How do we make clinical judgments about ocular appearance? This is a critical question as it relates to diagnosis (and prognosis) in eye care since so much diagnostic skill is based on recognising the unique macro and micro appearance of many diseases/conditions affecting the eye. In addition to diagnosis, though, appearance judgment is vital in assessing the changes to the eye that occur with treatment to determine if an intervention is effective. The latter can be particularly problematic because in deciding on differences we are automatically assigning numbers to the appearance we are judging even if those are simple binary decisions worse/not worse or worse/better.

Figure 1: The problem can be stated in relationship to this image: How red is this eye and is it getting better or worse?

These types of judgement are not solely to diagnose and track clinical changes; the words and numbers used are clinical summaries that specify some useful attribute about the patient. If the words or numbers faithfully represent the underlying condition described, these words or numbers are measurements and the numbers form the basis of a scale. Utilising this scale is clinical grading with the standard being the CCLRU grading scales.

Clinical grading of the anterior segment is something of a mystery. Although there has been some scientific study of it, we know almost nothing about how the skills are acquired and how clinicians actually make the grading judgments. All of the experiments have been about how the scales themselves are used or how to automate the process and very few have been about scale design and verification. So although we are starting to understand how clinicians use the various scales that are available, we do not know whether the scales actually measure the attribute they are designed to measure!

There are a number of things we do know about anterior segment scaling. In their simplest forms we are very good at the basic judgments required. Humans can discriminate colour, form, depth and texture very well, so the basic building blocks of judging appearance are present. If we complicate the task by making all of these basic visual judgements on eyes (or sometimes as is done experimentally, on images of eyes), we know that we can reliably perform the grading. Although there are slight differences, it generally doesn’t matter much what kind of scale is used; one with just reference words is more or less the same as one with reference pictures, one based on many pictures (or even a movie of the condition of an eye worsening) is used with surprisingly similar results to the other 2.For example, the next time a red eye is seen it will generally be judged to be red. This suggests that clinicians have rules about using scales that they use similarly from one time to the next. There are big problems though with repeatability between observers; a red eye judged by someone may be judged to be not so red by another. What this implies is that even though clinicians have access to the same scales (for example a set of photographic reference pictures that define a range of a condition), the rules each clinician chooses to use when applying the scale differs. There are suggestions that perhaps training may affect this, but there are also results showing that it is unaffected by training! Finally there is one more thing about how we grade; we like to use “pretty” numbers that divide the scale into predictable amounts. This results in grades that cluster in particular positions on the scale.

So what can we do? There have been a number of demonstrations of the feasibility of automating the clinical grading of bulbar redness. This allows us the luxury of objectively extracting the salient data from the eye being assessed. This is illustrated using the McMonnies redness scale showing that two objective measures of the images that form the basis of the scale behave in remarkably similar ways and reflect the scale quite well. One (fractal dimension) captures details of the blood vessels in the images and the other (chromaticity component CIE x) captures the overall redness in the image. The same logic and similar techniques should work with other types of ocular redness, corneal and conjunctival staining and perhaps even with something as complicated as tarsal roughness. With the availability of high speed desktop computers, this is almost a reality. The major difficulty is still that all the algorithms described need some sort of operator to define areas to be measured and the exact details of the measurement. Perhaps eventually, though we will have computerised techniques developed that will allow us to completely objectively quantify ocular appearance.

What can we do in the mean time? Individually, our grading is repeatable, so we should continue to use the scales in ways similar to what we are doing now. What we are not that good at is being consistent with our colleagues. If we are working in settings where patients are being seen by multiple practitioners, it is critical that the same scales be used and the rules applied, to ensure that assigned numbers to appearance by different clinicians are the same. Finally, we should think about establishing clinical standards that would promote the international use of the same scales, using the same rules for each scale. This is not trivial and entails developing standards committees, methods to design and verify scales and eventually the international promotion of these scales in the professions who would benefit from their use.

Figure 2: McMonnies Redness Fractal Dimension and Chromaticity.