Introduction

UK Biobank is a community-based prospective cohort study, currently underway in the United Kingdom (UK), which is unprecedented in terms of both its data collection “breadth” and “depth”.[1–3] In this study, 500,000 participants, aged 40–69 years at enrollment, have been recruited, and will be followed over a period of at least 25 years. For each subject, exhaustive baseline data collection has already been performed based on questionnaires, physical measurements, and biological samples. Questionnaires will assess a range of diverse factors, including general health and disability, socio-demographic profile, smoking/alcohol usage, and dietary habits. Physical measurements included electrocardiography and exercise tolerance, spirometry, and bone density measurement, amongst others. Biological samples collected included blood, urine, and saliva. Using DNA extracted from the blood samples, high throughput genotyping is underway on all 500,000 participants. As such, UK Biobank has the potential to profoundly transform our understanding of the risk factors for disease.[3]

Although not included among the physical measurements from the initial cohort of subjects, a detailed examination of ocular health was later incorporated into UK Biobank.[1, 3] This ocular evaluation included measurements of 1) best-corrected visual acuity, 2) refractive error, and 3) intraocular pressure. Imaging of the eye was also performed, with color photography and optical coherence tomography (OCT). OCT was first described in 1991,[4] and has revolutionized the diagnosis and management of ocular disease.[5] By providing high-resolution cross-sectional (tomographic) images of the neurosensory retina in a completely non-invasive manner, OCT imaging has become indispensable for the assessment of patients with retinal disease, the commonest causes of blindness in the developed world.[6–9] Furthermore, by allowing direct visualization of central nervous system (CNS) tissue and its associated vasculature, retinal imaging with OCT and color photography may provide unique insights into the aging process and into systemic diseases such as those affecting the cardiovascular and neurological systems.[10–12]

A unique advantage of OCT imaging is its extremely high axial resolution–typically 3–8 μm when imaging the retina.[13] Image acquisition is also extremely fast, allowing comprehensive retinal scanning in seconds (typically 100+ macular scans). As a result, OCT imaging has sometimes been described as “in vivo clinical biopsy”. Due to its excellent resolution, OCT allows for accurate measurements of thickness of the neurosensory retina.[14–16] OCT is also well suited to visualization of the multi-layered architecture of the retina, and measurement of individual retinal sublayers is possible.[15] In clinical research, OCT image “segmentation” (delineation of boundaries to allow measurements) is often performed manually by trained image graders.[17, 18] While highly accurate, such an approach is time-consuming and therefore not feasible for large studies such as UK Biobank. Automated segmentation algorithms have been developed, although many are inaccurate, slow, and do not allow for batch processing of image sets from large studies.[19] As OCT imaging is increasingly incorporated into large, population-based epidemiological studies, approaches to allow for rapid, automated, quantitative analysis of OCT image sets will become increasingly necessary.

In this report, we describe an approach to the use of OCT imaging in large, population-based studies, including methods for OCT image acquisition, storage, remote analysis, and–most importantly–rapid, automated analysis of retinal thickness.

Materials and Methods

Results

Discussion

In this report, we describe methods used for the acquisition, storage, and remote, automated analysis of OCT image sets from the UK Biobank study. Our approach provides rapid, non-invasive, quantitative measures of retinal thickness (including measures of individual retinal sublayers) for a large population based cohort involving >100,000 eyes. To our knowledge, this is the first study that involves quantitative analysis of OCT images sets on this scale. By comparison, the Beaver Dam Eye Study has recently reported the results of spectral domain OCT imaging; this was also performed with the Topcon 3D-OCT system, but only involved 1544 individuals, and did not include measurements of retinal sublayers.[24] The Beijing Eye Study has also included spectral domain OCT imaging, but with the Heidelberg Spectralis system and involving 3468 individuals. In this study, measurements of subfoveal choroidal thickness were obtained manually using a calipers.[25]

We present these methods in isolation from the specific retinal thickness results for a number of reasons. Firstly, UK Biobank is an open-access resource that encourages researchers from around the world–including those from the academic, nonprofit, public, and commercial sectors–to access the data and biological samples for any health-related research that is in the public interest.[2] As such, the retinal thickness measurements provided by our study will be incorporated back into the resource and made publically available so that others can evaluate their significance as risk factors for disease. Secondly, we believe that our approach has implications for ongoing and future studies incorporating OCT imaging, whether they be large population-based epidemiological studies, phase IV or phase V clinical trials, “real-world” outcome studies, or national screening programs for ocular and systemic disease. For example, the use of electronic medical record (EMR) systems offer the ability to capture and pool a large proportion or even all data from patients undergoing a specific treatment.[26] Such systems have the benefit that all data can be collected as a by-product of routine clinical practice and can be designed to mandate capture of defined minimum datasets. Consequently, they offer a unique opportunity to assess how clinical trial results translate into “real-world” outcomes. In the recent UK Neovascular Age-Related Macular Degeneration (AMD) Database study, use of an EMR allowed assessment of visual outcomes following 92,976 treatments with ranibizumab for this condition.[27, 28] In almost all cases, OCT imaging was obtained at each treatment episode. However, without a method for automated analysis this vast quantity of clinically relevant information is not easily accessible. Similarly, our approach may be of use for screening programs for diseases such as diabetic retinopathy, where OCT is increasingly being incorporated.[29] At present, this typically involves manual assessment of images by trained “graders”–an approach that is expensive, time-consuming, subjective, and often only semi-quantitative. Without the use of rapid, automated OCT analysis techniques, such an approach may not be feasible for inclusion in screening programs on a national scale.

In our study, OCT image sets provided cross-sectional images of the neurosensory retina in the macular region, covering a 6 x 6 mm² area of each participant’s eye. By allowing detailed quantitative analysis of individual retinal sublayers, OCT imaging may thus be of considerable value for the assessment of systemic disease in epidemiological studies. For example, reductions in the thickness of the retinal nerve fiber layer (RNFL) have recently been reported in patients with mild cognitive impairment, Alzheimer’s disease, and Parkinson’s disease.[30, 31] Interestingly, in patients with multiple sclerosis, RNFL thinning appears to correlate with atrophy in both white matter and deep gray matter structures as visualized by magnetic resonance imaging (MRI).[32] In addition to ocular and neurological disease, OCT may be useful for the study of cardiovascular, metabolic, and endocrine disease–in patients with diabetes mellitus, for example, preliminary evidence from small studies suggests that neurodegeneration may precede vascular degeneration.[33, 34] We specifically highlight these medical specialties as, in May 2014, UK Biobank began a multimodal imaging extension study in 100,000 participants. This study will encompass MRI scanning of the brain, heart, and abdomen, carotid artery ultrasonography, and whole-body dual-energy x-ray absorptiometry (DXA) of the bones and joints (http://imaging.ukbiobank.ac.uk, accessed October 1^st, 2014). Correlation of these findings with OCT measures of retinal thickness is likely to be of particular interest.

While the opportunities afforded by current OCT technology are numerous, they likely represent only the tip of the proverbial iceberg. Since its initial description in 1991,[4] and even since its utilization in UK Biobank in 2009, OCT technology has continued to evolve rapidly.[35] Commercially available OCT systems now allow cross-sectional imaging of the choroid, a tissue with the highest vascular flow rate in the human body.[36–38] The choroidal circulation lacks the autoregulation of the retinal circulation and thus choroidal thickness may be affected by factors such as age,[39] refractive error,[40] diurnal variation,[41] inflammatory disease,[42] renal disease,[43] and numerous medications.[44, 45] Such variability is likely to be of value when studied in large, cross-sectional epidemiological studies. The approaches to automated analysis of retinal thickness described herein have already been modified to incorporate automated measures of choroidal thickness in newer OCT systems.

Recent commercial OCT systems also demonstrate greatly increased image acquisition speed, providing new capabilities such as “widefield” imaging of the ocular fundus (e.g., 12 x 9 mm² in area or greater, incorporating the macula and optic nerve regions in a single scan), and so-called “OCT angiography”, allowing non-invasive mapping of the retinal and choroidal vasculature.[46–48] Recently developed high-speed (100KHz or higher scan rate) OCT systems employ wavelength tunable “swept source” lasers as their light source.[35] The first commercially available swept source OCT system is the DRI OCT-1 Atlantis from Topcon. Swept source lasers are also small and robust lasers and may thus allow future OCT devices to become more readily portable, and even handheld.[49] The adoption of “binocular” designs may further remove the need for additional personnel to acquire OCT by enabling patients to align the optical axes of the instrument with the optical axes of their own eyes.[50] The potential cost-saving implications for large epidemiological studies are clear.

Our approach to automated analysis of OCT image sets has a number of potential limitations and caveats. Although the accuracy and reproducibility of the analysis software has previously been reported in patients with glaucoma, and in healthy volunteers of varying ages, it is less likely to produce accurate results in the presence of ocular disease where there is complex morphological disturbance of the retina (e.g., in patients with advanced neovascular AMD).[20] In such cases, manual segmentation of images at an OCT image-reading center, or using a crowd-sourced approach,[51] is likely to be required. Of note, UK Biobank did not specifically exclude patients with macular disease and this will have affected the accuracy of retinal boundary detection in a proportion of imaged eyes. Another limitation to consider is that although automated segmentation was completed in over 99% of eyes, this should not be confused with accuracy of automated retinal and sublayer boundary detection. Segmentation accuracy depends on a variety of factors including image quality and indeed the prevalence of morphological abnormalities in the sample of OCT images analyzed. In the UK Bioabnk OCT images we excluded 22% of the sample based on indicators of segmentation accuracy when reporting and analyzing retinal thickness in the cohort. In addition, our algorithm provides measures of retinal sublayer thickness but does not provide measures of other morphologic features that may be present as a result of retinal fluid exudation, hemorrhage, or scarring. Again, manual image analysis is likely to be required to achieve this aim.[52] Efforts are underway to develop algorithms that allow for automated detection of ocular diseases, and which place less emphasis on direct measurements of retinal thickness.[53] These algorithms may facilitate selection of those image sets most likely to require reading center grading in large studies. A further limitation of our approach is that the software program employed for this study was OCT system specific (i.e., it is only capable of performing automated analysis of OCT images from the Topcon OCT system). However, the principle of utilizing dual scale gradient information is not OCT vendor specific, and studies are underway utilizing updated versions of the software to perform automated analysis of both Spectralis OCT (Heidelberg Engineering) and Cirrus OCT (Carl Zeiss Meditec) datasets.

Conclusion

In conclusion, we report an approach to the rapid, automated measurement of retinal thickness from OCT images in the UK Biobank study. Analysis of images from ~140,000 eyes was completed in an entirely automated fashion over a 28 day period. Measurements for the neurosensory retinal thickness as whole, and for individual retinal sublayers, were obtained. In the near future, these measurements will be publically available for utilization by researchers around the world, and thus for correlation with the wealth of other data collected in UK Biobank. Finally, the automated analysis approaches we describe may be of utility for future large population-based epidemiological studies, clinical trials, and screening programs that employ OCT imaging.

Acknowledgments

This research has been conducted using the UK Biobank Resource. The collection of eye & vision data in UK Biobank was supported in part by a grant from the NIHR Biomedical Research Centre at Moorfields Eye Hospital and UCL Institute of Ophthalmology. The UK Biobank Eye and Vision Consortium is supported by a grant from The Special Trustees of Moorfields Eye Hospital. The main contact for this consortium is Prof Paul Foster (p.foster@ucl.ac.uk)

The members of the UK Biobank Eye & Vision Consortium are:

Prof Tariq ASLAM—Manchester University

Dr Sarah BARMAN—Kingston University

Prof Jenny BARRETT—University of Leeds

Prof Paul BISHOP—Manchester University

Mr Peter BLOWS—Moorfields Eye Hospital, London

Dr Catey BUNCE—King’s College London

Dr Roxana CARARE—University of Southampton

Prof Usha CHAKRAVARTHY—Queens University, Belfast

Miss Michelle CHAN—Moorfields Eye Hospital, London

Mrs Antonietta CHIANCA—UCL Institute of Ophthalmology

Dr Valentina CIPRIANI—UCL Institute of Ophthalmology

Prof David CRABB—City University, London

Mrs Philippa CUMBERLAND—UCL Institute of Child Health

Dr Alexander DAY—Moorfields Eye Hospital, London

Miss Parul DESAI—Moorfields Eye Hospital, London

Prof Bal DHILLON—University of Edinburgh

Prof Andrew DICK—University of Bristol

Prof Paul FOSTER—UCL Institute of Ophthalmology

Dr John GALLACHER—University of Oxford

Prof David (Ted) GARWAY-HEATH—UCL Institute of Ophthalmology

Mr Dan GORE—Moorfields Eye Hospital, London

Mr Srini GOVERDHAN—University of Southampton

Prof Jeremy GUGGENHEIM—Cardiff University

Prof Chris HAMMOND—King's College London

Prof Alison HARDCASTLE—UCL Institute of Ophthalmology

Prof Simon HARDING—University of Liverpool

Dr Ruth HOGG—Queen's University, Belfast

Prof Anne HUGHES—Queen's University, Belfast

Dr Pirro HYSI—King's College London

Mr Pearse A KEANE—UCL Institute of Ophthalmology

Prof Sir Peng Tee KHAW—UCL Institute of Ophthalmology

Mr Anthony KHAWAJA—Moorfields Eye Hospital, London

Mr Gerassimos LASCARATOS—Moorfields Eye Hospital, London

Prof Andrew LOTERY- University of Southampton

Prof Phil LUTHERT—UCL Institute of Ophthalmology

Dr Tom MACGILLIVRAY—University of Edinburgh

Dr Sarah MACKIE—St James’s University Hospital, Leeds

Prof Keith MARTIN—University of Cambridge

Ms Michelle MCGAUGHEY—Queen’s University Belfast

Dr Bernadette MCGUINNESS—Queen’s University Belfast

Dr Gareth MCKAY—Queen's University Belfast

Mr Martin MCKIBBIN—Leeds Teaching Hospitals NHS Trust

Dr Danny MITRY—University of Edinburgh

Prof Tony MOORE—UCL Institute of Ophthalmology

Prof James MORGAN—Cardiff University

Ms Zaynah MUTHY—UCL Institute of Ophthalmology

Mr Eoin O'SULLIVAN—University of Cambridge

Dr Chris OWEN—St George's, University of London

Mr Praveen PATEL—Moorfields Eye Hospital, London

Dr Tunde PETO—Queen's University Belfast

Dr Axel PETZOLD—UCL Institute of Neurology

Prof Jugnoo RAHI—UCL Institute of Child Health

Dr Alicja RUDNICKA—St George's, University of London

Miss Carlota Grossi SAMPEDRO—University of East Anglia

Mr David STEEL—Newcastle University

Mrs Irene STRATTON—Gloucestershire Hospitals NHS Foundation Trust

Mr Nicholas STROUTHIDIS—Moorfields Eye Hospital, London

Prof Cathie SUDLOW—University of Edinburgh

Dr Caroline THAUNG—UCL Institute of Ophthalmology

Miss Dhanes THOMAS—Moorfields Eye Hospital, London

Prof Emanuele TRUCCO—University of Dundee

Mr Adnan TUFAIL—Moorfields Eye Hospital, London

Dr Marta UGARTE—Moorfields Eye Hospital, London

Dr Veronique VITART—University of Edinburgh

Prof Stephen VERNON—University Hospital, Nottingham

Mr Ananth VISWANATHAN—Moorfields Eye Hospital, London

Miss Cathy WILLIAMS—University of Bristol

Dr Katie WILLIAMS—King's College London

Prof Jayne WOODSIDE—Queen's University Belfast

Prof John YATES—University of Cambridge

Dr Max YATES—University of East Anglia

Ms Jennifer YIP—University of Cambridge

Dr Yalin ZHENG—University of Liverpool

Dr Haogang ZHU—City University, London