The principles and technical details of the SD PS-OCT setup used in this work are published elsewhere [23
]. Here is a brief summary: Light emitted from a super luminescent diode (Superlum, Moscow; center wavelength 839 nm, FWHM bandwidth 53 nm) illuminates, after being vertically polarized, a bulk optics Michelson interferometer, where it is split by a non polarizing beam splitter (NPBS) into a sample and a reference beam. The reference light transmits a quarter wave plate (QWP) oriented at 22.5°, and is reflected by a mirror. After double passage of the QWP, the orientation of the polarization plane is at 45° to the horizontal, providing equal reference power in both channels of the polarization sensitive detection unit. The sample beam passes a QWP oriented at 45°, which provides right-handed circularly polarized light to the sample. An x-y galvanometer scanner scans the beam over the retina (fast scan direction: x = horizontal; 1/e2
diameter at cornea ~ 2 mm; diffraction limited focal spot size at retina ~ 12 μm).
After recombination of the two beams at the NPBS, light is directed towards a polarization sensitive detection unit, where it is split into orthogonal polarization states by a polarizing beam splitter. The two orthogonally polarized beam components are coupled into two polarization maintaining fibers (PMFs) and directed towards two separate spectrometers which are designed identically, consist of similar components, and have to be aligned with sub-pixel accuracy with respect to each other (for details of alignment requirements, see ref. [23
]). Each spectrometer consists of a reflection grating (1200 lines/mm), a camera lens with a focal length of 200 mm, and a 2048 element line scan CCD camera with a pixel size of 14 × 14 μm2
(Atmel Aviiva M2 CL 2014). To reduce the data transfer rate, only 1024 pixels of each camera were read out (for better image quality, the data were expanded again to 2048 pixels per spectrum by zero padding in postprocessing). The maximum line rate of the camera is 29 kHz, and via camera link and a high speed frame grabber board (PCI 1428 National Instruments) data could be transferred continuously to a personal computer. The resolution of the camera is 12 bit per pixel. The sensitivity of our system was 98 dB with an integration time of 50 μs and a power of 700 μW onto the sample. The sensitivity decay, due to the finite spectrometer resolution, was 14 dB (equal in both channels) along the measurement range of 3 mm.
Our system was operated at an A-scan rate of 20k A-lines/sec. 3D data of the human retina, covering a scan field of 15° × 15° and consisting of 60 B-scans (1000(x) × 1024(z) pixels) are acquired within ~ 3 seconds. Imaging depth (in air) is ~ 3 mm. The theoretic depth resolution within the retina (assuming a refractive index of 1.38) is 4.5 μm. Pixel spacing of original datasets is ~ 4.5 μm (x) × 75 μm (y) × 3 μm (z, measured in air). Because of a trigger delay between x-scanner and B-scan data acquisition, images were truncated by 50 pixels in x-direction at one side of the image, leaving 950 transverse pixels (or ~ 14.25°). To improve 3D data processing and interactive image manipulation speed with OSA ISP software, the data were further reduced (after all the image processing steps described below) before generating the final OSA ISP data files: in x-direction, data were downsampled by a factor of 2; in z-direction, data were downsampled by a factor of 2, then the images were cropped in z-direction to remove areas that contain no information (vitreous, noisy areas below the choroid). The final z-extension of the images was 1.5–1.8 mm (in air, details are given in figure captions). summarizes pixel spacings, pixel counts, and approximate physical extensions of the OSA ISP datasets.
Dimensions of OSA ISP Datasets
After data collection the following data pre-processing steps were performed: At first fixed pattern noise, originating from the camera readout, was eliminated. This procedure consists of subtracting a mean spectrum (averaged over 1000 A-scans) from each spectral dataset, inverse Fourier transforming the dataset, removing two remaining sharp frequencies generated by the camera, and Fourier transforming the data back to obtain a spectrum free of fixed pattern noise. Afterwards, zero padding to 2048 pixels was performed, the data were rescaled into k-space (including residual dispersion compensation), and the inverse FFT of both signals was calculated, finally providing the pre-processed A-scan data of the horizontal and vertical polarization channels. Prior to calculation of the polarization data, the influence of anterior segment birefringence was compensated by a numerical method [33
In a next step, datasets of polarization independent reflectivity R, retardation δ, and optic axis orientation θ were derived from the corrected horizontal and vertical channel data as previously described [19
with A being the amplitude and Φ the phase of the respective channel, and the indices H and V denoting the horizontal and the vertical polarization channel, respectively. ΔΦ = ΦV
is the phase difference between the two channels. The unambiguous measurement ranges are 90° for δ and 180° for θ.
Apart from retardation, there is another light-tissue interaction mechanism of considerable interest for retinal imaging: We recently demonstrated that the RPE scrambles the polarization state of backscattered light, i.e., acts as a depolarizing layer [27
]. This can be used for identifying and segmenting the RPE [34
]. Since OCT is a coherent method, the light detected from a single speckle is always fully polarized, i.e., the degree of polarization (DOP) cannot be directly measured by OCT [35
]. However, depolarization can be observed in PS-OCT images because it leads to uncorrelated polarization states of adjacent speckles, i.e., their polarization state varies randomly. In a recent paper, we defined a new quantity, the degree of polarization uniformity DOPU [34
] that is closely related to the DOP and measurable by OCT. DOPU exploits the random variation of the polarization state of adjacent speckles and is obtained in the following way: At first, we calculate the Stokes vector S of each pixel:
where I, Q, U, V denote the four Stokes vector elements. Then we average Stokes vectors over adjacent pixels by calculating the mean value of each Stokes vector element within a floating rectangular window (size 15(x) × 6(z) pixels, or ~70(x) × 18(y) μm) and derive DOPU within this window by the following equation:
where the indices m indicate mean Stokes vector elements. As can be seen by Eq. (5)
, DOPU is closely related to DOP well known from polarization optics. However, we want to emphasize the statistical origin of DOPU, i.e., it can only be derived by local averaging of Stokes vector elements. It might therefore also be regarded as a spatially averaged DOP and is closely related to the apparent degree of polarization obtained by temporal averaging [35
] and to the quantity
that describes the local correlation of polarization states and was recently used for detection of multiply scattered light by OCT [36
]. (It should be mentioned that DOPU is different from the degree of circular polarization (DOCP) that was recently discussed in the context of PS-OCT [37
]. Since DOCP is based on only one Stokes vector element (V), it does not really provide a measure for the degree of polarization; e.g., retardation can transform a circularly polarized beam into a linear polarized beam with DOCP = 0 but DOP and DOPU still equal to 1
In case of a polarization preserving or birefringent tissue, the value of DOPU is approximately 1, in case of a depolarizing layer, DOPU is lower than 1. To avoid erroneous data points caused by noise (which also gives rise to random Stokes vector elements), we first apply a thresholding procedure based on the intensity data to gate out areas with low signal intensity (threshold value: 3 times local intensity noise). To reduce computation time, software compensation of corneal retardation can be omitted for the calculation of DOPU (since randomness of polarization states can be judged also from uncorrected data).
Prior to generation of OSA ISP ready data, motion artifacts were corrected by correlation of B-scans (using ImageJ module “StackReg” [38
]). Reflectivity data are displayed on a logarithmic gray scale. Polarization data δ, θ, and DOPU are displayed on a false color scale. For display of polarization data in 3D volume renderings and in cross sectional images through the 3D datasets by OSA ISP software, intensity and polarization datasets are fused to a combined dataset where the polarization values are encoded by color while the intensity data form the alpha channel (i.e., the opacity of a data point corresponds to the intensity) [39
]. This method of display ensures that areas of low reflectivity appear transparent in a volume rendering, allowing an unobstructed view of the polarization data in high-reflectivity areas. Furthermore, erroneous polarization data caused by noise in low-intensity areas are gated out of 2D cross sections by this novel display method.
More than 200 eyes of healthy subjects and patients with various diseases were imaged by SD PS-OCT. All measurements were approved by the ethics committee of the Medical University of Vienna and followed the tenets of the Declaration of Helsinki. Informed consent was obtained from the subjects and patients after explanation of the nature and possible consequences of the study. In this paper, a few selected cases are shown to demonstrate the 3D imaging and display capabilities. Full 3D datasets are provided to allow the reader to interactively view and explore the datasets, vary aspect angles, change transfer functions and coloring schemes. Finally, Stokes vector data with full resolution that were only pre-processed (i.e., prior to motion correction, cornea compensation, downsampling) are provided in the Appendix to give readers the opportunity to develop and test their own algorithms for processing polarization data and compare their results with the images shown here.