• Users Online: 123
  • Home
  • Print this page
  • Email this page
Home About us Editorial board Ahead of print Current issue Search Archives Submit article Instructions Subscribe Contacts Login 


 
 Table of Contents  
ORIGINAL ARTICLE
Year : 2019  |  Volume : 4  |  Issue : 2  |  Page : 85-89

Quantification of retinal nerve fiber layer changes in optical coherence tomography images reveals differential progression of glaucomas


1 Department of Biostatistics, Public Health Dynamics Laboratory (PHDL), Graduate School of Public Health, Pittsburgh, PA, USA
2 Department of Clinical Epidemiology and Biostatistics, LV Prasad Eye Institute, Hyderabad, Telangana, India
3 Department of Pharmacy and Therapeutics, Program Evaluation and Research Unit, University of Pittsburgh School of Pharmacy, Pittsburgh, PA, USA
4 Department of Glaucoma, Narayana Nethralaya, Bengaluru, Karnataka, India

Date of Submission27-Nov-2019
Date of Acceptance06-Dec-2019
Date of Web Publication09-Jan-2020

Correspondence Address:
Dr. Saumyadipta Pyne
Department of Biostatistics, Graduate School of Public Health, A 718, Public Health, 130 De Soto Street, Pittsburgh, PA 15261
USA
Login to access the Email id

Source of Support: None, Conflict of Interest: None


DOI: 10.4103/bjhs.bjhs_74_19

Rights and Permissions
  Abstract 


BACKGROUND: For monitoring the progression of glaucoma, a major cause of irreversible blindness, clinicians measure retinal nerve fiber layer (RNFL) changes in the eye.
MATERIALS AND METHODS: Based on a clinical cohort of patients, we computed the temporal differences in the RNFL patterns in their glaucomatous eyes using optical coherence tomography (OCT) images. To gain insights into disease progression, we quantified the precise changes in the RNFL clock-hour sector phenotypes in each glaucomatous eye between the first and second clinical visits.
CONCLUSION: Further, we identified 2 groups of patients using unsupervised clustering based only on their initial RNFL phenotypes, which may be investigated further to develop clinically useful models for prediction of glaucoma progression.

Keywords: Clustering, disease progression, glaucoma, optical coherence tomography, retinal nerve fiber layer phenotypes, cosine distance


How to cite this article:
Pyne S, Ali MH, Aruru M, Rao HL. Quantification of retinal nerve fiber layer changes in optical coherence tomography images reveals differential progression of glaucomas. BLDE Univ J Health Sci 2019;4:85-9

How to cite this URL:
Pyne S, Ali MH, Aruru M, Rao HL. Quantification of retinal nerve fiber layer changes in optical coherence tomography images reveals differential progression of glaucomas. BLDE Univ J Health Sci [serial online] 2019 [cited 2020 Jul 4];4:85-9. Available from: http://www.bldeujournalhs.in/text.asp?2019/4/2/85/275438



Globally, glaucoma is the most common cause of irreversible blindness and the second leading cause of blindness.[1] Glaucoma is a chronic progressive optic neuropathy associated with the loss of retinal ganglion cells, resulting in permanent visual field (VF) loss and irreversible blindness. As damage to the retinal ganglion cells cannot be detected clinically, clinicians often rely on indirect methods such as changes in the retinal nerve fiber layer (RNFL). Visible, localized RNFL defects are often considered an early sign of glaucoma and are associated with neuronal losses, thereby making these structural observations more relevant in documenting the early changes in glaucoma.[4] RNFL changes in glaucoma may be present with or without VF loss.[3]

Primary open-angle glaucoma (POAG) and primary angle-closure glaucoma (PACG) are the two most common forms of primary glaucoma worldwide. Intraocular pressure (IOP) is the most important risk factor for the development and progression of glaucoma.[1],[2],[3],[4] The two most commonly used diagnostic tests to detect glaucoma and determine its severity are the standard automated perimetry (SAP), which tests the VF loss and the optical coherence tomography (OCT), which evaluates structural changes in the RNFL. OCT is a widely used noninvasive imaging technology that uses low-coherence light to capture in vivo two- and three-dimensional images of the eye at micrometer resolution. Numerous studies have reported excellent diagnostic abilities of OCT-measured RNFL changes to discriminate glaucomatous eyes from normal eyes.[5],[6],[7],[8]

However, perhaps, the most challenging aspect of managing glaucoma is the ability to predict the course of progression of the disease. RNFL evaluation remains by far the most promising protocol for glaucoma detection and monitoring progression. In this direction, the calculation of precise changes in RNFL patterns as captured by longitudinal measurements of OCT could be particularly insightful since glaucoma is a slowly progressing disease. In the present study, based on a clinical cohort, we explore a new computational approach based on the well-known cosine distance to quantify the temporal changes in RNFL patterns across the clock-hour sectors in OCT data in glaucomatous eyes. Further, our unsupervised clustering of the initial RNFL patterns showed its potential utility for distinguishing early the groups of patients that undergo differential disease progression.


  Materials and Methods Top


Data

The dataset was acquired from an ongoing, prospective, longitudinal study at Narayana Nethralaya, a tertiary eye care center in Bengaluru in Southern India. Written informed consent was obtained from all participants, and the study was approved by the Ethics Committee of Narayana Nethralaya. Participants of the study, all of the Indian ancestry, include glaucoma patients (POAG and PACG subtypes), glaucoma suspects who are longitudinally evaluated every 6–12 months with SAP, optic disc stereo photographs, and OCT for the detection of progression.

POAG patients had open angles on gonioscopy and glaucomatous changes on optic nerve head (ONH) examination (neuroretinal rim narrowing, notching, and RNFL defects) as documented by glaucoma experts on dilated examination and confirmed by the masked evaluation of stereoscopic optic disc photographs by graders. PACG patients had occludable anterior chamber angles on gonioscopy, presence of goniosynechiae, a history of IOP >21 mm Hg, and had undergone laser peripheral iridotomy before the study inclusion. The anterior chamber angle was examined using an indentation gonioscope and was considered occludable if, in primary position, the posterior trabecular meshwork was not seen in three or more quadrants.[9] Pretreatment IOP or VF loss was not used to define POAG and PACG.

The inclusion criteria for all participants included age ≥18 years, corrected distance visual acuity of 20/40 or better, and refractive error within ± 5 D sphere and ± 3 D cylinder. The exclusion criteria included the presence of any media opacities that prevented good quality OCT scans or any retinal or neurological disease other than glaucoma, which could confound the evaluation. All participants underwent a comprehensive ocular examination, which included corrected distance visual acuity measurement, slit-lamp biomicroscopy, Goldmann applanation tonometry, gonioscopy, dilated fundus examination, stereoscopic optic disc photography, VF examination, and OCT imaging with Cirrus HD-OCT (model 5000, Carl Zeiss Meditec Inc., Dublin, CA, USA) every 6–12 months.

OCT examination was conducted using 200 × 200 ONH cube scan protocol on Cirrus HD-OCT which uses a superluminescent diode laser with a center wavelength of 840 nm at an axial resolution of 5 μm. The ONH protocol is based on a tridimensional scan of a 6 × 6 mm2 area centered on the optic disc where information from a 1024 (depth) × 200 × 200-point parallelepiped is collected. RNFL thickness was calculated along a circle 3.46 mm in diameter, positioned evenly around the center of the optic disc. The calculation circle was also divided into smaller 12 clock-hours sectors of 30° each, and RNFL thickness in each of the 12 clock-hours sectors was measured. Image quality was assessed for all OCT scans. Poor quality images, which were defined as those with signal strength <6 or images with motion artifacts and segmentation errors, were excluded from the analysis.

Thus, for the present study, we obtained OCT RNFL data for glaucomatous eyes of n = 29 patients, along with their clinical covariates. The cohort consisted of 6 (21%) female and 23 (79%) male patients with 18 (68%) POAG and 11 (32%) PACG subtypes. Their baseline clinical measurements are described in [Table 1].
Table 1: Baseline measurements of the clinical variables of the study cohort (n=29)

Click here to view


Analysis

In the present study, we used cosine distance, an known method in linear algebra for measuring distances between a pair of vectors, to compute the differences in RNFL thickness phenotypes {ai, j, t} for the ith glaucomatous eye {i = 1… N} across the 12 OCT clock-hour sectors j {j = 1…12} between the first and second clinical visits at time-points t {t = 1,2}. Note that ai, j, t 0 for all values of i, j, t.

The cosine distance (dcos) between a given pair of eyes e, e' (samples indexed i and i' respectively), in terms of their OCT clock-hour phenotypes {ai, j, t | j = 1…12} and {ai', j', t'| j = 1…12} measured during time-points t and t', is defined as follows:



Thus, at a fixed time point the distance between two eyes e, e' is given by



and the distance for a fixed eye across two time points t and t':



For unsupervised learning of groups in RNFL clock-hour sector phenotype data, we used agglomerative hierarchical clustering (implemented in R by the “hclust” function) using the matrix of cosine distances {dcos (et, et')} between the RNFL phenotypes of the glaucomatous eyes (e, e') of each pair of (n = 29) patients as measured by OCT during their first clinical visit (at time t).

In addition, we computed the intervisit phenotypic difference in terms of the cosine distance D = dcos (et, et,) between the RNFL clock-hour sector data for each patient's glaucomatous eye e as measured by OCT during the first and the second clinical visits at time points t and t' respectively. The mean duration between these visits is 10.01 months (SD = 7.69). The median (M = 0.059) of these (n = 29) values of D were used to define two groups of patients: (i) “LowDiff” consisting of those with intervisit difference of RNFL phenotypes D < M and (ii) “Highdiff” containing the rest, i.e., for D ≥ M. Illustrative examples of intervisit RNFL phenotypic differences are shown in [Figure 1].
Figure 1: Inter-visit changes in retinal nerve fiber layer clock hour sector phenotypes: The radar plots show the examples of low (Sample 13) and relatively high (Sample 2) levels of phenotypic change between the optical coherence tomography scans of glaucomatous eyes taken during the first (green area) and second (red area) clinical visits. The changes are measured by cosine distances, which are 0.03 and 0.108 for the Samples 13 and 2, respectively. The 12 clock hour sectors are marked on the circular axis

Click here to view



  Results Top


To gain insight into disease progression, each patient's inter-visit RNFL phenotypic change was calculated with cosine distance D = dcos (et, et') as described in the methods section. This allowed us to the identify groups with low and high intervisit differences, namely LowDiff and HighDiff. Interestingly, we observed statistical variations in certain features of their glaucomatous eyes. For instance, the distributions of the baseline optic disc diameter measurements of these two groups while overlapping have distinct shapes and modes [Figure 2].
Figure 2: Optic disc diameter distributions. The distinct density curves of the baseline disc diameter measurements of the glaucomatous eyes from the LowDiff and the HighDiff groups are shown in blue and red, respectively

Click here to view


This motivated us to ask the question as to whether the existence of groups of patients who may go on to progress in their disease differently could be predicted with early signatures in their RNFL phenotypes as captured in OCT data from their first clinical visit. To address this problem, we conducted unsupervised learning of patient groups using agglomerative hierarchical clustering of the “early” RNFL clock-hours phenotypes of the glaucomatous eyes based only on their first OCT data and using no other covariates. Notably, the clustering dendrogram based on cosine distances between every pair of glaucomatous eyes in our data reveals the clear existence of two clusters of early RNFL phenotypes (as shown in green and orange) in [Figure 3].
Figure 3: Hierarchical clustering of the glaucomatous eyes: Based on the cosine distances of the initial retinal nerve fiber layer phenotypes, the dendrogram reveals two main groups, namely clusters 1 (green) and 2 (orange). The numeric sample ID's are shown below. The samples marked with asterisks are primary angle closure glaucoma and those without are primary open angle glaucoma. Each sample's membership in LowDiff/HighDiff group (this information was not used for clustering) was noted with blue/red ID, respectively

Click here to view


Interestingly, we noted a statistically significant overlap between the two clusters identified in an unsupervised analysis based only on the early phenotypes, and the two groups of patients defined as HighDiff and LowDiff based on disease progression. Fisher's exact test [with 2 × 2 table shown in [Figure 4] gives the P value (for odds of HighDiff for cluster 2) as 0.045. In contrast, the same test showed no statistically significant overlap between the same two clusters and the two subtypes of glaucoma patients (POAG and PACG).
Figure 4: Contingency table: The counts of samples in the table show the overlap among the early phenotypic clusters and the groups with differential disease progression

Click here to view






Glaucoma affects 1-in-200 individuals aged 50 years and younger and 1-in-10 individuals over the age of 80 years. It is estimated that 11 million individuals will be blind from glaucoma and a total of 80 million individuals will be affected by glaucoma by the year 2020.[2] It is, therefore, of critical importance to develop accurate and rigorous statistical models that can predict the progression of glaucoma for a given patient as early as possible following his/her diagnosis. In the present study involving the comparison of changes in RNFL phenotypes in glaucomatous eyes, our computational approach to quantify the changes in the clock-hour sector patterns illustrates the potential use of such objective means to capture the diversity in complex phenotypes and obtain early insights into disease progression.

In our future work, we plan to extend the power of our study with larger samples sizes and more time points to further explore these findings. It might be possible to validate our quantified differences in OCT RNFL phenotypes in glaucoma against genotypic and transcriptomic variations in a larger cohort. With more time points, we can study the effects of duration of time on RNFL phenotypic changes as well as the variability of such measurements across visits. Finally, given the inherent richness of the OCT data, we believe that new statistical models that can address the rich spatio-temporal dynamics and the complexity in the circular space of intraocular phenotypes need to be developed.

Acknowledgement

We thank Dr. Sirisha Senthil for technical advice.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.



 
  References Top

1.
Patel NB, Sullivan-Mee M, Harwerth RS. The relationship between retinal nerve fiber layer thickness and optic nerve head neuroretinal rim tissue in glaucoma. Invest Ophthalmol Vis Sci 2014;55:6802-16.  Back to cited text no. 1
    
2.
Quigley HA, Broman AT. The number of people with glaucoma worldwide in 2010 and 2020. Br J Ophthalmol 2006;90:262-7.  Back to cited text no. 2
    
3.
Sommer A, Katz J, Quigley HA, Miller NR, Robin AL, Richter RC, et al. Clinically detectable nerve fiber atrophy precedes the onset of glaucomatous field loss. Arch Ophthalmol 1991;109:77-83.  Back to cited text no. 3
    
4.
Choi JA, Park HY, Shin HY, Park CK. Optic disc characteristics in patients with glaucoma and combined superior and inferior retinal nerve fiber layer defects. JAMA Ophthalmol 2014;132:1068-75.  Back to cited text no. 4
    
5.
Rao HL, Addepalli UK, Yadav RK, Choudhari NS, Senthil S, Garudadri CS. Factors affecting the ability of the spectral domain optical coherence tomograph to detect photographic retinal nerve fiber layer defects. PLoS One 2014;9:e116115.  Back to cited text no. 5
    
6.
Rao HL, Yadav RK, Addepalli UK, Chaudhary S, Senthil S, Choudhari NS, et al. Peripapillary retinal nerve fiber layer assessment of spectral domain optical coherence tomography and scanning laser polarimetry to diagnose preperimetric glaucoma. PLoS One 2014;9:e108992.  Back to cited text no. 6
    
7.
Rao HL, Babu JG, Addepalli UK, Senthil S, Garudadri CS. Retinal nerve fiber layer and macular inner retina measurements by spectral domain optical coherence tomograph in Indian eyes with early glaucoma. Eye (Lond) 2012;26:133-9.  Back to cited text no. 7
    
8.
Rao HL, Yadav RK, Addepalli UK, Chaudhary S, Senthil S, Choudhari NS, et al. Retinal nerve fiber layer evaluation of spectral domain optical coherence tomograph and scanning laser polarimeter to diagnose glaucoma. Eye (Lond) 2014;28:654-61.  Back to cited text no. 8
    
9.
Weinreb RN, Friedman DS, editors. Angle Closure and Angle Closure Glaucoma: Reports and Consensus Statements of the 3rd Global AIGS Consensus Meeting on Angle Closure Glaucoma. The Hague, The Netherlands: Kugler Publications; 2006. p. 1-20.  Back to cited text no. 9
    


    Figures

  [Figure 1], [Figure 2], [Figure 3], [Figure 4]
 
 
    Tables

  [Table 1]



 

Top
 
 
  Search
 
Similar in PUBMED
 Related articles
Access Statistics
Email Alert *
Add to My List *
* Registration required (free)

 
  In this article
Abstract
Materials and Me...
Results
Discussion
References
Article Figures
Article Tables

 Article Access Statistics
    Viewed260    
    Printed2    
    Emailed0    
    PDF Downloaded56    
    Comments [Add]    

Recommend this journal


[TAG2]
[TAG3]
[TAG4]