ODTiD: Optic Nerve Head SD-OCT Image Dataset.

INTRODUCTION: Optic disc tilt (ODT) or tilted optic disc is a common finding in the general population. It is due to anomalous development caused by the malclosure of the embryonic optic fissure. ODT is commonly associated with high myopia as well as other conditions. In recent days, the common method to image the optic disc (OD) is by optical coherence tomography (OCT). To the best of our knowledge, there are no datasets of ODT available in the public domain. This dataset aims to make open access raw ODT OCT images to test out new image processing segmentation algorithms.
METHODS: This dataset of ODT images contains both horizontal and vertical cross-sectional images obtained using spectral-domain optical coherence tomography (SD-OCT, Cirrus 5000, Carl Zeiss Meditec Inc., Dublin, CA). The optic disc cube 200×200 program was used and all the images are aligned with the center of the optic nerve head. This dataset includes images from both clinically normal (20 eyes) and myopic subjects (101 eyes).
RESULTS: The dataset consists of clear (121) and manually marked (121) images resulting in a total of 242 images. The age distribution for all subjects combined is 27.24 ± 9.28 (range, 11.0-69.0) years. For normal subjects mean ± SD age distribution is 32.40 ± 17.23 years. Similarly, the myopia age distribution is 26.22 ± 6.37 years. Ground truth images, ie, manually segmented by a clinical expert are provided along with other meta-data includes age, gender, laterality, refractive error classification, spherical equivalent (SE), best-corrected visual acuity (BCVA), intraocular pressure (IOP), and axial length (AXL).
CONCLUSION: This open, public database is online at the ICPSR website of the University of Michigan. The dataset can be used to test and validate newly developed automated segmentation algorithms.

Introduction

Optic disc tilt (ODT) or Tilted optic disc is a common finding in the general population and is due to anomalous human development1,2 caused by the malclosure of the embryonic optic fissure.2,3 High myopia,1–4 astigmatism,2 visual field loss,1,5 defective color vision,1 and retinal abnormalities are commonly associated with ODT. Usually ODT is considered to be non-progressive1 except in cases of progressive myopia. The anomalous ODT can be misdiagnosed, as for example, in glaucoma.1,5 The prevalence of ODT is reported to be the highest (37.0%) amongst myopic Asian subjects.6

In myopic eyes with increasing axial length, the optic nerve head loses its original anatomical size and shape.7,8 In addition to change of the optic disc to a vertical oval shape, a parapapillary gamma zone develops and enlarges at the temporal disc border.7,8 In eyes with long axial lengths, (>26.0 mm) the Bruch’s membrane opening (BMO) diameter increases both horizontally and vertically.9 Likewise, the Gamma zone may develop due to an axial elongation associated with BMO enlargement.9 The characteristics of the peripapillary retinal nerve fiber layer (RNFL) thickness is also associated with the degree of temporal myopic ODT.10,11 Hence, while interpreting the RNFL thickness in myopic eyes the degree of myopic ODT should be considered.10,11

The common methods to image the optic disc is by fundus photography, optical coherence tomography (OCT) and confocal scanning laser ophthalmoscopy (Table 1). OCT helps the clinician to image the layers of the retina non-invasively and provides high-speed 3D images of high-quality retinal, optic nerve head, and choroidal vasculature images. OCT images is a useful tool to differentiate true condition/diseases from pseudo status. For example, OCT optic disc images can help in differentiate between true optic disc edema and pseudoedema.12 Table 1 summarizes the various techniques that have been presented in the literature.

Table 1

A Summary of Optic Disc Tilt Assessment Methods

S No	Author (Reference #)	Imaging Method	ODT Quantification Tool	Remarks
1	Gudapati13	SD-OCT	Imaging processing	Automated using a ground truth
2	Fraser14	SD-OCT	NA	NA
3	Cho15	CFP	Deep learning algorithm	Automated using a ground truth
4	Dervisevic16	Ophthalmic examination	Descriptively assessed	Clinical observation
5	Chen17	CFP	ImageJ	Manual method
6	Park18	CFP	ImageJ	Manual method
7	Kim3	Swept-Source OCT	NA	Manual method
8	Kosekahya19	Ophthalmic examination	Fundus appearance	Clinical observation
9	Choudhury20	Stereoscopic FP	NA	Clinical observation
10	Pan21	SD-OCT images	NA	Manual method
11	Shoeibi22	CFP	Adobe Photoshop CS6	Manual method
12	Marsh-Tootle23	SD-OCT images	NA	Manual method
13	Kim24	CFP centered on the OD	ImageJ	Manual method
14	Han5	CFP	ImageJ	Manual method
15	Sharif25	CFP	Adobe Photoshop CS6	Manual method
16	Rebolleda26	FP	FP Tilted index	Manual method
17	Sung27	OD centered CFP	ImageJ	Manual method
18	Lee9	Red-free OD centered FP	ImageJ	Manual method
19	Ando28	CFP	NA	NA
20	Hwang29	NA	NA	NA
21	Pichi30	Stereoscopic FP	FP appearance	Clinical observation
22	Chang31	CFP	Adobe Photoshop CS5	Manual method
23	Shinohara32	Stereoscopic fundus examination	Fundus appearance	Clinical observation
24	You33	SD-OCT images	NA	NA
25	Cohen34	CFP	FP appearance	Clinical observation
26	Takasaki35	HRT image	HRT printout and ruler	Manual method
27	Kim36	CFP	ImageJ	Manual method
28	Hwang10	SD-OCT images	ImageJ	Manual method
29	Samarawickrama6	Stereo CFP	ImageJ	Manual method
30	Chung37	Stereoscopic FP	Adobe Photoshop CS3	Manual method
31	Fledelius38	CFP	Slide calipers	Manual method
32	Kaimbo39	Ophthalmic examination and FP	FP appearance	Clinical observation
33	You40	OD centered CFP	FP appearance	Clinical observation
34	Tong41	Stereoscopic OD centered FP	FP appearance	Clinical observation
35	Gürlü42	SLP	NA	NA
36	Vongphanit2	CFP	Pickett small circles no. 1203	Manual method
37	Gündüz43	CFP, FFA & stereoscopic photographs	Planimetric	NA
38	Chihara44	Stereoscopic BW photographs	Slide calipers	Manual method

Abbreviations: BW, black and white; CFP, color fundus photography; FFA, fundus fluorescein angiography; FP, fundus photography; NA, not available; OD, optic disc; SD-OCT, spectral-domain optical coherence tomography; SLP, scanning laser polarimetry.

Currently, to the best of our knowledge no imaging instrument has an inbuilt method/algorithm to quantify the ODT. There are many ways to segment the ODT and hence quantify the angle of tilt. Clinicians can use manual or system generated marking. Manual marking is desired but is dependent upon the availability of a trained clinician. In general, during clinical examination the ODT is not quantified due to non-availability of easy methods or tools. Recently, authors presented an automated segmentation ODT algorithm for use with OCT images.13 The results from this methodology were compared with ground-truth (manually marked by an expert clinician) and the accuracy was reported to be 80.00%.

The availability of real world datasets is essential in accelerating health data science data analytics, including the use of routinely collected data to drive new discoveries and innovations.45 Khan et al,45 reported out of 140 unique datasets, 94 raw datasets alone were available for open access. The current paper describes here a dataset for optic nerve head OCT images from myopic subjects. This is to the best of knowledge only dataset dealing with this. This dataset aims to make open access raw ophthalmic ODT OCT images for further analysis and to test out new image processing segmentation algorithms.

Construction and Content

Image Resources

Data from Subjects who visited a tertiary care ophthalmic center in Chennai, India between January 2019 to December 2020 for ophthalmic consultation and underwent OCT imaging are included. All individuals who came for comprehensive ophthalmic examination had signed the written informed general consent agreement prior to their eye examination and approved the use of their data for research purposes. The current study was approved by the IRB of the Vision Research Foundation, Chennai, India and was conducted in accordance with the tenets of the declaration of Helsinki.

The optic nerve head was imaged using a commercially available Spectral-Domain OCT (Cirrus 5000, Carl Zeiss Meditec Inc., Dublin, CA). The optic disc cube 200×200 program was used and all the images were aligned with the center of the optic nerve head. All OCT images were 8-bit grayscale images of dimensions of 200×200 pixels corresponding to 6 mm x 6 mm (894 x 596 pixels). Images with a signal strength of 7 or higher than was included.

Demographic and Clinical Parameters

This dataset consists of a set of optic disc images (vertical and horizontal cross-sectional) from 67 subjects (34 Female, 33 Male) imaged by OCT. These datasets cover both clinical normal and also images of myopic subjects. Table 2, gives details on the dataset which includes 20 healthy normal and 101 myopic OCT images (total 121 images, 60 males, 61 females). These images are divided into three groups: 20 emmetropes (EMM) (SE 0.00 to > −0.50 D), 70 low-moderate myopes (LMM) (SE <-0.50 to −6.00 D), 31 high myopes (HM) (SE <-6.12 D).46

Table 2

Image Details of ODTiD Dataset

	Number of Images*	OD:OS	Gender Ratio (Female: Male)
Emmetropic images	20 X 2	10:10	10:10
Low-Moderate Myopic images	70 X 2	35:35	36:34
High Myopic images	31 X 2	16:15	15:16
Total images	121 X 2	61:60	61:60

Note: *Each subjects image contains both horizontal and vertical images.

In addition to unsegmented optic disc OCT images, the dataset also contains corresponding ground-truth images (each image was manually segmented by an experienced clinician), as well as meta-data, namely age in years, gender, their refractive error as spherical equivalent, refractive classification, BCVA, IOP measured (in mmHg) with Goldmann applanation tonometer, and axial length (in mm) data measured using the non-contact and high-resolution optical biometric device IOLMaster 700 (Carl Zeiss Meditec AG, Jena, Germany).

Characteristics of the Dataset

The Dataset consists of clear (121) and manually marked (121) images resulting in a total of 242 images. These 121 images include patients with myopia (101) and clinically normal (20) images. The age distribution for all subjects combined is 27.24 ± 9.28 (range, 11.0–69.0) years. For normal subjects mean ± SD age distribution is 32.40 ± 17.23 years. Similarly, the myopia age distribution is 26.22 ± 6.37 years.

Manual Marking of Images

Manual marking of all images was done by a trained single clinical expert (JJB). The Cirrus 5000 (Carl Zeiss Meditec Inc., Dublin, CA) provides a cross-sectional image for both horizontal and vertical. The clinician manually drew two straight line aligning the upper boundary RPE layers using a mouse and MS Paint. The boundary line with red-green color used is shown in Figure 1. A caveat should be inserted here - since these boundaries were marked using a mouse it is prone to error because of excessive sliding of the mouse and/or parallax.

Figure 1

ODT images (A) non marked horizontal scanned image, (B) manually marked horizontal scanned image (ground truth), (C) non marked vertical scanned image, and (D) manually marked vertical scanned image.

Segmentation of RPE Boundary

New segmentation algorithms can be developed using the clear images. Manually marked images can then be used to compare the segmentation achieved with new algorithms. For comparisons the boundary method suggested by Gudapati et al,13 or other methods can be used. Gudapati et al,13 for example, introduced various methods and comparisons were then made for each parameter. The source code for the image processing algorithm can be found at .13

Utility and Discussion

Differentiating physiological ODT from a disease involved ODT is clinically important.12 Recent reports have suggested that optic disc imaging with OCT can improve differential diagnosis involving optic nerve head diseases. Creating and making accessible large and real-world datasets has been essential in accelerating public health database research.45 To the best of our knowledge this the first publicly available dataset on optic nerve head cross-sectional imaged with OCT. Detailed calculations of ODT parameters from the ODT dataset has been completed and their results have been published elsewhere.13 These calculations can be used as a reference for future algorithms. The ODTiD database can be divided into training and test sets for application in machine learning/deep learning methods. This database is available for use by researchers and can be downloaded from the ICPSR website at the University of Michigan ().47 In the future additional marked and non-marked images will be included with their detailed characteristics.

Conclusions

This publicly available, open-access OCT images collection will serve as a dataset for use in biomedical image processing. This dataset will be optimal for researchers aiming to develop quantitative relationships between ODT and pathological conditions such as myopia.