Although spectral/Fourier domain OCT technology operates at speeds of 20,000 to 40,000 axial scans per second, imaging performance is limited in practice by the relatively long times needed to acquire dense 3D OCT data sets. In this study, OCT imaging speeds of 249,000 axial scans per second were demonstrated, which is almost an order of magnitude faster than commercial spectral/Fourier domain OCT systems.
Several papers have demonstrated applications of high-speed imaging including 3D imaging, high-definition imaging, and quantitative mapping.11–14,33
The advances in imaging speed reported herein have the potential to improve the performance of these scan protocols in the ophthalmology clinic. The demonstrated improvements in imaging speed will reduce motion artifacts significantly in patients with tear film dysfunction, short attention, excessive blinking, or gross tremor. The 1060-nm center wavelength should improve imaging through dense or opaque media, and enhance visualization of the choroid, an effect first demonstrated with time-domain OCT34,35
and later with spectral/Fourier domain OCT.36
An added benefit of the longer wavelength is that the beam scanning on the retina is virtually invisible to the subject and therefore does not pose a distraction. Although water absorption in the ocular media has the potential to reduce sensitivity for retinal imaging, permissible exposure levels are higher at 1060 nm compared with 800 nm and help to offset water absorption.
In addition to improving performance of existing scan patterns, the advances in imaging speed presented herein enable very dense raster scanning protocols. The OCT fundus image generated from these dense raster scanning protocol has excellent quality due to the high transverse pixel density, as shown in . The results illustrate that axial motion is present in 3D OCT data sets (). However, with axial motion correction it is possible to achieve high quality cross-sections at any orientation ().
3D OCT data sets in were corrected for axial motion by cross-correlation of consecutive images from the raster scan. This approach is similar to the approach of the StratusOCT instrument, which cross-correlates consecutive axial scans. Cross-correlation of consecutive images improves motion correction compared with correlation of consecutive axial scans. In an ultrahigh-speed, densely sampled 3D OCT data set, consecutive images are acquired in a short period and are highly correlated due to their close transverse spacing. Cross-correlation corrects local motion artifacts well, effectively removing undulations in the vitreoretinal interface and RPE caused by eye motion, as shown in . However, cross-correlation may not recover the larger scale retinal contour in the vertical direction. Retinal contour is preserved in the horizontal direction, because the imaging speed is so rapid that negligible eye motion occurs during acquisition of a single cross-sectional image. Alternative approaches to motion correction are possible, such as the use of vertical registration scans perpendicular to the fast horizontal axis of the raster scan. In the future, the performance of different eye motion correction algorithms must be clinically validated in eyes with retinal diseases.
Image compounding by median filtering can be used to reduce speckle and improve visualization of the outer retina as shown in . In addition, it can be used to improve visualization of deeper structures in the optic nerve head, such as the lamina cribrosa (). Although previous studies have demonstrated the use of retinal tracking and image averaging to reduce speckle,37
they depend on small eye movements to decorrelate speckle in multiple images. In this study, due to the ultrahigh imaging speed, multiple images can be rapidly acquired at different transverse positions before the eye has moved significantly, which should enable more controlled and repeatable speckle reduction.
The 3D data can also be processed in various ways to generate en face images. High-quality en face visualizations require very high-density 3D OCT data sets since each pixel in the en face image requires one axial scan. The advantage of en face images is that they condense information from the 3D OCT data set, and enable assessment of microstructure which would not otherwise be visible. En face visualization of the NFL bundles (), GCL vessels (), INL capillary network (), photoreceptors and RPE (), choroid () and the lamina cribrosa () was demonstrated. In addition to this study, previous spectral/Fourier domain OCT studies have demonstrated similar methods of en face visualization of individual retinal layers12,20,30,38
(Ishikawa H, et al. IOVS
2008;49:ARVO E-Abstract 1886). These methods differ from traditional en face OCT, which is a time-domain method providing en face images of the retina.39–43
En face OCT typically acquires individual transverse priority images at one axial depth, which may intersect multiple layers of the retina depending on how the retina is tilted. Swept source/Fourier domain OCT imaging speed enables acquisition of a densely sampled volumetric data set that can be processed to create en face images by using flexible image-processing methods to increase the contrast of or selectively display the layers of interest.
One hypothesis for the pathogenesis of glaucoma is that intraocular pressure causes stress and strain resulting in mechanical failure of the connective tissues of the lamina cribrosa, scleral canal wall, and peripapillary sclera.44
Therefore, in vivo measurements of the lamina cribrosa and connective tissues may aid in the understanding of glaucoma pathogenesis. Our results show that the longer wavelength and capability for image compounding enabled by ultrahigh imaging speeds may improve visualization of these structures.
Typically, the lamina cribrosa is not visible in normal eyes in standard fundus photographs. In glaucoma, neural damage exposes more of the lamina cribrosa surface, which makes the lamina funduscopically visible. This enables quantification of parameters such as pore size and shape in standard clinical photographs. A recent longitudinal study showed that individual pore size (mean pore-to-disc area ratio) decreased between baseline and follow-up (3.9 ± 0.7 years) measurements of glaucomatous eyes.45
Although it is possible to evaluate the inner surface of the lamina cribrosa funduscopically, the outer layers of this structure may have different morphology. Therefore, a method of 3D imaging is required. In addition, it is desirable to visualize structures such as the scleral canal and flange tissue, the sclera, and pial sheath.46
This requires an imaging method with high sensitivity and capability to image the deeper portions of the sclera and lamina cribrosa.
Scanning laser ophthalmoscopes (SLO) have been used to perform morphometric measurements on the lamina cribrosa in normal and glaucomatous eyes.47–49
The SLO uses confocal detection to suppress light backscattered from axial positions outside of the focal plane. With standard implementations of this technology, confocality is limited by ocular aberrations. The adaptive optics SLO has therefore been used to achieve improved axial (longitudinal) resolution and image contrast in the presence of ocular aberrations.50
OCT, by comparison, achieves axial resolution by using low coherence interferometry, and typically achieves larger penetration depths than confocal detection alone. The lamina cribrosa has been visualized using en face OCT with dynamic focusing.51
Although this method achieves high speed and transverse resolution, consecutive en face images may not be precisely registered due to axial eye motion, and therefore 3D morphology is difficult to assess. The 3D imaging capability demonstrated herein enables visualization of the laminar morphology at depths precisely registered to the RPE (), which may eventually facilitate quantitative volumetric measurements.
Although motion artifacts are reduced at imaging speeds of 249,000 axial scans per second, the data sets are still not free from eye motion. Eye motion can result from tremor, drifts, and microsaccades.52
Tremor and drifts have small amplitudes of <50 μ
m, whereas microsaccades can cause rapid displacements of several hundred micrometers in 25 ms. Microsaccades generally occur at <5-Hz frequency, but are dependant on conditions. Axial motion is also clearly present in the 3D OCT data sets, as shown in . Therefore, further improvements in imaging speed are desirable to reduce motion artifacts. By increasing interferometer detection efficiency and increasing the average power incident on the cornea to approximately 3.3 mW35
swept-source OCT technology will scale to 500,000 to 1,000,000 axial scans per second with comparable sensitivity. Presently, higher speed line scan CCD cameras are becoming available with speeds of 300 kHz, and therefore spectral/Fourier domain OCT may be an alternative method of achieving ultrahigh imaging speeds. However, swept-source OCT has the potential to achieve more efficient detection and better sensitivity, which may ultimately make it the preferred technology for imaging at very high speeds.
In conclusion, OCT imaging of the retina was demonstrated at 249,000 axial scans per second, a factor of ~5 to 10 times higher speed than commercial spectral/Fourier domain OCT technology. The results presented herein demonstrate that high-quality imaging of the retina is possible even at ultrahigh imaging speeds. The ultrahigh imaging speeds enable high-density sampling for improved en face visualization of the architectural morphology of the retina and optic nerve head such as the nerve fiber layer bundles, intraretinal layer blood vessels and capillary network, photoreceptors, RPE, and the lamina cribrosa. The combination of the longer wavelength and ultrahigh imaging speed enables excellent visualization of the choroid, sclera, and lamina cribrosa.