Abstract
Objective
To compare the abilities of a new Glaucoma Probability Scoring (GPS) system and Moorfields regression analysis (MRA) to differentiate between glaucomatous and normal eyes using Heidelberg retinal tomograph (HRT)‐III software and race‐specific databases.
Methods
In this prospective study, one eye (refractive error ⩽5 D) each of consecutive normal patients and those with glaucoma was enrolled. All patients underwent a full eye examination, standard achromatic perimetry (Swedish Interactive Threshold Algorithm‐standard automated perimetry (SITA‐SAP), program 24‐2) and confocal scanning laser ophthalmoscopy (HRT‐II) within 1 month. Normal patients had two normal visual fields in both eyes (pattern standard deviation (PSD) >5% and Glaucoma Hemifield Test within 97% normal limits) and a normal clinical examination. Glaucoma was defined on the basis of SITA‐SAP visual field loss (PSD<5% or Glaucoma Hemifield Test outside normal limits) on two consecutive visual fields. HRT‐II examinations were exported to the HRT‐III software (V.3.0), which uses an enlarged race‐specific database, consisting of 733 eyes of white people and 215 eyes of black people. Race‐adjusted MRA for the most abnormal sector (operator‐dependent contour line placement) was compared with the global race‐adjusted GPS (operator independent). MRA sectors outside the 99.9% confidence interval limits (outside normal limits) and GPS ⩾0.64 were considered abnormal.
Results
136 normal patients (72 black and 64 white patients) and 84 patients with glaucoma (52 black and 32 white patients) were enrolled (mean age 50.4 (SD 14.4) years). The average visual field mean deviation was −0.4 (SD 1.1) db for the normal group and −7.3 (SD 6.7) db for the glaucoma group (p<0.001). Mean GPS values were 0.21 (SD 0.23) and 0.73 (SD 0.27) for normal and glaucomatous eyes, respectively (p<0.001). Sensitivity and specificity values were 77.1% and 90.3% for GPS, and 71.4% and 91.9% for MRA, respectively.
Conclusions
In this cohort, GPS software sensitivity and specificity values are similar to those of MRA, which requires placement of an operator‐dependent contour line. The development of software to detect glaucoma without a contour line is critical to improving the potential use of HRT as a tool for glaucoma detection and screening.
Primary open‐angle glaucoma is a leading cause of blindness worldwide.1 Glaucomatous changes to the optic nerve head and retinal nerve fibre layer (RNFL) loss often precede achromatic visual field defects, which may become manifest only after a large percentage of retinal ganglion cells have been damaged.2,3 As such, early disease detection may be important in disease management strategies. Confocal scanning laser ophthalmoscopy has become an important tool for detecting structural damage of the optic nerve head and RNFL, and may assist in early glaucoma detection.4,5 There seem to be important racial differences between people of African and European ancestry regarding optic disc configuration6,7,8,9,10,11 and neuroretinal rim area.12,13 Given these variations, questions exist as to whether the current Heidelberg Retinal Tomograph (HRT)‐II database of people of European ancestry can be applied to other racial groups.10
Heidelberg retinal tomography (Heidelberg Engineering, GmBH, Dossenheim, Germany) uses confocal scanning laser technology to calculate topographic measurements of the optic nerve and parapapillary RNFL. One method of the HRT‐II software analysis, Moorfields regression analysis (MRA), uses an algorithm to compare measured optic nerve parameters with those from a normative database.9 Although there is good reproducibility of HRT measurements,14,15,16 a major limitation of this device is the need for an operator to draw a contour line at the border of the optic disc. This can result in variability in measurements between different observers.17 The HRT‐II uses a normative database consisting of 349 normal eyes of white people, whereas HRT‐III uses an enlarged race‐specific database, consisting of eyes of 733 white and 215 black people (Heidelberg Engineering, personal communication, Heidelberg Engineering, 2006).
HRT‐III software V.3.0 includes the calculation of the Glaucoma Probability Score (GPS), a new, automated algorithm that evaluates both optic disc and parapapillary RNFL topography to estimate the probability of having glaucoma. The GPS uses two measures of parapapillary RNFL shape (horizontal and vertical RNFL curvature) and three measures of optic nerve head shape (cup size, cup depth and rim steepness) for input into a vector machine‐learning classifier that estimates the probability of having damage consistent with glaucoma. No contour line or reference plane is used in the GPS calculation, and therefore the analysis is operator independent. This is based on mathematical modelling of the optic nerve shape, which typically exhibits a cup with varying width and depth, as well as curvature of the rim region.18 The normally convex RNFL curvature, caused by the ganglion cell axons converging towards the optic nerve, flattens as axons are lost as a result of glaucoma.
In the present study, we compared the abilities of GPS and MRA to differentiate between glaucomatous and normal eyes using HRT‐III software and race‐specific databases.
Methods
Institutional Review Board approval was obtained and all participants provided signed informed consent. Normal people, those suspected of having glaucoma and patients with glaucoma aged ⩾18 years were enrolled. All participants had a best‐corrected visual acuity of at least 20/40 in the study eye, spherical refractive error ⩽5.0 D and cylinder correction ⩽3.0 D. Patients with anatomically narrow angles, retinal disease, considerable ocular surface disease, non‐glaucomatous optic neuropathy or a history of intraocular surgery other than uncomplicated cataract surgery were excluded. If both eyes were eligible, one eye of each participant was randomly enrolled.
All participants underwent a complete ophthalmic examination, which included a review of relevant medical history, best‐corrected visual acuity, slit‐lamp biomicroscopy, gonioscopy, Goldmann applanation tonometry, dilated funduscopy, stereoscopic ophthalmoscopy of the disc using a 78‐D lens, and simultaneous stereoscopic photographs of the optic disc. Two baseline standard achromatic perimetry examinations (Swedish Interactive Threshold Algorithm‐standard automated perimetry, program 24‐2, Humphrey Field Analyzer II; Carl Zeiss Meditec, Dublin, California, USA) and confocal scanning laser ophthalmoscopy (Heidelberg Retinal Tomograph II; HRT‐II Heidelberg Engineering) were carried out within 1 month of enrolment. All participants had reliable visual field test results, defined as <33% false‐positive errors, false‐negative errors and fixation losses.
For HRT‐II imaging, a mean topographic image was automatically obtained from three scans (10° field of view), centred on the optic disc using HRT‐II software V.1.4.1. Corneal curvature measurements were recorded using keratometry to correct for magnification error. Good image quality was defined as follows: acquisition sensitivity <90%, topography standard deviation (SD) <40 μm, more than three quarters of the disc within the target circle, minimal movement during the acquisition movie, no floaters over the disc, and good imaging clarity and exposure. A trained technician, relying on stereophotographs of the respective optic disc, outlined the optic disc margin on the mean topographic image.
Racial groups were defined by self‐report. Glaucoma was defined by the existence of reproducible standard automated perimetry loss (pattern standard deviation <5% or Glaucoma Hemifield Test outside normal limits) on two consecutive fields. Participants were considered to be normal if both visual fields in both eyes were unremarkable (pattern standard deviation >5% and Glaucoma Hemifield Test within 97% normal limits) and the clinical examination was normal. HRT‐II data results were exported to the HRT‐III software (V.3.0) and the appropriate racial database was selected before analysis. Moorfields regression analysis (MRA) for the most abnormal sector Moorfields Classification (MFC) result (operator dependent) was compared with the global GPS (operator independent). Sectors outside the 99.9% confidence interval limits (outside normal limits) were determined to be abnormal. Sectors declared normal or borderline were considered to be normal. As per the manufacturer's recommendations, a GPS of ⩾0.64 was considered to be outside normal limits with an approximate sensitivity and specificity of 70% and 90%, respectively.
Data were statistically analysed using JMP (SAS Institute, Cary, North Carolina, USA). Student's t test was used to compare the distributions of continuous variables between groups. Values of p<0.05 were regarded as significant. Sensitivities and specificities were calculated for race‐adjusted MRA‐III MFC result (result of the worst sector) and for global GPS results.
Results
In all, 136 (72 black and 64 white) normal patients and 84 (52 black and 32 white) patients with glaucoma were enrolled. Three participants were excluded from the final analysis because the software failed to calculate GPS values. Mean (SD) age was 50.4 (14.4) years (45.5 (13.6) and 58.5 (11.8) years for normal patients and for those with glaucoma, respectively; p<0.001; table 1). The mean (SD) visual field mean deviation was −0.40 (1.05, range 2.00 to −2.84) db for normal eyes and −7.31 (6.66, range 1.22 to −30.51) db for glaucomatous eyes (p<0.001). The mean (SD) visual field pattern was 1.44 (0.25, range 0.99–1.99) db for normal eyes and 6.58 (3.85, range 1.66–17.19) db for glaucomatous eyes (p<0.001).
Table 1 Patient demographics and mean values.
| Normal | Glaucoma | p Value* | |
|---|---|---|---|
| Total = n (W, B) | 134 (62 W, 72 B) | 83 (31 W, 52 B) | |
| Age, years† | 45.5 (13.6) | 58.5 (11.8) | <0.001 |
| Visual field mean deviation, db† | −0.40 (1.05) | −7.31 (6.66) | <0.001 |
| Range | 2.00 to −2.84 | 1.22 to −30.51 | |
| Visual field pattern standard deviation, db† | 1.44 (0.25) | 6.58 (3.85) | <0.001 |
| Range | 0.99 to 1.99 | 1.66 to 17.19 | |
| Global GPS%† | 21.25 (23.36) | 73.19 (27.11) | <0.001 |
B, black patients; GPS, Glaucoma Probability Score; W, white patients.
*p Value by Student's t test.
†Values are mean (SD).
The GPS model could not be calculated in 3 of the 220 patients (two normal patients and one with glaucoma); these patients were excluded from the final analysis. GPS fails to calculate the model as a result of absence of an optic cup or due to very small disc size, which was seen in two of our patients in whom disc sizes were 0.872 and 0.953 mm. The HRT‐III MFC result analysis detected glaucoma in 59 of the 83 glaucomatous eyes, yielding a sensitivity of 71.1%; it confirmed normal tests in 123 of 134 normal eyes, yielding a specificity of 91.8%. Global GPS analysis detected glaucoma in 64 of the 83 glaucomatous eyes, yielding a sensitivity of 77.1%, and confirmed normal tests in 121 of 134 normal eyes, yielding a specificity of 90.3% (table 2).
Table 2 Sensitivity and specificity of Moorfields and Glaucoma Probability Score classifications.
| MFC result HRT‐III | Global GPS | |
|---|---|---|
| Sensitivity | 71.1% (59/83) | 77.1% (64/83) |
| Sensitivity for early glaucoma | 59.6% (28/47) | 72.3% (34/47) |
| Specificity | 91.8% (123/134) | 90.3% (121/134) |
GPS, Glaucoma Probability Score; MFC, Moorfields classification.
MFC Result HRT‐III: the worst sector in the race‐adjusted MFC analysis.
Global race‐adjusted GPS: Early glaucoma, mean deviation ⩽−5 db.
Further analysis was carried out on a subgroup of 47 patients with early glaucoma, defined as mean deviation on visual field ⩽−5 db. As shown in table 2, sensitivity levels decreased in early glaucoma for both MRA and GPS (59.6% and 72.3%, respectively).
Discussion
Computerised optic nerve imaging is often used to augment traditional methods of optic disc examination in the management of glaucoma. The HRT ancillary study to the Ocular Hypertension Study proved the ability of confocal laser ophthalmoscopy to identify early glaucomatous injury and predict the conversion from ocular hypertension to glaucoma using the Ocular Hypertension Study criteria.5 One limitation of HRT technology has been the need to place an operator‐dependent contour line at the border of the optic disc before computerised analysis. Iester et al17 reported that even in ideal circumstances, outlining the optic disc with the aid of optic disc photographs, intraobserver variability cannot be entirely eliminated.
We assessed the ability of the new HRT‐III software to differentiate between normal and glaucomatous eyes. This software takes advantage of an enlarged race‐specific database and a machine‐learning classifier to help better identify glaucoma. GPS without the use of an operator‐dependent contour line successfully differentiated glaucomatous eyes from normal ones, with sensitivity and specificity levels comparable to those in MRA.
GPS was better at detecting patients with early glaucoma (mean deviation ⩽−5 db), with a sensitivity of 72.3% compared with 59.6% for MRA. As it is difficult to place the contour line accurately, particularly in a fast‐paced general ophthalmology setting, sensitivity and specificity of MRA in clinical practice are likely to be less than we reported herein, and thus the benefits of an operator‐independent system such as the GPS will be more evident.
A global parameter of the GPS performed as well or even better than any abnormal individual MRA sector for detecting glaucoma. Interestingly, a global parameter was sufficient for discriminating healthy and glaucomatous eyes; the diagnostic accuracy of these algorithms should be tested in future studies.
In summary, the GPS of the HRT‐III represents an alternative strategy for diagnosing glaucoma, with a considerable advantage over the MRA in early stages of the disease. The GPS bypasses the “Achilles heel” of the HRT, which is the need for an operator‐dependent contour line. Automated image analysis free of contour line placement suggests that the HRT‐III may have a role in glaucoma detection and screening.
Abbreviations
GPS - Glaucoma Probability Score
HRT - Heidelberg retinal tomograph
MFC - Moorfields classification
MRA - Moorfields regression analysis
RNFL - retinal nerve fibre layer
Footnotes
Funding: This study was supported in part by the Shelley and Steven Einhorn Research Fund of the New York Glaucoma Research Institute, New York, USA.
Competing interests: None.
Presented in part at the Annual Meeting of the Association for Research in Vision and Ophthalmology, Fort Lauderdale, Florida, USA, May 2006.
References
- 1.Resnikoff S, Pascolini D, Etya'ale D.et al Global data on visual impairment in the year 2002. Bull World Health Organ 200482887–888. [PMC free article] [PubMed] [Google Scholar]
- 2.Quigley H A, Katz J, Derick R J.et al An evaluation of optic disc and nerve fiber layer examinations in monitoring progression of early glaucoma damage. Ophthalmology 19929919–28. [DOI] [PubMed] [Google Scholar]
- 3.Harwerth R S, Carter‐Dawson L, Shen F.et al Ganglion cell losses underlying visual field defects from experimental glaucoma. Invest Ophthalmol Vis Sci 1999402242–2250. [PubMed] [Google Scholar]
- 4.Wollstein G, Garway‐Heath D F, Hitchings R A. Identification of early glaucoma cases with the scanning laser ophthalmoscope. Ophthalmology 19981051557–1563. [DOI] [PubMed] [Google Scholar]
- 5.Zangwill L M, Weinreb R N, Beiser J A.et al Baseline topographic optic disc measurements are associated with development of primary open angle glaucoma. The Confocal Scanning Laser Ophthalmoscopy Ancillary Study to the Ocular Hypertension Treatment Study. Arch Ophthalmol 20051231188–1197. [DOI] [PubMed] [Google Scholar]
- 6.Beck R W, Messner D K, Musch D C.et al Is there a racial difference in physiologic cup size? Ophthalmology 198592873–876. [DOI] [PubMed] [Google Scholar]
- 7.Chi T, Ritch R, Stickler D.et al Racial differences in optic nerve parameters. Arch Ophthalmol 1989107836–839. [DOI] [PubMed] [Google Scholar]
- 8.Tsai C S, Zangwill L, Gonzalez C.et al Ethnic differences in optic nerve head topography. J Glaucoma 19954248–257. [PubMed] [Google Scholar]
- 9.Wollstein G, Garway‐Heath D F, Hitchings R A. Identification of early glaucoma cases with the scanning laser ophthalmoscope. Ophthalmology 19981051557–1563. [DOI] [PubMed] [Google Scholar]
- 10.Girkin C A, McGwin G, Xie A.et al Differences in optic disc topography between black and white normal subjects. Ophthalmology 200511233–39. [DOI] [PubMed] [Google Scholar]
- 11.Racette L, Boden C, Klienhandler S L.et al Differences in visual function and optic nerve structure between healthy eyes of blacks and whites. Arch Ophthalmol 20051231547–1553. [DOI] [PubMed] [Google Scholar]
- 12.Zangwill L M, Weinreb R N, Berry C C.et al Racial differences in optic disc topography. Arch Ophthalmol 200412222–28. [DOI] [PubMed] [Google Scholar]
- 13.Poinoosawmy D, Fontana L, Wu J X.et al Variation of nerve fiber layer thickness measurements with age and ethnicity by scanning laser polarimetry. Br J Ophthalmol 199781350–354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Miglior S, Albe E, Guareschi M.et al Intraobserver and interobserver reproducibility in the evaluation of optic disc stereometric parameters by Heidelberg retina tomograph. Ophthalmology 20021091072–1077. [DOI] [PubMed] [Google Scholar]
- 15.Rohrschneider K, Kruse F E. Reproducibility of retinal nerve fiber layer evaluation by dynamic scanning laser ophthalmoscopy. Am J Ophthalmol 1995119666–667. [DOI] [PubMed] [Google Scholar]
- 16.Orgül S, Cioffi G A, Van Buskirk E M. Variability of contour line alignment on sequential images with the Heidelberg retina tomograph. Graefes Arch Clin Exp Ophthalmol 199723582–86. [DOI] [PubMed] [Google Scholar]
- 17.Iester M, Mikelberg F S, Courtright P.et al Interobserver variability of optic disk variables measured by confocal scanning laser tomography. Am J Ophthalmol 200113257–62. [DOI] [PubMed] [Google Scholar]
- 18.Swindale N V, Stjepanovic G, Chin A.et al Automated analysis of normal and glaucomatous optic nerve head topography images. Invest Ophthalmol Vis Sci 2000401730–1742. [PubMed] [Google Scholar]