Purpose
To assess the determinants of image quality of Heidelberg Retina Tomography II (HRT II) and its association with optic disc parameters in a population-based setting.
Design
Population-based, cross-sectional study involving 3280 (78.7% response) Asian Malays aged 40 to 80 years.
Methods
Three thousand fifty-six participants completed the HRT II test. Image quality was assessed using the mean pixel height standard deviation generated by the HRT II, with lower mean pixel height standard deviation indicating higher quality. Participants underwent an interviewer-administered questionnaire and a standardized ophthalmic examination, including visual acuity, applanation tonometry, gonioscopy, refraction, automated perimetry, and lens and fundus photography.
Results
The mean (standard deviation) and median of mean pixel height standard deviation in the study population were 34 (34) and 23 μm respectively. In multivariate regression models, older age and the presence of visual impairment, blindness, high myopia, and cataract were significantly associated with greater mean pixel height standard deviation ( P < .05 for all). People with a higher mean pixel height standard deviation were more likely to have smaller rim area and greater cup depth.
Conclusion
People who are older or have high myopia, visual impairment, blindness, or cataract are more likely to have poor HRT II image quality. Poorer image quality is in turn associated with smaller optic rim area and greater cup depth. These data provide useful information when the HRT instrument is used for diagnosing glaucoma in the general population.
The Heidelberg Retinal Tomograph (HRT, Heidelberg Engineering GmbH, Dossenheim, Germany) is an imaging tool that provides quantitative measurements of optic disc parameters. HRT is commonly used to aid the diagnosis and monitoring of progression of glaucomatous optic neuropathy. The quality of HRT images is an essential factor for reliable interpretation of optic disc measurements (eg, optic rim area). According to the manufacturer’s guidelines, the HRT image quality is assessed by the mean pixel height standard deviation, which is defined as the standard deviation for the height measurement of each pixel for three consecutive images. A higher mean pixel height standard deviation indicates poorer HRT image quality.
Two clinically relevant issues remain unanswered regarding HRT image quality. First, the factors associated with poor HRT image quality are not fully understood. Previous clinic-based studies in selected subjects have shown that dense cataracts, advanced age, small pupil size, and increased degree of astigmatism are associated with poorer HRT image quality. However, whether these influences apply more broadly in a general population setting is uncertain. It is also unclear whether socioeconomic status, systemic diseases (eg, diabetes mellitus), and ocular factors (eg, high myopia) could affect HRT image quality. Second, previous studies have used various mean pixel height standard deviation cutoff points (eg, 20 μm, 25 μm, 30 μm, 40 μm, and 50 μm ) for image quality, and there has not been a consensus on the minimum mean pixel height standard deviation value for an “acceptable” HRT image. Although mean pixel height standard deviation is a known contributor to the test-retest variability of optic disc measurements, it remains unclear to what extent the mean pixel height standard deviation could affect optic disc measurements.
The purpose of this study was twofold: 1) to examine the influences of a broad range of systemic and ocular factors on HRT image quality as assessed by mean pixel height standard deviation; and 2) to examine the relationship of mean pixel height standard deviation with optic disc parameters in a population-based study of adults aged 40 to 80 years.
Subjects and Methods
Population Sampling
The Singapore Malay Eye Study (SiMES) examined 3280 (78.7% response rate) persons of Malay ethnicity aged 40 to 80 years between August 1, 2004 and June 30, 2006. Study methods have been published elsewhere, a summary of which is provided below.
Ophthalmic Examination
Presenting visual acuity (VA) was measured using a logarithm of the minimal angle of resolution (logMAR) number chart (Lighthourse International, New York, New York, USA) at a distance of 4 m. Visual impairment was defined as presenting VA worse than 20/40 but better than 20/200 and blindness as VA of 20/200 or worse. Intraocular pressure (IOP) was measured using a Goldmann applanation tonometer (Haag-Streit, Bern, Switzerland) prior to pupil dilation. Refraction was measured using an autorefractor (Canon RK-5 Auto Ref-Keratometer; Canon Inc Ltd, Tokyo, Japan) and finally determined by subjective refraction by an optometrist. Myopia was defined as spherical equivalent (SE) less than -0.5 diopter (D), high myopia as SE equal to or less than -5 D, astigmatism as cylinder less than -0.5 D. Lens photographs were graded according to the Wisconsin cataract grading system. Nuclear cataract was defined when nuclear opacity was at least as great as the standard 4 photograph. Cortical cataract was defined when cortical opacity involved at least 5% of the total lens area and posterior subcapsular cataract (PSC) was defined when opacity comprised at least 1% of the total lens area.
Digital fundus photography (Canon CR-DGi with a 10D SLR back, Canon, Japan) was taken after pupil dilation. Diabetic retinopathy was graded according to the modified Airlie House classification system. Retinopathy was considered present if any of the following lesions were present: microaneurysms, hemorrhages, cotton-wool spots, intraretinal microvascular abnormalities, hard exudates, venous beading, and new vessels. Age-related macular degeneration (AMD) was graded according to the Wisconsin Age-Related Maculopathy grading system. Early AMD was defined as either soft indistinct or reticular drusen or soft distinct drusen with retinal pigment epithelium abnormalities. Late AMD was defined as the presence of either neovascular AMD or geographic atrophy. The presence of AMD, defined as the presence of either early or late AMD, was used for our statistical analyses.
Vertical cup-to-disc ratio (VCDR) was measured using a +78-D lens at 16× magnification with a measuring graticule (Haag-Streit, Bern, Switzerland). Automated perimetry (SITA FAST 24-2 program, Humphrey Visual Field Analyzer II; Carl Zeiss Meditec, Dublin, California, USA) was performed with near refractive correction on every glaucoma suspect (defined below) and one in five continuous participants without suspected glaucoma (n = 641) before the ophthalmic examination. The visual field test was repeated on another occasion without pupil dilation if the test reliability criteria were not satisfied (fixation losses >20%, false positives >33% and/or false negatives >33%) or if there was a visual field defect. Glaucomatous visual field defect was defined if the following were found: 1) glaucoma hemifield test (GHT) outside normal limits; and 2) a cluster of three or more non-edge, contiguous points on the pattern deviation plot, not crossing the horizontal meridian with a probability of <5% being present in age-matched normals, present on two separate occasions. Glaucoma suspects were defined as those participants with any of the following criteria: 1) IOP greater than 21 mm Hg; 2) VCDR >0.6 or VCDR asymmetry >0.2; 3) abnormal anterior segment signs consistent with pseudoexfoliation or pigment dispersion syndrome; 4) “occludable” angle, defined as posterior trabecular meshwork seen for no more than 180 degrees of the angle circumference during static gonioscopy; 5) peripheral anterior synechiae or other findings consistent with secondary glaucoma; and 6) known history of glaucoma. The “glaucoma suspects” underwent gonioscopy, visual field testing, and a repeated IOP measurement, usually on another day. “Glaucoma cases” were defined according to the International Society of Geographical & Epidemiological Ophthalmology (ISGEO) criteria. Final definition, adjudication, and classification of glaucoma cases were reviewed by the senior author (T.A.).
Heidelberg Retinal Tomograph II Imaging
Heidelberg Retinal Tomograph II (HRT II, Heidelberg Engineering GmbH, Dossenheim, Germany) employs a diode laser (670-nm wavelength). A three-dimensional image is yielded as a series of 16 to 64 consecutive and equidistant two-dimensional optic section images. Each of the two-dimensional images includes 384 × 384 picture elements. Each HRT output is coupled with a mean pixel height standard deviation to reflect the image quality. According to the manufacturer’s guideline, an HRT image with a mean pixel height standard deviation <10 μm would be classified as “excellent”; 10 to 20, “very good”; 20 to 30, “good”; 30 to 40, “acceptable”; 40 to 50, “look for ways to improve”; >50, “low quality image and not recommended as a baseline image.” In our study, the HRT II imaging was operated by two trained technicians after pupil dilation. Corneal curvature radius was entered in the HRT II software and cylindrical lens was adapted for those who had astigmatism greater than or equal to 1.0 D. After completion of the SiMES, the optic disc margin (inner edge of Elschnig’s ring) was manually defined by a trained ophthalmologist (S.C.L.). This critical step was accomplished by plotting a series of dots around the margin of disc on the reflectance image provided by the computer. The default standard reference plane was applied.
Socioeconomic and Systemic Factors
An interviewer-administered questionnaire was used to collect data on education levels, individual income, and housing type, which were subsequently summarized as an overall socioeconomic status (SES) factor for statistical analysis. Educational level was classified as: 1) primary or lower (≤6 years); 2) secondary (7-10 years); and 3) postsecondary (≥11 years). Based on Singapore dollars (SGD), individual income level was classified as: 1) low (<SGD1000); 2) middle (SGD from 1000 to <2000); and high (≥SGD2000). Housing type was classified as: 1) small (1- to 2-room public flat); 2) medium (3- to 4-room public flat); and 3) large (5-room public flat or private housing). The overall socioeconomic factor was categorized as: 1) having all three low SES factors; 2) having a combination of any two of the three low SES factors; and 3) having ≤1 low SES factor.
Height was measured with a wall-mounted tape and weight with a digital scale (SECA, model 782 2321009; Vogel & Halke, Hamburg, Germany). Blood pressure was measured with a digital automatic blood pressure monitor and hypertension was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg or a self-reported previously diagnosed hypertension. Diabetes was diagnosed as a random plasma glucose ≥11.1 mmol/L or self-reported physician-diagnosed diabetes or use of glucose-lowering medication.
Data Analysis
Statistical analysis was performed using STATA (version 8.2; Stata Corp, College Station, Texas, USA). Data on both eyes were used to assess the distribution of mean pixel height standard deviation and included in the generalized estimating equation (GEE) regression analyses. Bonferroni correction was used for multiple comparisons.
First, we used GEE logistic regression models to examine the influences of socioeconomic status and systemic and ocular factors on mean pixel height standard deviation. We constructed univariate and age-adjusted models. Factors that showed significant association with mean pixel height standard deviation in age-adjusted models would be further included in the multivariate GEE regression models.
Second, we examined the effect of HRT image quality (mean pixel height standard deviation) on the optic disc measurements. We excluded the eyes with a mean pixel height standard deviation >50 μm given that the optic disc measurements were largely inaccurate and thus unreliable in this range. Image quality was treated as either a continuous or categorized variable (1 = <10 μm; 2 = 10-20 μm; 3 = 20-30 μm; 4 = 30-40 μm; 5 = 40-50 μm). We then used the GEE models to examine the relationship between image quality and HRT optic disc parameter while controlling for age, sex, spherical equivalent, disc area size (except for the model for disc area), and the presence of glaucoma, high myopia, nuclear cataract, cortical cataract, post-subcapsular cataract, visual impairment, and blindness.
Results
Out of the 3280 participants, 3056 completed the HRT II test. Table 1 shows the characteristics of the SiMES cohort. Figure 1 shows the distribution of mean pixel height standard deviation. The mean (standard deviation) and median of mean pixel height standard deviation were 34 μm (34 μm) and 23 μm, respectively. A total of 16.1% (987/6112 eyes) had a mean pixel height standard deviation ≥50 μm; 22.0% (1342/6112) had mean pixel height standard deviation ≥40 μm; and 32.7% (1998/6112) had a mean pixel height standard deviation ≥30 μm.
| Characteristic | Values |
|---|---|
| Age (years), mean (SD) | 58.2 (10.9) |
| Height (cm), mean (SD) | 158.6 (9.1) |
| Weight (kg), mean (SD) | 66.3 (13.6) |
| Spherical equivalent (D), mean (SD) | −0.1 (2.0) |
| Astigmatism (D), mean (SD) | 0.4 (0.5) |
| IOP (mm Hg), mean (SD) | 15.3 (3.6) |
| Mean pixel height standard deviation (μm), mean (SD) | 34.2 (34.4) |
| Sex (female), n (%) | 1563 (51.2%) |
| Low SES (yes), n (%) | 113 (3.7%) |
| Hypertension (yes), n (%) | 2071 (67.8%) |
| Diabetes (yes), n (%) | 699 (22.9%) |
| High myopia (yes), n (%) | 89 (2.9%) |
| Visual impairment (yes), n (%) | 932 (30.5%) |
| Blindness (yes), n (%) | 110 (3.6%) |
| Nuclear cataract (yes), n (%) | 402 (14.7%) |
| PSC (yes), n (%) | 268 (10.2%) |
| Cortical cataract (yes), n (%) | 639 (24.3%) |
| Glaucoma (yes), n (%) | 79 (2.6%) |
| Retinopathy (yes), n (%) | 400 (13.1%) |
| AMD (yes), n (%) | 18 (0.6%) |
We treated the mean pixel height standard deviation as a dichotomous variable by using 30 μm as a cutoff. In the multivariate GEE logistic regression model, we found that age, negative degree of spherical equivalent, and the presence of high myopia, nuclear cataract, PSC, visual impairment, and blindness were associated with the presence of mean pixel height standard deviation ≥30 μm (all with P < .05), whereas sex, height, weight, IOP, and the presence of hypertension, diabetes, cortical cataract, glaucoma, retinopathy, and age-related macular degeneration were not significantly associated with the presence of mean pixel height standard deviation ≥30 μm ( Table 2 ). Similar associations were seen when mean pixel height standard deviation was dichotomized based on the cutoff of 40 or 50 μm. Consistently, when we treated the mean pixel height standard deviation as a continuous variable, the multivariate GEE model showed that increased age, decreased spherical equivalent, and the presence of high myopia, visual impairment, blindness, nuclear cataract, cortical cataract, and PSC were significantly associated with increasing level of mean pixel height standard deviation (all with P < .05).
| Presence of Mean Pixel Height Standard Deviation ≥30 μm a | Mean Pixel Height Standard Deviation (as a Continuous Variable) b | |||||||
|---|---|---|---|---|---|---|---|---|
| Age-Adjusted OR (95% CI) | P | Multivariate Adjusted OR (95% CI) c | P | Beta Coefficient (95% CI) | P | Multivariate Adjusted Beta Coefficient (95% CI) c | P | |
| Age (years) | 1.08 (1.07, 1.09) | <.001 | 1.06 (1.05, 1.07) | <.001 | 1.16 (1.07, 1.24) | <.001 | 0.58 (0.46, 0.72) | <.001 |
| Sex (female) | 1.24 (1.09, 1.41) | .001 | 0.94 (0.75, 1.17) | .58 | 2.32 (0.52, 4.13) | .01 | −1.28 (−3.92, 1.36) | .34 |
| Low SES (y/n) | 1.35 (0.99, 1.84) | .05 | – | – | 4.57 (2.32, 6.66) | .11 | – | – |
| Height (cm) | 0.98 (0.98, 0.99) | <.001 | 0.99 (0.97, 1.00) | .06 | −0.20 (−0.30, −0.10) | <.001 | −0.08 (−0.23, 0.08) | .12 |
| Weight (kg) | 0.99 (0.99, 1.00) | .05 | – | – | −0.13 (−0.20, −0.06) | <.001 | −0.08 (−0.15, 0.01) | .32 |
| Hypertension (y/n) | 0.99 (0.86, 1.16) | .97 | – | – | −0.40 (−2.50, 1.71) | .72 | – | – |
| Diabetes (y/n) | 1.22 (1.05, 1.41) | .007 | 1.17 (0.99, 1.40) | .07 | 2.84 (0.64, 5.04) | .01 | 2.08 (−0.13, 4.30) | .07 |
| Spherical equivalent (D) | 0.87 (0.84, 0.90) | <.001 | 0.92 (0.88, 0.97) | .002 | −3.46 (−3.91, −3.01) | <.001 | −1.89 (−2.51, −1.27) | <.001 |
| Astigmatism (D) | 1.38 (1.21, 1.57) | <.001 | 1.14 (0.97, 1.33) | .11 | 5.05 (3.14, 6.94) | <.001 | 0.48 (−1.44, 2.41) | .62 |
| High myopia (y/n) | 2.99 (2.05, 4.34) | <.001 | 2.65 (1.09, 3.27) | .02 | 22.11 (16.4, 27.8) | <.001 | 11.98 (6.47, 17.48) | .001 |
| Visual impairment (y/n) | 1.77 (1.56, 2.01) | <.001 | 1.52 (1.30, 1.77) | <.001 | 8.16 (6.28, 10.03) | <.001 | 6.35 (4.39, 8.30) | <.001 |
| Blindness (y/n) | 2.32 (1.51, 3.31) | <.001 | 1.58 (1.08, 2.31) | .03 | 39.04 (33.43, 44.65) | <.001 | 26.68 (20.41, 32.95) | <.001 |
| Nuclear cataract (y/n) | 1.81 (1.50, 2.18) | <.001 | 1.63 (1.30, 2.03) | <.001 | 20.39 (17.54, 23.24) | <.001 | 12.78 (9.81, 15.75) | <.001 |
| PSC (y/n) | 1.52 (1.24, 1.87) | <.001 | 1.34 (1.06, 1.71) | .02 | 14.66 (11.56, 17.76) | <.001 | 7.76 (4.61, 10.90) | .002 |
| Cortical cataract (y/n) | 1.18 (1.01, 1.38) | .04 | 1.10 (0.93, 1.31) | .26 | 4.75 (2.50, 7.01) | <.001 | 4.02 (1.80, 6.23) | <.001 |
| Glaucoma (y/n) | 1.44 (1.03, 2.01) | .03 | 1.35 (0.89, 2.05) | .15 | 4.31 (−0.75, 9.37) | .10 | – | – |
| IOP (mm Hg) | 1.02 (1.00, 1.04) | .01 | 1.02 (0.99, 1.05) | .06 | 0.35 (0.11, 0.60) | .005 | 0.22 (−0.03, 0.47) | .08 |
| Retinopathy (y/n) | 1.00 (0.98, 1.02) | .86 | – | – | 0.17 (−0.22, 0.55) | .40 | – | – |
| AMD (y/n) | 2.25 (0.98, 5.16) | .06 | – | – | 11.11 (−0.79, 23.01) | .07 | – | – |
a Generalized estimating equation (GEE) linear regression models treating mean pixel height standard deviation as a binary response variable (cutoff = 30 μm).
b GEE logistic regression models treating mean pixel height standard deviation as a continuous variable.
c Factors that showed significant association ( P < .05) in the age-adjusted regression analysis were included in the multivariate regression models.
The significant associations shown in Table 2 remained similar if: 1) the presence of visual impairment and presence of blindness (as dichotomized variables) were replaced by logMAR presenting visual acuity (as a continuous variable); or 2) the presence of nuclear, cortical, or PSC cataract (as dichotomized variables) were replaced by the severity scale of nuclear, cortical, or PSC cataract (as continuous variables). In another subsidiary analysis after excluding persons with glaucoma (n = 133), we found that the results were similar to the ones generated from the whole sample. For instance, age ( P < .001) and the presence of high myopia ( P < .01), nuclear cataract ( P < .001), cortical cataract ( P = .001), visual impairment ( P < .03), and blindness ( P = .01) were significantly associated with the presence of mean pixel height standard deviation ≥30 μm.
Table 3 shows that most of the optic disc parameters were significantly associated with mean pixel height standard deviation (as a continuous variable). When we treated the mean pixel height standard deviation as a categorized variable, the optic disc parameters varied among different image quality groups, except for optic disc area, rim volume, and cup shape measure. Generally, these associations obtained statistical significance when mean pixel height standard deviation was higher than 30 μm. Figure 2 shows the associations between image quality category and neuroretinal rim area across different optic quadrants (temporal-superior, temporal-inferior, nasal-superior, and nasal-inferior). In all four quadrants, neuroretinal rim area was significantly smaller in the “30-40 μm” group compared with the “<10 μm” group ( P < .05 for all). Statistical significance remained after Bonferroni correction (data not shown).
