Eleven subjects (six male and five female) aged 20 to 31 years (mean age 23 years) were recruited for the study. All subjects had corrected visual acuity of 6/6 or better in both eyes and were emmetropic or had a slight refractive error (sphere power ranged between +/−1 D, cylinder power no more than 1D). All subjects had good ocular and general health and each was screened for anterior eye conditions that could contraindicate contact lens wear. No subject reported a history of significant dry eye symptoms and none had worn contact lenses for at least one month prior to the commencement of the study. All subjects gave informed consent and the study was approved by the university research ethics committee.

The study was conducted over four consecutive weeks although not all subjects completed all four weeks. During the first week, no contact lenses were worn and bare eye measurements were taken from both eyes on the first and seventh days of the week (11 subjects). In the second week, each subject wore a hydrogel lens (FDA group IV, 58%, etafilcon A), in one eye (randomly chosen) and measurements were taken in the lens-wearing eye on the first and seventh days of the week (10 subjects). Following a week of no lens wear, the hydrogel lens was exchanged for a silicone hydrogel lens (FDA group I, 47%, galyfilcon A) and measurements were taken on the first and seventh day of the fourth week (5 subjects). The reduction in the number of subjects during the week of silicone hydrogel lens wear was the result of subject dropout due to discomfort and/or time constraints. All statistical tests were adjusted to account for the different number of subjects in each section of the study.

The two contact lenses used in this study were clinically assessed as being good fits and the subjects reported them to be comfortable. All lenses had a diameter of 14.0mm, a base curve of 8.3mm and power of –1.00 D. The lenses were worn in one eye only (monovision, but not necessarily providing optimal visual correction) on a daily wear schedule. Subjects used a range of common multipurpose solutions to care for the lenses. It would have been preferable to use only one type of solution to control this study variable, but unfortunately the one solution brand we chose to use was withdrawn from the market for safety reasons during the study.

Subjects had measurements taken at three different times each day; in the morning (between 8-10 am), at lunchtime (12-2 pm) and in the afternoon (4-6 pm). Five individual high speed videokeratoscopy (HSV) measurements were taken at each measurement session. Subjects were advised to fixate on the green light in the instrument (Medmont E300 videokeratoscope) and were instructed to blink, then open their eyes (not wide, but naturally open), and to avoid further blinking during the short period of data collection. Each measurement was taken over a period of 8seconds at a videokeratoscope sampling rate of 25Hz (equivalent to one image every 40ms). Therefore each 8second measurement resulted in 200 individual measurement frames. Each subject's initial blink was included at the beginning of each 8sec recording (analysis procedure will be described later). Measurements that showed poor fixation or significant variation in the corneal apex distance to the imaging device were excluded. If the analysis area was obstructed by the eyelashes or the eyelid itself, the measurement was rejected and the subject was asked to open their eyes slightly wider during the next recording. Measurements were also repeated if tear debris or mucin were seen in the recorded images. In such cases, the subject was asked to blink a few times to flush away the debris and a new measurement was taken. The measurements were carried out in a laboratory where the temperature ranged between 22.1°C and 25.6°C with a mean of 23.6°C (±0.7°C), and the humidity ranged between 30% and 65%, with a mean of 46.9% (±7.1%) over the course of the four weeks of the study. At each measurement session, the subjects were also asked to rate the subjective ‘feeling of dryness’ in their eyes on an analogue scale from 0 to 10, where 0 represents very dry eyes and 10 represents no dry eye symptoms.

During the first week of bare eye measurements, on the morning of the first day, a range of other tear function assessments were undertaken. The tear film break-up time was measured using blue light and a slit lamp biomicroscope by instilling fluorescein and subjects were asked to blink several times and then to keep their eyes open and to suppress blinking. The McMonnies dry eye questionnaire31,32 was administered and scored. The Zone Quick (FCI Ophthalmics Inc. Marshfield Hills, MA, USA) phenol red thread test was used according to the manufacturer's recommendations to estimate tear volume. An evaluation of the anterior eye (including assessment of lid margins, lid eversion and surface staining) was also conducted using a slit-lamp biomicroscope.

A set of images captured during contact lens wear is shown in figure 1. The quality of the ring reflections is closely associated with the quality of the tear film surface.23 Thus, the clinical problem of estimating tear film surface quality can be reduced to the technical problem of estimating the image quality of videokeratographs. In order to estimate the image quality of a HSV ring pattern, we first select a square image section, centred on the instrument's axis, and defined as intensity matrix I (x,y), x= 1,2,…, L, y= 1,2,…, L where L denotes its size in pixels. We then define a set of radial image profiles

Figure 1.

An example of a HSV recording (link to movie in electronic PDF version, available at: www.journalofoptometry.com) for an eye with a contact lens.

(0.14MB).

Figure 2.

Examples of Cartesian to polar transformation of the analysis area (denoted by light circles) for two subjects with different pupil sizes. The intensity of the polar image of subject with smaller pupil (b) is less uniform than for subject with larger pupil size (a).

(0.13MB).

The number of rings is then counted for each column of the matrix Ip that corresponds to a semi-meridian in the original HSV image to find out the discontinuities in the ring pattern. These discontinuities often indicate instabilities in tear film surface. To count the rings, edges in the polar image need to be detected and this could have been performed by the Marr-Hildreth edge detection algorithm,34 for example, in a similar way as described in our earlier work.23

However, changes in the characteristics of the human anterior eye pattern, such as the variation in iris color, different pupil shapes and sizes, and the natural changes of pupil size during a long measurement with HSV makes the ring detection task difficult. There are no “off-the-shelf” edgedetecting algorithms that are universally applicable for all types of eyes. Examples of how pupil size affects the reflection of the Placido rings pattern and subsequently its polar image representation are shown in figure 2. If the pupil size encompasses the analysis area, as it is shown in figure 2a, the resulting radial image representation remains of relatively uniform intensity. However, when the pupil is smaller than the analysis area (Figure 2b), the resulting radial image shows significant variations in the intensity that cannot be simply removed with standard histogram equalisation techniques.35 In particular, it can be seen that for a small pupil, part of the subject's iris is included in the sampled radial image. The contrast of Placido rings on the light colour iris is degraded, which makes it even harder to detect the sometimes-already-blurred rings.

To overcome these deficiencies, a customised edge detection method was developed, in which we first estimated the average local radial intensity image profile that was then fit to a parametric function composed of two parts. An iterative least-square procedure was used to find the optimal combination of the two parametric parts of the function. The modelled local average intensity was then subtracted from the corresponding local part of the original radial image prior to edge detection. The edge detection involved estimating all rising slopes of the radial profile using adaptive thresholding technique. The effect of using the background subtraction technique is graphically explained in figure 3 using, as an example, the image from figure 2b. The original radial image (Figure 3, top-left) has a significantly uneven intensity distribution, which causes the edge detection routine to miss significant portion of the rings in the iris area (Figure 3, bottom-left). After estimating the average intensity radial profile and fitting a parametric function to it (Figure 3, centre), we are able to subtract the average background intensity and perform edge detection on a residual radial image (Figure 3, top-right) which results in a noisier but also more complete edges (Figure 3, bottom-right). The resulting noise is of no significance because in the final step we simply count the number of edges leading to the ring number estimator

Figure 3.

The effect of local background subtraction on the performance of edge detection algorithm for radial profile images that contain significant part of lighter iris. Note that without the background subtraction a significant number of rings are missed in the edge detection process.

(0.29MB).

Because the tear film surface instability results in blurring and discontinuities in the ring patterns, we use the variance of the above ring number estimator

In order to achieve a parameter that would be proportional to the surface quality of tear film and that would be bounded between 0 and 1, we will use in the remainder of the paper the normalized TSQ indicator given by TSQnorm =(10-TSQ)/10, where the approximate value of 10 corresponds to the highest empirically recorded value of the TSQ during the blink (i.e., when only part of the Placido ring image is visible). Such normalization allows the study of tear film behaviour in its build-up phase. During the tear film stability phase TSQnorm is normally above 0.9.

A typical example of estimating tear film surface quality is shown in figure 4 in which one can clearly see the blink, the tear film build-up phase (up to 1st second),20 and the phase in which the tear film is relatively stable. In general, the tear film surface quality (TSQ) indicator estimated from the set of HSV data can be used to estimate three important parameters: tear film build-up time (TBLD), tear film break-up time (TBUT), and the average tear film surface quality in the relatively stable interblink phase of the tear film. In this study, we concentrated on the tear surface quality during the inter-blink period to address the question of whether contact lenses cause a measurable change in the tear surface compared with the bare eye.

Figure 4.

A typical example of the estimation of the tear film surface quality for a subject using high speed videokeratoscopy images. The tear surface quality score (TSQ) is normalised such that a score of 1 is best (i.e., good tear surface quality).

(0.13MB).

Analysis of the videokeratoscope data (5×8 sec recordings per measurement session) was done at a later time. The first frame after the initial blink was identified in the recording. Each sample was made up of this frame plus the following 149 frames (equivalent to a period of 6seconds, see figure 4). To remove the tear build-up phase following each blink from the analysis, the TSQ data was sampled from the 26th (1sec post-blink) to the 150th frame (6sec post-blink) of each measurement (i.e., during the relatively stable inter-blink phase of the tear film). By limiting the analysis to no more than 6sec post-blink and choosing subjects with good tear quality, we also minimized the variability in TSQ that is potentially associated with both the tear build-up and break-up phases of the tear dynamics.20,23,36 The area of the tear film/cornea that was analysed was defined as the inferior half of a circular area centred on the rings with a diameter of 6mm (see figure 3). This was done to avoid possible interference from the reflections from the eyelashes, which could have been mistaken by the algorithm as changes in tear film.

Each 6second sample of 150 frames was filtered to remove outlier values and to improve the reliability of the tear surface quality mean for that sample. Firstly, the measurements in each set were detrended and their signal powers (variances) calculated. Out of these, the measurement with the smallest variance was taken as the reference measurement. Subsequently all other measurements from that data set were compared to this reference measurement. Only signals with powers within 3 dB of the reference measurement were included in the analysis.

In some cases, the filtering procedure outlined above eliminated a number of the measurements from a particular measurement session. Of the 5×8 second recordings taken at each measurement session, the lowest number remaining after filtering was two at a session and for most sessions, there were 3 or 4 recordings available for analysis and subsequent averaging. After reviewing the recordings that were eliminated by the filtering procedure, the poor measurements were mainly caused by transient local disturbances due to tear film debris and not consistent changes in the tear film quality.

Statistical analysis of normalised TSQ was performed using SPSS 15.0.1 software. Statistical procedures including the Pearson's correlation test and repeated mesures ANOVA were used. For all statistical tests a p-value of less than 0.05 was considered significant. TSQ data was assumed to be normally distributed.

The normalized tear surface quality values, TSQnorm, for the eleven subjects with bare eyes ranged from 0.8956 to 0.9897, (0.9685±0.0118; that is, mean±standard deviation) across the six measurement sessions during the first week of the study. There was no evidence of a systematic change in TSQnorm as a function of time of day from the morning 0.9685±0.0092, to lunch 0.9681±0.0101 to afternoon 0.9688±0.0075 measurement sessions. A repeated measures ANOVA showed no statistically significant effect of time of day (P=0.911) or day of week (day 1 versus day 7). Correlation in values of TSQnorm between the right and left eyes of the bare eyes of the same individuals showed a significant association (Pearson's r=0.61).

The wearing of both the hydrogel and the silicone hydrogel lenses caused the TSQ to significantly worsen (see figure 5). The group mean TSQnorm for the bare eye condition averaged over the 6 measurement sessions was 0.9668±0.0120 for the group of measurements corresponding to those of contact lens wear, while during hydrogel lens wear it was 0.9427±0.1640, and for the silicone hydrogel wearing period it averaged 0.9407±0.0179.

Figure 5.

Normalized tear film surface quality for bare eye and contact lens wear lens. The group means and SDs are derived from 10 subjects for the bare eye and hydrogel lens wear condition and a subgroup of 5 subjects for the bare eye and silicone hydrogel lens wear condition.

(0.1MB).

A repeated measures ANOVA on the effect of the type of lenses, time of day (morning, lunch and afternoon) and time of week (day 1 or day 7), showed a significant effect of contact lens wear on TSQnorm (P<0.01), but no systematic time of day or time of week effects (P>0.05). Post-hoc analysis showed that both the hydrogel and silicone hydrogel lenses gave significantly worse TSQnorm values than the bare eye condition, but that there was no significant difference in TSQnorm values between the hydrogel and the silicone hydrogel lens wear periods (P>0.05).

Repeated measures ANOVA showed a significant time of day effect for the TSQnorm values derived when wearing the silicone hydrogel lenses. The TSQnorm values improved slightly over the course of the day on day 1 of the week (P=0.045). Similar analysis of the hydrogel lens wear period showed no significant change in the TSQnorm values over the course of the day (P>0.05). There were no systematic changes in TSQnorm values for the hydrogel or silicone hydrogel lens wear periods as a function of time of week (day 1 versus day 7).

We compared the tear surface quality values obtained with the bare eye condition (week 1, visit 1) with common tear function tests (McMonnies dry eye questionnaire, fluorescein TBUT, phenol red thread test and subjective dryness rating) conducted at the same visit. The results of Pearson's correlations are shown in table 1 with no significant correlations for the McMonnies dry eye questionnaire (r=0.13, P>0.05), fluorescein TBUT (r=0.22, P>0.05) or phenol red thread test (r=0.06, P>0.05). However there was a weak, but statistically significant correlation between the bare eye subjective dryness rating and the TSQ values (r=0.62, P<0.05). This association between TSQ and subjective dryness was not maintained during contact lens wear.