Development of a Widefield Phantom Eye for Retinal ...

Development of a Widefield Phantom Eye for Retinal Optical Coherence Tomography Anthony T. Corcoranab , Gonzalo Muyob , Jano I. van Hemertb , and Andrew R. Harveya a University

of Glasgow, Physics and Astronomy Department, Glasgow, G12 8QQ, Scotland; PLC, Research Department, Dunfermline, KY11 8GR, Scotland

b Optos

ABSTRACT We report the design, manufacture and assessment of a phantom eye that can be used to measure the performance and accuracy of ophthalmic-OCT devices. We base our design on a wide-field schematic eye, R. Navarro, J. Opt. Soc. Am. A 2 (1985), to allow the assessment of device performance relative to ± 70◦ external field of view. We have fabricated the phantom eye and have verified the structural dimensions of the multi-material 3D-printed retinal targets using calibrated-OCT images. Keywords: Phantom Eye, Optical Coherence Tomography, 3D-Printing, Visual Optics, Image Quality Assessment, Optical Design and Fabrication

1. INTRODUCTION Optical coherence tomography (OCT) has been developed to a considerable maturity for imaging the retina in three dimensions1 and its uptake for the quantitative monitoring of disease progression in ophthalmology has accelerated following the commercial availability of spectral-domain OCT devices with their superior imaging performance over previous system.2, 3 We describe here a wide-field phantom eye (WPE) that provides a closer approximation to a real eye for wide-field OCT imaging than do the planar-geometry phantom eyes that have previously been employed successfully across narrow-field imaging.4–6 OCT is typically used to measure pathology and layer thickness about the fovea and optic nerve head. This requirement has allowed previous generations of OCT devices to have a field of view (FOV) of only ±15o external field angle. In the past two years, the growing recognition of peripheral pathologies in the retina has fueled a trend towards increasing the FOV accessible by OCT.7–9 The development of wide-field OCT is justified by the growing evidence of clinically significant pathology found using wide-field scanning laser reflectance and fluorescence imaging.10–12 Current wide-field devices have shown success in monitoring disease indicators such as nonperfusion and haemorrhaging of the peripheral vasculature; in diabetic retinopathy and the increase in lipofuscin symptomatic of damage to the retinal-pigment epithelium, from age-related macular degeneration.13, 14 The capabilities of OCT make its use standard in providing quantitative information on the disease severity and the efficacy of treatments, such as anti-VEGF drugs;15, 16 however, this functionality cannot be achieved for peripheral disease indicators as there are no devices capable of accessing the retinal-periphery with repeatability. This constraint not only limits the early treatment of some of the most common retinal diseases but also prevents the investigation into new disease indicators. In the next section, we present the optical specifications of the wide-field phantom eye (WPE). In section 3, we outline the design for the housing and for the interchangeable phantom retinas. In section 4, we describe the manufacture of the phantom eye and verification of the structural dimensions. Finally, in section 5 we describe the use of the phantom eye for assessment of imaging performance by an OCT device. Further author information: Anthony Corcoran: E-mail: [email protected], Telephone: +44 (0)1383 843780 Jano van Hemert: E-mail: [email protected], Telephone: +44 (0)1383 843327

Design and Performance Validation of Phantoms Used in Conjunction with Optical Measurement of Tissue VI, edited by Robert J. Nordstrom, Jean-Pierre Bouchard, David W. Allen, Proc. of SPIE Vol. 8945, 89450F © 2014 SPIE · CCC code: 1605-7422/14/$18 · doi: 10.1117/12.2037767 Proc. of SPIE Vol. 8945 89450F-1 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/14/2014 Terms of Use: http://spiedl.org/terms

2. OPTICAL DESIGN We have developed a WPE that mimics salient imaging and retinal characteristics of a human eye across ± 70◦ external FOV (equivalent to 180◦ around the centre of the eye). The phantom was designed to closely match the optical performance of a schematic-eye model with the added restriction that it should be manufacturable with an acceptable complexity. Schematic-eye models are empirically-determined to mimic the optical properties of a typical healthy human eye. We have selected the schematic-eye model by Navarro et al., summarized in Table 1, as the basis for the design of our WPE, since it closely replicates the imaging performance and optical path-lengths of the human eye.17 More complete models have been reported, for example containing GRIN lenses (mimicking the graded index property of the human eye) and reflecting the variations in eye parameters with demographics,18 but their complexity does not readily facilitate manufacture of an accurate model. Table 1: Optical prescription for wide-field schematic eye (relaxed for 532nm).19 Surface Corneal Lens Aqueous Crystalline Lens Vitreous Retina

Radius 7.72 6.5 10.2 -6 -12

Thickness 0.55 3.05 4 16.3203 -

Ref. Index 1.38 1.34 1.42 1.34 -

Conic -0.26 0 3.316 -1 0

We aim to achieve a close similarity to the imaging performance of the schematic eye in terms of optical pointspread function (PSF), OPL and image distortion given the restrictions associated with practical implementation. One restriction is that commonly available optical glasses have higher refractive indices than do the corneal and crystalline lens of the human eye; however, we have identified that fused silica and CaF2 provide acceptably close matches to the ideal refractive indices for the corneal and crystalline lens respectively. This combination of low-refractive-index materials gives a lower deviation in OPL compared to the Navarro eye than do other combinations with larger refractive indices. Using CaF2 for the cornea, as well as for the lens, would offer superior OPL matching; however, diamond machining of CaF2 for this surface would produce a significantly poorer optical finish. The curvatures and thickness of the first lens matched those of the schematic-eye cornea to mimic the geometric reflection properties of human eyes. The optical performance was optimised for dual wavelengths of 532nm and 830nm to ensure the phantom was also able to be used for characterisation of devices such as SLOs, which typically operate in the visible region and are used in tandem with OCT at 830nm, to allow the validation of multi-modal image registration. Finally the WPE is water-filled to provide a close match to the chromatic dispersion of the human eye and hence to provide a match to the axial resolution achieved with a human eye.20 The specifications of our model eye design are summarised in Table 2. Table 2: Optical prescription for wide-field phantom eye. Surface Corneal Lens Aqueous Crystalline Lens Vitreous Retina

Radius 7.72 6.5 11.22 -5.9 -12

Thickness 0.55 3.05 3.93 16.32 -

Ref. Index 1.46 1.33 1.44 1.33 -

Conic -0.26 0 0 -0.55 0

The WPE was modelled for a 2mm entrance pupil at field angles 0-70◦ relative to the optical axis and then compared against the schematic eye model using Zemax ray tracing. This pupil diameter is the most common beam diameter used in commercial OCT systems. The spot diagrams, showing the location and spread of rays along the image plane for the WPE model and Navarro schematic model, are shown in Fig. 1 along with the tangential and sagittal RMS spot sizes, distortion and chief-ray intersection with the retina in Fig. 2.

Proc. of SPIE Vol. 8945 89450F-2 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/14/2014 Terms of Use: http://spiedl.org/terms

Target features across an entire hemisphere of the retina enable assessment of wide-field imaging performance. Manufacture of the target presents some significant challenges: techniques such as laser ablation and guided deposition are ideal for planar surfaces but their small depth-of-focus make their use cost-prohibitive on large topographies. Recent advances in additive manufacturing, often referred to as 3D-printing, have made it possible to create structures consisting of multiple materials with a tolerance of a few microns. The 3D-print process is ideal for creating large feature targets that can be used to calibrate the image-scanning patterns and the measurements made with an OCT device. The Objet Eden 350V used to manufacture the targets has a 30µm transverse resolution and a 16µm axial resolution.22 Fig. 3 (c-e) show the three target designs used to assess the suitability of 3D-printing as a method for manufacturing OCT-device calibration targets. Once the suitability of 3D-printed targets has been determined then the next step would be to generate targets with anatomy. An embedded-feature (EF) target, is composed of two materials: a transparent bulk material with similar refractive properties to PMMA (commercial name Fullcure720 ) and a highly scattering material that is embedded with Ti02 particles (commercial name Verowhite). The EF target includes structures such as imitation-vessels and alphabetic letters to aid subjective assessment of image quality. The axial-layers (AL) target has 10 alternating thin layers of Fullcure720 and Verowhite of 60µm thickness that are followed by 10 alternating layers of 120µm thickness that extend across the hemisphere of the target. The known dimensions of this target allow assessment of the accuracy of OCT measurement tools. These thickness were chosen as multiples of 2 and 4 of the 3D-printing transverse resolution to minimise aliasing of the layers. A bullseye target, was designed to assess the distortions of a wide-field SLO and en face OCT. This target uses black resin and white resin, Veroblack and Verowhite, to alter the absorption properties across the target. The target contains a bullseye pattern that has concentric circles of white and black with alternating thicknesses of 1mm and 0.1mm thick, respectively, intersected with a black cross hair.

4. MANUFACTURING ACCURACY OF THE WPE The accuracy of the WPE is limited by the manufacturing accuracy of the components within the phantom. The lenses were manufactured with a precision of ± 0.01mm for each radius of curvature, ± 0.2mm for each lens thickness and ± 0.1mm for the housing dimensions which allows a further angular freedom ± 0.56◦ on the corneal lens and ± 0.17◦ on the crystalline lens. The estimated performance change that resulted from this precision was assessed in Zemax using Monte-Carlo Analysis. Modelling indicated that 98% of manufactured eyes would exhibit diffraction-limited on-axis performance and 98% of the phantoms would have an RMS spot-radius at 30◦ off axis that is less than 12.1µm - an increase from the nominal value of 8.6µm. The distortions and layer thickness of the 3D-printed retinal targets were assessed for suitability for device calibration. Large-scale distortions of the target cause a non-linearity in the ring spacing of the bullseye target, which were measured by imaging with a macro camera followed by image processing to locate ring interfaces. This assessment showed the variation in ring spacing was below the 30µm transverse resolution of the 3D-printing process which is negligible with respect to the image-resolution capabilities of wide-field ophthalmic devices. The thickness of the layers in the AL target were measured using a sub-micron precision, calibrated OCT device. From these images, the mean layer thickness was measured as 59.9µm ± 2.8µm for each of the ten thin layers and 121.3µm ± 3.2µm for each of the ten thick layers. The deviations of these measurements are mainly due to the irregular edges of the layers. The measurement of the layer thickness was initially performed using microscope images of a cleaved target. The low confocality of this method and potential distortions from to the cleaving process resulted in a low confidence in the results from the automated segmentation; however, this method found the thick layers to be 121µm ± 8.8µm which concurs with the subsequent OCT measurement.

5. OCT AND SLO IMAGES OF THE WPE The features of the bullseye target displayed with low contrast in OCT as the black and white resin had a low contrast in their scattering properties. The en face OCT image of the bullseye target seen in Fig. 4 (a) revealed a displacement artefact, highlighted by an arrow, in the rendering of the volume scan, which was shown to originate in the ophthalmoscope. Fig. 4 (b) shows a ± 70◦ SLO with a stereographic projection taken on the Optos 200TX.


As can be seen in Fig. 2, the 532nm spot size closely matches that of the schematic eye from 0-40◦ , but for larger field angles the tangential spot size exceeds that of the schematic eye. At 830nm the spot size is smaller than the schematic eye, but this deviation is not critical as it is below the on-axis diffraction limit. A small increase in off-axis spot size related to field curvature exists but is unlikely to be observable compared to typical focus errors and will normally be removed by automated focal-correction of an OCT device. The maximum chief-ray deviation, represented here as the arc-length difference in the location of the chief-ray intersection with the retina, is 0.46mm at 830nm is shown in Fig. 2 (c). The maximum change in OPL seen in Fig. 2 (d) for 830nm is 64µm, which for a 2mm A-scan constitutes a 3.2% deviation in the location of the reference arm. This shift is less than an order-of-magnitude lower than the standard deviation in axial eye length and hence will not significantly impact algorithms used for segmentation, dispersion correction or retinal-flattening.21

3. MECHANICAL AND TARGET DESIGN The critical dimensions for the phantom housing are shown in Fig. 3. The lenses are mounted with 100µm precision in the anodised aluminium housing using RTV silicon and the water-tight seals between parts of the housing are maintained by rubber O-rings.

3.05 3.93

12

0.55

r.

Cornea 4mm Iris Crystalline Lens

12 9 4

Bullseye Target

R11.22

R5.9 C-0.55

(b)

(a)

120µm (c)

(d)

(e)

Figure 3: CAD drawings of all the components that comprise the phantom. (a) A 2D-ray trace of the WPE. Each beam corresponds to a 10◦ field angle. The surface apertures that must be maintained to ensure that there is no vignetting are (mm left to right) S1: 9.2 S2: 8 S3: 2 S4: 8 (b) A cross section of WPE housing containing the bullseye target. (c) A cross section of the embedded-features target. (d) A cross section of the axial-layers target. (e) The bullseye target.


Target features across an entire hemisphere of the retina enable assessment of wide-field imaging performance. Manufacture of the target presents some significant challenges: techniques such as laser ablation and guided deposition are ideal for planar surfaces but their small depth-of-focus make their use cost-prohibitive on large topographies. Recent advances in additive manufacturing, often referred to as 3D-printing, have made it possible to create structures consisting of multiple materials with a tolerance of a few microns. The 3D-print process is ideal for creating large feature targets that can be used to calibrate the image-scanning patterns and the measurements made with an OCT device. The Objet Eden 350V used to manufacture the targets has a 30µm transverse resolution and a 16µm axial resolution.22 Fig. 3 (c-e) show the three target designs used to assess the suitability of 3D-printing as a method for manufacturing OCT-device calibration targets. Once the suitability of 3D-printed targets has been determined then the next step would be to generate targets with anatomy. An embedded-feature (EF) target, is composed of two materials: a transparent bulk material with similar refractive properties to PMMA (commercial name Fullcure720 ) and a highly scattering material that is embedded with Ti02 particles (commercial name Verowhite). The EF target includes structures such as imitation-vessels and alphabetic letters to aid subjective assessment of image quality. The axial-layers (AL) target has 10 alternating thin layers of Fullcure720 and Verowhite of 60µm thickness that are followed by 10 alternating layers of 120µm thickness that extend across the hemisphere of the target. The known dimensions of this target allow assessment of the accuracy of OCT measurement tools. These thickness were chosen as multiples of 2 and 4 of the 3D-printing transverse resolution to minimise aliasing of the layers. A bullseye target, was designed to assess the distortions of a wide-field SLO and en face OCT. This target uses black resin and white resin, Veroblack and Verowhite, to alter the absorption properties across the target. The target contains a bullseye pattern that has concentric circles of white and black with alternating thicknesses of 1mm and 0.1mm thick, respectively, intersected with a black cross hair.

4. MANUFACTURING ACCURACY OF THE WPE The accuracy of the WPE is limited by the manufacturing accuracy of the components within the phantom. The lenses were manufactured with a precision of ± 0.01mm for each radius of curvature, ± 0.2mm for each lens thickness and ± 0.1mm for the housing dimensions which allows a further angular freedom ± 0.56◦ on the corneal lens and ± 0.17◦ on the crystalline lens. The estimated performance change that resulted from this precision was assessed in Zemax using Monte-Carlo Analysis. Modelling indicated that 98% of manufactured eyes would exhibit diffraction-limited on-axis performance and 98% of the phantoms would have an RMS spot-radius at 30◦ off axis that is less than 12.1µm - an increase from the nominal value of 8.6µm. The distortions and layer thickness of the 3D-printed retinal targets were assessed for suitability for device calibration. Large-scale distortions of the target cause a non-linearity in the ring spacing of the bullseye target, which were measured by imaging with a macro camera followed by image processing to locate ring interfaces. This assessment showed the variation in ring spacing was below the 30µm transverse resolution of the 3D-printing process which is negligible with respect to the image-resolution capabilities of wide-field ophthalmic devices. The thickness of the layers in the AL target were measured using a sub-micron precision, calibrated OCT device. From these images, the mean layer thickness was measured as 59.9µm ± 2.8µm for each of the ten thin layers and 121.3µm ± 3.2µm for each of the ten thick layers. The deviations of these measurements are mainly due to the irregular edges of the layers. The measurement of the layer thickness was initially performed using microscope images of a cleaved target. The low confocality of this method and potential distortions from to the cleaving process resulted in a low confidence in the results from the automated segmentation; however, this method found the thick layers to be 121µm ± 8.8µm which concurs with the subsequent OCT measurement.

5. OCT AND SLO IMAGES OF THE WPE The features of the bullseye target displayed with low contrast in OCT as the black and white resin had a low contrast in their scattering properties. The en face OCT image of the bullseye target seen in Fig. 4 (a) revealed a displacement artefact, highlighted by an arrow, in the rendering of the volume scan, which was shown to originate in the ophthalmoscope. Fig. 4 (b) shows a ± 70◦ SLO with a stereographic projection taken on the Optos 200TX.


(b)

(a) s:0-...,

; ¡ -'

4'.'7.1....j.,0,_,.;'?----

}

.

-'

.

yVII .

V

v

i

.

. _ " "--"' ,

- '44= i

--

..

-

;

y

.±

J.4f ¡--

Q'-

l

r

^ r

(c)

-f..'. ^y 'l, fti.

(d)

iel

m.

*'

)i

(e)

(f)

Figure 4: (a) An en face OCT image of the bullseye target with an artefact in the volume projection. (b) c of the bullseye target taken with the Optos 200TX. (c) An image showing the A wide-field SLO optomap surface of the axial-layers target. (e) A deeper image of the axial-layers target. (d) An on-axis image of the embedded-feature target. The height and thickness of the letters are 0.2mm and 1.2mm. (f ) A 30◦ off-axis image of the embedded-feature target. The AL target yielded high contrast layers in the OCT images as shown in Fig. 4 (c) and (e). The highscattering and low-scattering materials displayed with similar amplitudes in OCT images to the RNFL and the aqueous humour. Directly on-axis, both of the 120µm and 60µm layers of the AL target are well defined and resolved. With increasing field angle, the 60µm layers develop a more irregular surface and begin to merge as can be seen in Fig. 5 (b-e). Conversely the 120µm layers retain high visibility up to ±10mm from the optical axis (approximately ±35◦ ) as can be seen in Fig. 5 (e). Furthermore, there is significant intensity streaking in the images. Both of these artefacts are attributed to the irregular surface caused by the low transverse resolution of the 3D-printing process. The assertion that these artefacts are associated with the target rather than OCT device was confirmed by the recording of artefact-free images at these field angles for a target composed of stacked layers of polypropylene and cellulose adhered with acrylic (clear and matt Scotch tape) attached to a cylindrical 3D-printed mount, seen in Fig. 5 (f).


' ...

(a)

-tt