a clearly visible photoreceptor pattern after guide-star-based split aperture correction using tiles.
Finally, as we use a line illumination, the system itself introduces astigmatism, which adds to the aberration of the subject and decreases the dynamic range of the correction. The split aperture method is operating on the acquired image and thus corrects for the total aberrations introduced by the subject and the system.
C. Volumetric Aberration Correction
To demonstrate digital aberration correction for the full retinal volume, we first start with the corrected photoreceptor layer of the previously described step recorded at 7° eccentricity. We assume that the higher order aberrations will stay constant over depth and the major change along depth will be defocus. Under this assumption, we can simply take the geometric propagation distance of each layer from the in-focus region at the inner–outer photo segment layer and calculate the defocus phase according to wave optics as7 In the volume-rendered image in Fig. 7(a), the photoreceptor layer Fig. 7(b) is marked by the red boundary, the outer plexiform layer in blue in Fig. 7(c), and the nerve fiber layer in violet in Fig. 7(d). The correction enables us to extend the depth of focus beyond the Rayleigh range of approximately 70 μm to more than 300 μm. RMS for Fig. 7(b) is 0.24 μm, 0.25 μm for Fig. 7(c), and 0.26 μm for Fig. 7(d).
Due to its noninvasiveness and natural coregistration with OCT, OCT angiography (OCTA) became a major topic in ophthalmic imaging and diagnosis. OCTA, in general, enhances the contrast between static and dynamic tissue. For this task, multiple B-scans are acquired at the same position within a small period of time . Any change between the recorded tomograms is due to the fluctuating signal of flowing blood in the human retina and choroid, as static tissue remains unchanged. OCTA visualizes vessels in the retina down to the level of individual small capillaries, without any use of contrast agents. The advantage of enhancing the resolution for OCTA by combination with adaptive optics has been recently well exemplified . Performing this technique for high-resolution imaging with the possibility to use digital aberration correction (DAC) would further strengthen its clinical impact. The question is whether DAC is working properly in the presence of phase and signal decorrelation introduced by the blood flow. We acquired a retinal volume in vivo with four B-scans at each position. The focus was set to the photoreceptor layer. After applying the volumetric aberration correction explained in Section 2.C, we calculated the OCTA motion contrast following Blatter et al. . Figure 8(a) shows an original tomogram from the volume, with the corresponding OCTA tomogram in Fig. 8(b). The aberration-corrected tomograms are displayed in Figs. 8(f) and 8(g). The correction itself is best appreciated in the en face projections taken at the ganglion cell layer [Figs. 8(c) and 8(h)], the inner nuclear layer [Figs. 8(g), 8(d), and 8(i)], and the outer nuclear layer [Figs. 8(e) and 8(j)]. The phase correction algorithm removes further motion artifacts visible as stripes in the uncorrected images. DAC is improving the vessel definition, which is blurred mainly due to defocus in the uncorrected images. Figure 8(k) shows the volume-rendered angiogram of the inner retinal layers, with the colors corresponding to the different vascular beds. Visualization 1 allows appreciating interconnecting capillaries. The colors encode the three different depth regions as indicated in Fig. 8(b).
We demonstrate in vivo digitally aberration-corrected high-resolution retinal OCT employing a line-field spectral-domain system. The fast B-scan rate of up to 2.5 kHz is essential for successful DAC of volumetric retinal in vivo data. It furthermore ensures high sensitivity in single B-scans and enables an on-line tomogram view for easy adjustment. Both properties are indispensable for operation of the device in a relevant clinical setting. The strength of our core algorithm of DAC is that it is noniterative and does not need knowledge of system parameters, such as pixel size, focus distance, or refractive index. Our approach consists of three steps. In the first step, the defocus at the photoreceptor depth layer is corrected based on a single-split aperture method. In the second step, a guide star is selected and a multisplit aperture method is applied for correcting higher order aberration. In the last step, the other retinal layers are corrected by applying the extrapolated and conjugated defocus phase together with the previously determined higher order correction. This correction scheme works for the full retinal volume with an extension of the Rayleigh range by more than a factor of four. We compared the performance of DAC at different eccentricities from the fovea and quantified the necessary lateral phase correlation based on the actual axial motion measured with 36 subjects. This analysis enables us to decide whether recorded data can be corrected for residual axial motion or not. Given a maximum axial motion of , stable operation of DAC requires B-scan rates above 3.2 kHz.
Finally, we demonstrate for the first time, to the best of our knowledge, the impact of DAC on high-resolution OCTA to increase vascular contrast in out-of-focus regions. The limitation in FOV is mainly due to the speed of the camera since longer acquisition times increase the probability of lateral motion artifacts. A specific feature for high-resolution imaging is the reduction of projection artifacts in OCTA, due to the short Rayleigh range, in comparison to standard OCTA modalities. This helps to improve the visibility of axially connecting vessels across capillary beds.
Carl Zeiss Meditec AG (CD 10260501); EXALOS AG (CD 10260501); Bundesministerium für Wissenschaft, Forschung und Wirtschaft (BMWFW) (CD 10260501); Austrian Science Fund (FWF) (P29093-N36).
See Supplement 1 for supporting content.
1. J. Liang, D. R. Williams, and D. T. Miller, “Supernormal vision and high-resolution retinal imaging through adaptive optics,” J. Opt. Soc. Am. A 14, 2884–2892 (1997). [CrossRef]
2. A. Roorda, F. Romero-Borja, W. Donnelly III, H. Queener, T. Hebert, and M. Campbell, “Adaptive optics scanning laser ophthalmoscopy,” Opt. Express 10, 405–412 (2002). [CrossRef]
3. M. Pircher and R. J. Zawadzki, “Combining adaptive optics with optical coherence tomography: unveiling the cellular structure of the human retina in vivo,” Expert Rev. Ophthalmol. 2, 1019–1035 (2007). [CrossRef]
4. O. P. Kocaoglu, Z. Liu, F. Zhang, K. Kurokawa, R. S. Jonnal, and D. T. Miller, “Photoreceptor disc shedding in the living human eye,” Biomed. Opt. Express 7, 4554–4568 (2016). [CrossRef]
5. A. Stadelmaier and J. H. Massig, “Compensation of lens aberrations in digital holography,” Opt. Lett. 25, 1630–1632 (2000). [CrossRef]
6. Y. Yasuno, M. Wojtkowski, K. Taira, N. Yoshifumi, M. Shuichi, I. Masahide, and Y. Toyohiko, “Non-iterative numerical method for laterally super resolving Fourier domain optical coherence tomography,” Opt. Express 14, 1006–1020 (2006). [CrossRef]
7. P. S. Carney, B. J. Davis, D. L. Marks, T. S. Ralston, and S. A. Boppart, “Interferometric synthetic aperture microscopy,” in Computational Optical Sensing and Imaging (Optical Society of America, 2007), paper CTuC2.
8. J. R. Fienup and J. J. Miller, “Aberration correction by maximizing generalized sharpness metrics,” J. Opt. Soc. Am. A 20, 609–620 (2003). [CrossRef]
9. L. P. Murray, J. C. Dainty, and E. Daly, “Wavefront correction through image sharpness maximization,” in OPTO-Ireland (International Society for Optics and Photonics, 2005), pp. 40–47.
10. N. D. Shemonski, F. A. South, Y.-Z. Liu, S. G. Adie, C. P. Scott, and S. A. Boppart, “Computational high-resolution optical imaging of the living human retina,” Nat. Photonics 9, 440–443 (2015). [CrossRef]
11. F. Felberer, J.-S. Kroisamer, B. Baumann, S. Zotter, U. Schmidt-Erfurth, C. K. Hitzenberger, and M. Pircher, “Adaptive optics SLO/OCT for 3D imaging of human photoreceptors in vivo,” Biomed. Opt. Express 5, 439–456 (2014). [CrossRef]
12. D. Hillmann, H. Spahr, C. Hain, H. Sudkamp, G. Franke, C. Pfäffle, C. Winter, and G. Hüttmann, “Aberration-free volumetric high-speed imaging of in vivo retina,” Sci. Rep. 6, 35209 (2016). [CrossRef]
13. D. J. Fechtig, B. Grajciar, T. Schmoll, C. Blatter, R. M. Werkmeister, W. Drexler, and R. A. Leitgeb, “Line-field parallel swept source MHz OCT for structural and functional retinal imaging,” Biomed. Opt. Express 6, 716–735 (2015). [CrossRef]
14. A. Kumar, W. Drexler, and R. A. Leitgeb, “Subaperture correlation based digital adaptive optics for full field optical coherence tomography,” Opt. Express 21, 10850–10866 (2013). [CrossRef]
15. B. Grajciar, M. Pircher, A. Fercher, and R. Leitgeb, “Parallel Fourier domain optical coherence tomography for in vivo measurement of the human eye,” Opt. Express 13, 1131–1137 (2005). [CrossRef]
16. “International Electrotechnical Commission Safety of laser products—Part 1: Equipment classification and requirements,” IED-60825-12.
17. H. Gross, F. Blechinger, B. Achtner, and H. Gross, Survey of Optical Instruments (Wiley-VCH, 2008).
18. M. Shuichi, H. Youngjoo, Y. Masahiro, Y. Toyohiko, and Y. Yoshiaki, “Optical coherence angiography,” Opt. Express 14, 7821–7840 (2006). [CrossRef]
19. A. Kumar, W. Drexler, and R. A. Leitgeb, “Numerical focusing methods for full field OCT: a comparison based on a common signal model,” Opt. Express 22, 16061–16078 (2014). [CrossRef]
20. A. Kumar, T. Kamali, R. Platzer, A. Unterhuber, W. Drexler, and R. A. Leitgeb, “Anisotropic aberration correction using region of interest based digital adaptive optics in Fourier domain OCT,” Biomed. Opt. Express 6, 1124–1134 (2015). [CrossRef]
21. J. Ito, A. R. Nikolaev, M. Luman, M. F. Aukes, C. Nakatani, and C. van Leeuwen, “Perceptual switching, eye movements, and the bus paradox,” Perception 32, 681–698 (2003). [CrossRef]
22. R. de Kinkelder, J. Kalkman, D. J. Faber, O. Schraa, P. H. B. Kok, F. D. Verbraak, and T. G. van Leeuwen, “Heartbeat-induced axial motion artifacts in optical coherence tomography measurements of the retina,” Invest. Ophthalmol. Visual Sci. 52, 3908–3913 (2011). [CrossRef]
23. R. A. Leitgeb, L. Schmetterer, W. Drexler, A. F. Fercher, R. J. Zawadzki, and T. Bajraszewski, “Real-time assessment of retinal blood flow with ultrafast acquisition by color Doppler Fourier domain optical coherence tomography,” Opt. Express 11, 3116–3121 (2003). [CrossRef]
24. Y. Zhao, Z. Chen, C. Saxer, S. Xiang, J. F. de Boer, and J. S. Nelson, “Phase-resolved optical coherence tomography and optical Doppler tomography for imaging blood flow in human skin with fast scanning speed and high velocity sensitivity,” Opt. Lett. 25, 114–116 (2000). [CrossRef]
25. S. G. Adie, N. D. Shemonski, B. W. Graf, A. Ahmad, P. S. Carney, and S. A. Boppart, “Guide-star-based computational adaptive optics for broadband interferometric tomography,” Appl. Phys. Lett. 101, 221117 (2012). [CrossRef]
26. L. Ginner, C. Blatter, D. Fechtig, T. Schmoll, M. Gröschl, and R. A. Leitgeb, “Wide-field OCT angiography at 400 KHz utilizing spectral splitting,” Photonics 1, 369–379 (2014). [CrossRef]
27. C. A. Curcio, K. R. Sloan, R. E. Kalina, and A. E. Hendrickson, “Human photoreceptor topography,” J. Comp. Neurol. 292, 497–523 (1990). [CrossRef]
28. M. Salas, M. Augustin, L. Ginner, A. Kumar, B. Baumann, R. Leitgeb, W. Drexler, S. Prager, J. Hafner, U. Schmidt-Erfurth, and M. Pircher, “Visualization of micro-capillaries using optical coherence tomography angiography with and without adaptive optics,” Biomed. Opt. Express 8, 207–222 (2017). [CrossRef]
29. C. Blatter, J. Weingast, A. Alex, B. Grajciar, W. Wieser, W. Drexler, R. Huber, and R. A. Leitgeb, “In situ structural and microangiographic assessment of human skin lesions with high-speed OCT,” Biomed. Opt. Express 3, 2636–2646 (2012). [CrossRef]