Measuring Retinal Nerve Fiber Layer Birefringence, Retardation, and Thickness Using Wide-Field, High-Speed Polarization Sensitive Spectral Domain OCT

Abstract

Purpose.: Wepresented a novel polarization sensitive optical coherence tomography (PS-OCT) system for measuring retinal nerve fiber layer (RNFL) birefringence, retardation, and thickness, and report on the repeatability of acquiring these quantities.

Methods.: A new PS-OCT system, measuring at 840 nm, was developed that supports scan angles of up to 40° × 40° with an A-scan rate of 70 kHz. To test the performance and reproducibility, we measured 10 eyes of 5 healthy human volunteers five times each. All volunteers were imaged further with scanning laser polarimetry (SLP). The obtained RNFL birefringence, retardation, and thickness maps were averaged, and standard deviation maps were calculated. For quantitative comparison between the new PS-OCT and SLP, a circumpapillary evaluation within 2 annular segments (superior and inferior to the optic disc) was performed.

Results.: High quality RNFL birefringence, retardation, and thickness maps were obtained. Within the superior and inferior segments, the mean retardation for individual eyes ranged from 20° to 28.9° and 17.2° to 28.2°, respectively. The quadrant precision over the 5 consecutive measurements for each subject, calculated for the average retardation obtained within the superior and inferior quadrants ranged from 0.16° to 0.69°. The mean birefringence ranged from 0.106°/μm to 0.141°/μm superior and 0.101°/μm to 0.135°/μm inferior, with a quadrant precision of 0.001°/μm to 0.007°/μm. The mean RNFL thickness varied from 114 to 150 μm superior, and 111 to 140.9 μm inferior (quadrant precision ranged from 3.6 to 11.9 μm).

Conclusions.: The new PS-OCT system showed high image quality and reproducibility, and, therefore, might be a valuable tool for glaucoma diagnosis.

Introduction

Glaucoma, one of the most frequent eye diseases, causes the loss of retinal ganglion cells and leads to a thinning of the retinal nerve fiber layer (RNFL). ^1,2 Untreated, glaucoma leads to an irreversible loss of vision and blindness. Therefore, an early detection of glaucoma is important. It is known that the thinning of the RNFL layer is an early indicator for glaucoma. ³ One clinically established method to measure the RNFL thickness is scanning laser polarimetry (SLP), such as with the GDx-VCC and ECC devices (Carl Zeiss Meditec, Dublin, CA). SLP is based on a scanning laser ophthalmoscope with an integrated ellipsometer. It measures the phase retardation introduced by the birefringence of the RNFL. ⁴ 2D-RNFL thickness maps are calculated by assuming a constant birefringence value of the RNFL. ⁵ Clinical suspicion of glaucoma is supported by comparing the individual measurement results with a normative database. 

Polarization sensitive optical coherence tomography (PS-OCT) is a functional extension of intensity-based OCT that provides additional image information. ^6–10 PS-OCT combines the capability of intensity-based OCT to provide depth resolved 3D images of the human retina with the polarization sensitivity of SLP. Different polarizing properties of retinal layers form the basis of tissue-specific contrast that can be exploited by PS-OCT. The retinal layers can be differentiated into polarization preserving layers (e.g., photoreceptor layer), birefringent layers (e.g., RNFL, Henle fiber layer), ^11–14 and depolarizing layers (e.g., RPE). ^14,15 In addition to generating tissue-specific contrast, PS-OCT also is capable of providing quantitative data on the birefringence of the RNFL. ^8,16–18  

A fundamental advantage of PS-OCT in comparison with SLP is that RNFL retardation and thickness can be measured simultaneously but independently, providing access to birefringence, the quotient of the two quantities. In contrast, SLP measures only the phase retardation, which is a combined effect of RNFL thickness and birefringence. Furthermore, it is known that in a certain percentage of healthy and glaucomatous eyes, SLP with variable cornea compensation (VCC) shows image artifacts, so-called atypical retardation patterns. ^19–22 In these cases, RNFL thickness cannot be measured reliably with SLP. It was shown that these artifacts occur in eyes that provide a deep penetration of the sampling beam into the birefringent sclera, ²³ an effect likely related to a thin choroid. ²⁴ SLP has no depth segmentation capability and, therefore, the origin of the birefringence cannot be localized. Although recently the enhanced cornea compensation (ECC) version of the SLP has reduced these artifacts, ^21,25 PS-OCT can overcome this limitation completely because it provides depth resolved information. 

The purpose of our study was to introduce a novel wide-field, high speed spectral domain PS-OCT (SD PS-OCT) system for measuring RNFL retardation, thickness, and birefringence in vivo, and to demonstrate the high reproducibility of the measurement results. We measured 10 eyes of 5 healthy volunteers 5 times each and the results were displayed in various ways. Standard deviation maps between the 5 consecutive measurements were calculated, which proved the high reproducibility of our system. For a quantitative comparison between PS-OCT and SLP, a circumpapillary evaluation within 2 annular segments superior and inferior to the optic nerve head was performed. 

Methods

PS-OCT System

The new wide-field, high-speed SD PS-OCT system is based on a concept described previously. ²⁶ A super luminescent diode (SLD) with a center wavelength of 840 nm (full width at half maximum 50 nm) was used, providing an axial resolution of 5.8 μm in tissue. The incident light power at the cornea was 730 μW, which is below the limits for safe light exposure. ^27,28 Contrary to the previously described setup, ²⁶ in which two separate line scan cameras are used to record the orthogonal polarization states, a single camera approach was used in the new setup. ²⁹ This has the advantage of reduced system complexity; however, it requires several additional post-processing steps. ²⁹ With an A-scan rate of 70 kHz, we measured a maximum sensitivity of 98 decibels (dB, measured within a single polarization channel) and a sensitivity roll off of approximately 8 dB over 1.8 mm imaging depth. Our system supports scan angles of up to 40° × 40° and scan patterns from 512 × 125 A-scans up to 1024 × 250 A-scans. For this study we used a single scan pattern of 1024 × 250 A-scans (acquisition time approximately 4.2 seconds) and two different scan angles of 27° × 24° (for comparison with GDx-VCC) and 40° × 40° (for demonstrating large field of view imaging), respectively. More details about the new PS-OCT system have been reported previously. ³⁰  

Imaging of Healthy Volunteers

For this study we measured 10 eyes of 5 healthy volunteers with PS-OCT and SLP (GDx-VCC from Carl Zeiss Meditec). The mean age of our study group was 37 years (range 26–54 years). Of the volunteers, 4 men and 1 woman (all Caucasians) were included in the study. 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 each volunteer after explanation of possible risks. For testing the reproducibility of our PS-OCT system, 5 repeated measurements on each eye were performed. The volunteers were asked to sit back in between the measurements to include possible effects of realignment in the reproducibility data. 

PS-OCT Data Analysis

The new SD PS-OCT system was based on a single camera spectrometer. Both spectra of the two polarization channels were imaged onto a single line scan camera. This approach required several additional post-processing steps, which have been described previously. ²⁹ The processed 3D data contained amplitude and phase information for both polarization channels, which allowed us to calculate reflectivity, retardation, optic axis orientation, and degree of polarization uniformity (DOPU) values. ^9,31  

The next post-processing step was to correct the polarization sensitive data for an influence of the birefringence of the anterior segment. While the GDx-VCC measures the anterior segment birefringence in the macula and adjusts a variable wave plate for correction, ⁵ we obtained the anterior birefringence data from a measurement of the polarization state at the retinal surface and used a software-based correction approach. ³²  

In a next step, we used the 3D polarization sensitive data to generate 2D en face maps of the RNFL retardation. To observe the increased retardation due to the birefringence of the RNFL, we had to measure light that has double passed the RNFL and is backscattered from a polarization preserving layer. In principle, any layer between the posterior boundary of the RNFL and the choroid can be used (note that the signal from the sclera and the RPE cannot be used because the sclera is birefringent, whereas the RPE is depolarizing). However, to minimize phase noise one should use the signal from a layer with high reflectivity, for example inner/outer photoreceptor junction (IS/OS) or end tips of photoreceptors (ETPR). These two layers of approximately equal reflectivity are located directly anterior to the RPE. To isolate the signal from these layers, we segmented first the RPE based on its depolarizing (polarization scrambling) effect. ³¹ Then, we analyzed the signal anterior to the RPE within a window along the A-scan that extends from the anterior border of the segmented RPE until 80 μm inwards, the area in between the two green lines in Figure 1D. If the RPE could not be found in some areas of the B-scan (i.e., if the signal from the RPE is shaded by blood vessels or if the RPE is absent within the optic nerve head) our algorithm interpolated between these gaps. To find the retardation value that is assigned to the respective transverse (x-y) position of the final en face retardation map, we extracted the retardation values corresponding to the photoreceptor layer (i.e., within the window stretching from the RPE 80 μm inwards). These retardation values were weighted with the corresponding intensity value and afterwards plotted as a histogram (the weighting procedure was done to reduce the effect of noisy retardation values). The peak of this histogram then was regarded as the retardation value at the respective transverse position. This procedure was repeated for every A-scan to generate an en face retardation map. To exclude areas with low signal quality (e.g., areas that are shaded by blood vessels), intensity thresholding was used. 

Figure 1. 

 

3D PS-OCT scan (scan angle 27° × 24°, sampling 1024 × 250 A-scans). (A) Pseudo SLO en face image. Yellow line indicates the position of the B-scans shown in B to D. (B) Exemplary intensity B-scan on logarithmic gray scale. Yellow line indicates the position of the extracted A-scan values presented in E. (C) Corresponding retardation B-scan. Areas below a certain intensity threshold are displayed in gray. (D) Intensity B-scan overlaid with segmented anterior and posterior boundary of the RNFL (red lines) and window indicating the area from which retardation is derived (between green lines). (E) Extracted intensity (black dots) and retardation (red dots) values for a single A-scan marked as yellow line in B and C. Segmented RNFL is marked by red lines on the x-scale. Window indicating the area from which retardation is derived (marked by green lines on the x-scale). Linear regression fit of the retardation values within the RNFL is indicated by a solid black line within the retardation values.

To measure the birefringence and thickness of the RNFL, we needed to segment the RNFL within our 3D data. The algorithm used for this purpose relied on the fact that the RNFL has a higher reflectivity than the underlying layers. First, we applied a median filter on each intensity B-scan. Then, the anterior border of the RNFL (inner limiting membrane) was detected in each A-scan by searching for the first data point above a certain intensity threshold. Afterwards, the algorithm searched for the next data point within the A-scan below another threshold starting from the anterior border of the RNFL. In this way, anterior and posterior boundary of the RNFL were identified. By repeating this procedure for every B-scan within the 3D data set, we were able to generate a 2D en face RNFL thickness map. Areas with low signal quality (e.g., associated with blood vessel) that have been identified previously in the en face retardation map were excluded from the RNFL thickness map as well. Areas where the segmentation algorithm could not find the posterior boundary of the RNFL (e.g., within the optic nerve head) were marked in white. 

Finally, RNFL birefringence maps were calculated. For this purpose we used an algorithm that has been reported previously. ^8,12 First, the retardation values of each A-scan between the anterior and posterior boundary of the RNFL were plotted against depth. Afterwards, a linear regression fit was performed and the slope was regarded as the birefringence value at this A-scan. In this way, a 2D en face birefringence map of the RNFL was generated. Again, blood vessels and areas with low signal quality were excluded from the final image. In contrast to previously reported methods, ¹⁷ no averaging along the B-scan direction or median filtering of the final 2D en face RNFL retardation and birefringence maps was performed. It is worth noting here that the birefringence, in principle, also could be obtained by calculating the ratio between RNFL retardation and thickness. This method, however, is very sensitive to RNFL thickness segmentation and retardation errors. ^8,12,33  

For a quantitative comparison between PS-OCT and SLP, we performed a circumpapillary evaluation of the PS-OCT data. An annular area centered at the optic nerve head, with an inner radius of approximately 1.3 mm and an outer radius of approximately 1.7 mm, was used to extract RNFL birefringence, retardation, and thickness values as a function of the azimuth angle (see Fig. 4A). The values are averaged along 2048 radial lines. Additionally, the annular area is subdivided into 4 quadrants: temporal (spanning an angle of 50°, indicated by the red lines in Figure 4A), superior (120°), nasal (70°), inferior (120°), and the measurement parameters are averaged within the superior and inferior regions of interest. The obtained measurement results for each subject were presented in a table and compared to the results obtained from the GDx-VCC by statistical tests (t-test and Bland-Altman plot). 

In addition to the TSNIT curve evaluation, we further evaluated the RNFL retardation, thickness, and birefringence along the major nerve fiber bundles within the retina. Seven evaluation boxes were placed along the major nerve fiber bundles superior and inferior to the optic disc. The retardation, thickness, and birefringence values within each of the boxes were extracted and averaged, and the results were plotted. This evaluation was done for all 10 eyes included in this study and the resulting curves were averaged. 

Results

Figure 1 shows results of a 3D PS-OCT measurement obtained from a healthy volunteer (volunteer number 1, right eye). The scan is centered around the optic nerve head and covers a scan angle of 27° × 24°. Figure 1A shows a pseudo SLO en face projection of intensity data. The yellow line indicates the position of the extracted intensity B-scan shown in Figure 1B, which is displayed on a logarithmic gray scale. The corresponding retardation B-scan, which shows the increased retardation along the RNFL, is shown in Figure 1C. The yellow lines in Figures 1B and C correspond to the location of the extracted A-scan values plotted in Figure 1E. This graph shows the extracted intensity (black dots) and retardation (red dots) values along a single A-scan. One can see clearly that, within the RNFL, the retardation increases linearly with depth due to the birefringence of the RNFL. Figure 1D shows the anterior and posterior boundary of the RNFL detected by our segmentation algorithm (red lines), and the area where the values for the 2D en face retardation maps are extracted (between the green lines). 

Figure 2 shows 2D en face RNFL retardation, thickness, and birefringence maps recorded in the right eye of volunteer 1. Displayed are the results of 5 repeated individual measurements (scan angle 27° × 24°). The retardation maps presented in Figure 2A show an increased retardation superior and inferior to the optic nerve head compared to low retardation in the nasal and temporal region. Also the RNFL thickness (Fig. 2B) and birefringence maps (Fig. 2C) show clearly increased values in the superior and inferior regions, which is in good agreement with previous findings. ^12,16–18 Some areas of the birefringence maps are excluded from the final image and are marked in gray. One reason for this is that vessels are excluded from the image as it also is done for the RNFL retardation and thickness maps. Secondly, areas where the RNFL thickness was below a user defined threshold (60 μm) also are excluded because those areas do not allow a precise calculation of the RNFL birefringence. 

Figure 2. 

 

2D en face RNFL retardation, thickness, and birefringence maps of 5 repeated PS-OCT measurements recorded in the same healthy eye. Scan angle 27° × 24°. (A) RNFL retardation en face map (color scale: degree). Areas with low signal quality are displayed in gray. (B) RNFL thickness map (color scale: μm). Areas with low signal quality are displayed in gray. Regions where the RNFL thickness segmentation algorithm could not detect the posterior boundary of the RNFL (e.g., within the optic disc) are displayed in white. (C) RNFL birefringence map (color scale: degree per μm). Areas with low signal quality and areas below a RNFL thickness of 60 μm are displayed in gray.

A visual inspection of the image series presented in Figure 2 shows that the PS-OCT system is capable of measuring RNFL retardation, thickness, and birefringence with good reproducibility. Despite some slight displacements and in-frame distortions due to motion artefacts, the images are more or less identical. To quantify the reproducibility of our system we registered the individual images with respect to each other by an algorithm based on cross-correlation measurements. Afterwards the images were averaged and standard deviation maps were calculated. The results are presented in Figure 3. Figures 3A to 3C show the average of the 5 repeated measurements of RNFL retardation, thickness, and birefringence presented in Figure 2. Noticeable are the fine nerve fiber bundles that can be observed in the retardation and birefringence images. Such small details could not be resolved with previous PS-OCT systems. ¹⁷ Figures 3D to 3F show the corresponding standard deviation maps calculated for the 5 repeated measurements. Again, areas with low signal quality were removed and the standard deviation was calculated only if the corresponding pixel had a reliable value (i.e., was assigned a non-gray color value) in at least 3 of 5 repeated measurements. 

Figure 3. 

 

Averaged and SD 2D en face maps calculated for the image series presented in Figure 2. (A) Averaged RNFL retardation 2D en face map (color scale: degree). (B) Averaged RNFL thickness 2D en face map (color scale: μm). (C) RNFL birefringence map (color scale: degree per μm). (D–F) Corresponding standard deviation maps calculated for the 5 repeated measurements presented in Figure 2.

Figure 4. 

 

Comparison between PS-OCT and SLP recorded within the same eye. (A) Averaged retardation map from Figure 3A (color scale: 0°–50°). Red lines indicate the region for the circumpapillary evaluation. (B) GDx-VCC en face retardation map (color scale: 0°–61°).

To compare the results of the PS-OCT system with a clinically established method, we measured both eyes of the participating volunteers with SLP. A qualitative comparison between the measurement results of the two systems can be seen in Figure 4. Figure 4A shows the averaged retardation en face map from Figure 3A. Figure 4B presents the corresponding measurement result from the GDx-VCC (volunteer 1, right eye). Both retardation images are in good agreement, but obviously the PS-OCT image shows more details. Note that the data were calculated from the thickness map, provided by the GDx, assuming a constant conversion factor of 0.67 nm/μm and an imaging wavelength of 785 nm. ⁵ It is worth noting here that the retardation values measured by the GDx-VCC were recorded at an imaging wavelength of 785 nm and that the PS-OCT retardation values were recorded at 840 nm. To compare the values correctly (retardation in degrees) provided by both systems, the GDx-VCC results were multiplied by a factor of 785/840. 

For a quantitative comparison between the two methods, we further performed a circumpapillary evaluation of PS-OCT data similar to that provided by GDx-VCC. Figure 5A shows the retardation, measured in volunteer 1 right eye, as a function of the azimuth angle around the optic disc. The curve starts at the temporal region, and goes over to the superior, nasal, and inferior quadrants before coming back to the temporal starting point (TSNIT graph). It shows a typical double hump shape, which indicates the increased retardation superior and inferior of the optic nerve head (corresponding to the thick superior and inferior nerve fiber layer), as expected for a healthy eye. The small drops within the curve are due to blood vessels. The same evaluation was performed for the RNFL thickness and birefringence data, and the results are presented in Figures 5B and 5C. Both TSNIT graphs show a double hump pattern with a similar shape as the retardation. The data were calculated from the averaged images presented in Figures 3A to 3C. The red lines in Figures 5A to 5C indicate the mean value ± SD of the respective quantity. SD was calculated from the five individual measurements as a function of angular position. In areas where the signal quality was too low (gray areas in Figs. 3D–F) the standard deviation was not calculated and, therefore, the red lines in Figures 5A to 5C are interrupted at some points. Figure 5D shows the corresponding TSNIT graph of the GDx-VCC (green line) overlaid with the retardation values measured by PS-OCT (red line). Note that the GDx-VCC presents the results as RNFL thickness, which is calculated from the retardation map by assuming a constant RNFL birefringence of 0.67 nm/μm. ⁵ Again, a typical double hump pattern can be observed and the shape of the curve is at least comparable to the results observed by PS-OCT, although PS-OCT seems to measure higher retardation values than the GDx-VCC. It seems that the retardation curve obtained from the GDx-VCC is much smoother than the one from PS-OCT. Almost no drops in the retardation curve due to blood vessels can be observed, although these vessels are visible clearly in the en face image (Fig. 4B). 

Figure 5. 

 

Circumpapillary evaluation (TSNIT graphs) of RNFL retardation, thickness, and birefringence. Area where the circum papillary plot is calculated is indicated by two red rings in Figure 4. (A) RNFL retardation in degrees plotted as a function of the azimuth angle around the optic nerve head. Data were calculated from the averaged retardation image presented in Figure 3A. Red line indicates mean ± SD. (B) RNFL thickness and mean ± SD calculated from Figures 3B and 3E. (C) RNFL birefringence and mean ± SD calculated from Figures 3C and 3F. (D) Corresponding GDx-VCC circumpapillary evaluation in degree (left side, calculated out of the thickness data provided by the GDx-VCC and assuming a constant conversion factor of 0.67 nm/μm) and RNFL thickness in μm (right side). The green area around the GDx-VCC TSNIT curve corresponds to the normative database provided by the instrument. The red line corresponds to the retardation values measured by PS-OCT 3A.

TSNIT graphs, such as those presented in Figure 5, were calculated for all 10 eyes measured within this study. The obtained curves could be averaged further to obtain mean RNFL retardation, thickness, and birefringence TSNIT graphs. The results (Fig. 6) indicate the variation of RNFL retardation, thickness, and birefringence within the study group. Additionally, we also calculated the average RNFL retardation, thickness, and birefringence in the superior and inferior quadrants of the TSNIT graphs (areas denoted in Fig. 4A, corresponding to the scheme used for GDx-VCC). The obtained results for each volunteer are presented in the Table. The mean retardation for individual eyes within the study group ranged from 20° to 28.9° (mean value 24.6°) superiorly and 17.2° to 28.2° (mean value 22.5°) inferiorly. The mean single point precision (defined as the mean SD for each measurement point within the superior and inferior quadrants) within the defined segments over the 5 consecutive measurements ranged from 4.1° to 6° (mean value 4.9°). The quadrant precision over the 5 consecutive measurements for each subject, calculated for the average retardation values obtained within the superior and inferior quadrants, ranged from 0.16° to 0.69° (mean value 0.41°). The mean birefringence superior and inferior ranged from 0.106°/μm to 0.141°/μm (mean value 0.12°/μm) and 0.101°/μm to 0.135°/μm (mean value 0.117°/μm). The single point precision for the birefringence ranged from 0.031°/μm to 0.041°/μm (mean value 0.03°/μm), and the quadrant precision of the birefringence superior and inferior ranged from 0.001°/μm to 0.007°/μm (mean value 0.003°/μm). The mean RNFL thickness superior and inferior varied from 114 to 150 μm (mean value 129.6 μm) and 111 to 140.1 μm (mean value 124.1μm). The single point precision of the RNFL thickness measurements ranged from 4.51 to 12.1 μm (mean value 7.7 μm) and the quadrant precision of the RNFL thickness superior and inferior ranging from 2.1 to 7.8 μm (mean value 4.9 μm). Please note that the single point precision for the RNFL thickness appeared to be better than the retardation and birefringence values because for calculating the RNFL thickness maps individual intensity B-scans are median filtered in axial and transversal direction. Therefore, the quadrant precision of the RNFL thickness does not differ as much from the single point precision as for the RNFL retardation and birefringence. 

Figure 6. 

 

Averaged RNFL retardation (A), thickness (B) and birefringence (C) circumpapillary plots ± SD calculated from the measurement results of all 10 eyes included in this study.

Table. 

 

Results of TSNIT Evaluation

Subject	Eye	Mean Superior	Single Point Precision	Quadrant Precision	Mean Inferior	Single Point Precision	Quadrant Precision	Gdx Superior	Gdx Inferior
Birefringence [°/μm]
1	Right	0.134	0.034	0.004	0.135	0.032	0.003
	Left	0.141	0.032	0.002	0.126	0.032	0.003
2	Right	0.118	0.038	0.005	0.11	0.041	0.003
	Left	0.138	0.032	0.002	0.109	0.038	0.003
3	Right	0.126	0.033	0.001	0.116	0.03	0.003
	Left	0.126	0.031	0.003	0.118	0.032	0.007
4	Right	0.118	0.037	0.005	0.113	0.04	0.004
	Left	0.126	0.035	0.006	0.126	0.032	0.001
5	Right	0.106	0.038	0.003	0.115	0.036	0.003
	Left	0.108	0.039	0.003	0.101	0.036	0.006
	Mean	0.124	00.349	0.004	01.169	00.349	0.003
	Min	0.106	0.031	0.001	0.101	0.03	0.001
	Max	0.141	0.039	0.006	0.135	0.041	0.007
Retardation [°]
1	Right	28.93	5.88	0.29	28.21	5.10	0.27	19.78	21.16
	Left	27.66	4.89	0.55	27.35	4.92	0.16	21.62	18.78
2	Right	26.21	5.38	0.69	20.77	4.89	0.23	20.50	19.38
	Left	27.32	6.00	0.61	21.71	5.70	0.34	22.45	17.66
3	Right	20.84	4.07	0.45	20.57	4.48	0.38	19.12	21.77
	Left	21.85	4.33	0.60	19.68	4.23	0.44	18.84	20.13
4	Right	25.89	5.05	0.33	25.66	4.53	0.43	22.25	21.48
	Left	26.87	4.68	0.25	24.80	4.67	0.49	21.59	21.97
5	Right	20.30	5.55	0.50	19.34	4.57	0.22	16.63	18.32
	Left	20.05	5.19	0.47	17.19	4.70	0.45	15.56	17.83
	Mean	24.59	5.10	0.47	22.53	4.78	0.34	19.84	19.85
	Min	20.05	4.07	0.25	17.19	4.23	0.16	15.56	17.66
	Max	28.93	6.00	0.69	28.21	5.70	0.49	22.45	21.97
Retardation total superior + inferior				PS-OCT	23.56		GDX	19.84
Standard deviation					3.64			1.98
RNFL Thickness [μm]
1	Right	124.41	7.68	4.49	124.86	8.21	6.59
	Left	136.79	5.12	2.57	131.62	8.68	7.26
2	Right	124.74	11.9	5.58	110.97	9.81	6.84
	Left	150.07	7.21	3.27	128.44	7.1	4.84
3	Right	125.86	7.61	7.82	127.86	5.17	4.92
	Left	133.75	4.51	3.00	125.10	5.45	4.79
4	Right	131.99	9.93	4.88	118.29	12.11	4.43
	Left	139.09	8.08	4.49	140.89	9.43	6.56
5	Right	115.73	8.52	2.10	117.65	5.65	5.34
	Left	114.03	6.88	5.34	115.86	5.52	3.57
	Mean	129.65	7.744	4.35	124.15	7.713	5.51
	Min	114.03	4.51	2.10	110.97	5.17	3.57
	Max	150.07	11.9	7.82	140.89	12.11	7.26

For a quantitative comparison between retardation measured by PS-OCT and GDx-VCC, we performed a paired t-test between the retardation values in the superior and inferior quadrants of the 10 measured eyes. The mean retardation measured by PS-OCT is 23.56, SD 3.6, and for the GDx-VCC it was 19.84, SD 1.98, which results in a significant difference between the two methods (P < 0.0001). Furthermore the difference between the two methods is presented in a Bland-Altman plot shown in Figure 7. 

Figure 7. 

 

Bland-Altman plot of the mean retardation values in the superior and inferior quadrant measured by PS-OCT and GDx-VCC.

We further analyzed RNFL retardation, thickness, and birefringence as a function of distance to the ONH. Figures 8B to 8D show the mean RNFL retardation, thickness, and birefringence plotted against the number of the evaluation boxes. An example of how these evaluation boxes were placed along the major nerve fiber bundles superior and inferior to the optic disc is shown in Figure 8A (volunteer 3 left eye). One can see that the RNFL retardation and thickness decrease with distance from the optic nerve head, whereas the RNFL birefringence remains rather constant. 

Figure 8. 

 

(A) Averaged RNFL retardation map measured in volunteer 3 left eye. Numbered white boxes indicate evaluation areas along the large nerve bundles superior and inferior to the optic nerve head. RNFL retardation, thickness, and birefringence were averaged within these boxes. (B–D) Average RNFL retardation, thickness, and birefringence as a function of the evaluation boxes averaged from all ten eyes included in this study.

We investigated further the capability of the PS-OCT system to provide large field of view (40° × 40°) polarization sensitive images of the retina. For this purpose 5 repeated measurements on the eye of a healthy subject were recorded. Again the volunteer was asked to sit back between the measurements. Figure 9 shows an averaged retardation, thickness, and birefringence map (Figs. 9A–C) and the corresponding SD maps, which are calculated between the 5 consecutive measurements (Figs. 9D–F). Noticeable are the small nerve fiber bundles that can be observed by the new PS-OCT system and the effect of Henle's fiber layer, which generates a doughnut-shaped retardation distribution in the macula region. The red boxes in Figures 9B, 9C, 9E, and 9F denote an area around the fovea where the RNFL thickness segmentation algorithm failed to segment the posterior side of the RNFL. This segmentation error leads to an artefact in the RNFL thickness and birefringence map. 

Figure 9. 

 

Averaged and SD 2D large field of view en face maps calculated for 5 individual measurements. (A) Averaged RNFL retardation 2D en face map (color scale: degree). (B) Averaged RNFL thickness 2D en face map (color scale: μm). (C) RNFL birefringence map (color scale: degree per μm). (D–F) Corresponding standard deviation maps calculated for the 5 repeated measurements. The red box in B, C, E, and F denotes an area around the fovea where the RNFL thickness segmentation algorithm failed to segment the posterior side of the RNFL. This segmentation error leads to an artefact in the RNFL thickness and birefringence map.

Discussion

We presented a novel wide-field, high speed SD PS-OCT system and demonstrated its ability to measure the RNFL retardation, thickness, and birefringence. PS-OCT combines the ability of conventional OCT to provide cross-sectional intensity images of the retina that can be used to measure the RNFL thickness directly with the polarization sensitive detection of SLP, which provides information about RNFL retardation. By combining both informations that are acquired simultaneously, the birefringence of the RNFL can be obtained. The simultaneous acquisition of the parameters makes the measurements more robust against possible errors caused, for example, by motion artefacts or registration errors that can occur between separate measurements. All three parameters are believed to be an early indicator for glaucoma, ^34–36 which possibly makes PS-OCT a valuable tool for future glaucoma diagnostics. 

A qualitative comparison between retardation maps obtained from SLP and PS-OCT shows that both methods are in good agreement (Fig. 4). The circumpapillary retardation curves presented in Figures 5A and 5D have a similar shape. In the superior and inferior quadrant the GDx-VCC measures retardation values of 19.8° and 21.2°, respectively. PS-OCT measures a higher retardation value of 28.9° ± 0.3° superiorly and 28.2 ± 0.3° inferiorly. Also, the peak retardation values in the superior and inferior regions measured with PS-OCT are higher than the one measured with SLP. Furthermore, a paired t-test between the mean retardation values in the superior and inferior quadrants measured by PS-OCT and the GDx-VCC showed a significant difference (P < 0.0001). This deviation might be explained by the fact that the GDx-VCC uses an unknown averaging algorithm that possibly includes the signals from blood vessels (the small drops in the PS-OCT retardation curve, due to blood vessels, are not visible in the retardation curve presented by the GDx-VCC). Another explanation is that the GDx is based on SLO imaging, which provides no depth information. Therefore, the retardation values that correspond to light reflected at different depth positions are integrated and the signal from layers that exhibit lower retardation values (e.g., upper part of RNFL), random values (e.g., RPE) or even higher retardation values from the sclera are taken into account. ²³ On the other hand, PS-OCT provides depth information and allows us to extract the retardation values at the photoreceptor layer, which will exhibit the total retardation value introduced due the birefringence of the RNFL. Therefore, a quantitative comparison between the two methods remains difficult. Any quantitative comparison of the RNFL thickness values is not possible because SLP cannot measure RNFL thickness directly. Instead, SLP measures retardation and converts it into thickness by using a constant conversion factor of 0.67 nm/μm. Our study, and also previous reports, ^{12,16–18,37} indicated that the RNFL birefringence is not constant across the whole retina. Therefore, the absolute RNFL thickness values presented by SLP are incorrect, as it was shown in previous comparative studies. ^38,39 However, this limitation of SLP does not invalidate its usefulness for glaucoma detection. 

The results presented in Figure 6 show the mean RNFL retardation, thickness, and birefringence, and also the variance of these quantities, within the study group. Curves like these, obtained in a larger population, could be used to generate a normative database that might be useful for glaucoma detection. They also can be used to compare the results presented in our study to those of other studies that measured RNFL retardation, thickness, and birefringence using PS-OCT. ^16–18 Our results seem to be in agreement with previous reports. Cense et al. measured locally varying birefringence values ranging from 0.05°/μm to 0.175°/μm single pass birefringence. ¹² Götzinger et al. reported single pass birefringence values ranging from 0.02°/μm to 0.14°/μm, retardation values ranging from 7° to 34°, and RNFL thickness values ranging from 50 to 180 μm. ¹⁷ Yamanari et al. reported single pass birefringence values ranging from 0.05°/μm to 0.27°/μm, retardation values ranging from 5° to 40°, and RNFL thickness values ranging from 50 to 210 μm. ¹⁸ Differences, especially in the magnitude of the measurement results, might be explained by the fact that RNFL retardation, thickness, and birefringence varies from subject to subject. For a quantitative comparison between different PS-OCT systems, one would need to include a larger, age-, sex-, and race-matched population within the study or repeat the measurements of a smaller study group with different PS-OCT systems. 

One limitation of the new PS-OCT system, presented in our study, is that the RNFL birefringence cannot be calculated reliably in regions where the RNFL thickness is below 60 μm. In these areas, the number of independent data points in depth is too low, which limits the significance of the calculated birefringence value. To overcome this problem, one would need to use a light source with a broader spectrum and, hence, a better depth resolution, ²⁶ or evaluation algorithms that use transversal averaging. The latter would, however, sacrifice transverse resolution. 

The results presented in Figures 8B to 8D show how the mean RNFL retardation, thickness, and birefringence changes with distance from the optic nerve head. Note that the plots presented in Figures 8B to 8D show the averaged measurement quantities obtained from all 10 eyes within our study group (error bars indicate the mean ± SD). It is interesting to see that the RNFL retardation and thickness are reduced with distance from the optic nerve head, whereas the RNFL birefringence remained rather constant. This finding is in agreement with the previously reported results. ¹² Nevertheless, we want to point out that our results only cover the analysis of ten eyes that were measured mainly with the small scan angle (27° × 24°). The same evaluation was performed using the large scan angle (40° × 40°), such as the one presented in Figure 9. The results indicated that the RNFL birefringence remains constant over a distance of approximately 12° eccentricity from the center of the optic disc, but decreases further away from the optic disc (area is not covered by the small scan field of 27° × 24°). Therefore, the finding that the birefringence remains constant along the large nerve fiber bundles must be evaluated further using the larger scan field and a larger study population. 

We further presented novel high quality large field of view images that allowed us to visualize the distribution of nerve fiber bundles over a much wider area from the optic nerve head following the superior and inferior retinal blood vessel arcades. SLP images only cover a relatively small area around the optic nerve head, which may hinder detection of nerve fiber bundle defects. The large field of view together with the increased resolution of the en face RNFL maps may facilitate the identification of smaller RNFL bundle defects and their change over time. Being able to judge a longer part of the RNFL arcade and its bundle defects likely will improve the clinician's confidence to be confronted with early pathology instead of measurements error or physiologic variability. While the RNFL thickness and retardation decrease with increasing distance from the optic nerve head, the birefringence remains widely constant (Fig. 8). ³⁷ As a consequence RNFL bundle defects may be visible over much longer distance when using birefringence maps compared to retardation maps. Therefore, these large field of view PS-OCT images might become clinically relevant. 

In our study, we demonstrated further the good reproducibility of our new PS-OCT system. We measured 10 eyes of 5 healthy subjects 5 times repeatedly. The results demonstrated the high quality and reproducibility of the novel PS-OCT system for measuring RNFL retardation, birefringence, and thickness. The single point precision between 5 repeated measurements was 4.9° for retardation, 0.03°/μm for birefringence, and 7.7 μm for RNFL thickness. The quadrant precision over 5 repeated measurements was even better, with 0.41° for retardation, 0.003°/μm for birefringence, and 4.9 μm for RNFL thickness. These results showed that our new PS-OCT system is capable to resolve small changes in RNFL retardation, birefringence, and thickness. Therefore, decreases of the measurement quantities and nerve fiber bundle defects, which can occur in glaucoma or glaucoma suspect patients, could be monitored with high precision by the new PS-OCT system in follow-up studies. To validate this hypothesis, it would be necessary to monitor a reasonably large study group, including healthy, glaucoma, and glaucoma suspect patients, over a longer time period. 

The key question, whether PS-OCT or polarimetry in general may offer an advantage in detection of early RNFL damage during the development of glaucoma, still is open. There are, however, some findings that allow speculation in this direction. The RNFL birefringence is caused to a large amount by intracellular structures within the retinal nerve fibers, such as the microtubules. ⁴⁰ These microtubules have been demonstrated to disorganize rapidly after intravitreal injection of colchicine while the RNFL thickness remained unchanged. ^41,42 Following optic nerve transection in macaque monkeys, the RNFL birefringence started to decrease earlier and decreased more rapidly compared to the RNFL thickness. ⁴³ In glaucoma with its very slow disease process, there might be a time window of reversible damage to the RNFL with intracellular reduction of microtubules and, thus, reduced birefringence, but unchanged RNFL thickness. 

In the literature, to our knowledge there is no evidence of superiority of either OCT or SLP concerning the diagnostic power in patients with glaucoma compared to healthy subjects. ^36,44–46 However, these studies have been performed in patients with distinct glaucoma as expressed by presence of visual field defects, while the advantage of polarimetry might exist only in the very early stage of the disease, also referred to as preperimetric glaucoma. For this stage of the disease the literature is sparse and the results yet inconclusive. ^34,47 Follow-up studies over extended periods are needed to determine if polarimetric measurements provide superior diagnostic value in these cases. The high precision and reproducibility of PS-OCT, as demonstrated in this study, is a prerequisite for such studies. 

Footnotes

 Disclosure: S. Zotter, Canon, Inc. (F); M. Pircher, Canon, Inc. (F); E. Götzinger, Canon, Inc. (F); T. Torzicky, None; H. Yoshida, Canon, Inc. (E); F. Hirose, Canon, Inc. (E); S. Holzer, None; J. Kroisamer, None; C. Vass, None; U. Schmidt-Erfurth, Canon, Inc. (F); C.K. Hitzenberger, Canon, Inc. (F)