Ординатура / Офтальмология / Английские материалы / Field of Vision A Manual and Atlas of Perimetry_Barton, Benatar_2003

in each probability map, and at each of the 76 points there is a 1 in 20 chance of a normal subject getting a low score. Theoretically, if the odds at each point were independent—which they are not really, since they come from the same subject under similar con- ditions—the odds of getting every single point within 95% normal limits would be (0.95)76, or a mere 2%. Thus, it is highly likely that just by chance a normal subject will get a few low scores. If we make the criteria more stringent, and look only at points marked with black squares (p < 0.005), there is still a 31% chance of a normal subject getting one or more abnormal points.

The point is that a few abnormal scores do not make a story. Rather, one seeks patterns. A normal subject may have some points flagged as abnormal, but these will be randomly scattered. By contrast, the abnormal points of a subtle defect will be clustered in a pattern consistent with the clinical suspicion. On occasion, this may be a single point—as with enlargement of the blind spot—but most of the time it will be a cluster of several spots. If only a single spot is defective, one should be cautious about conclusions of either normality or pathology. A single point near fixation is more likely to be abnormal than one in the periphery, where performance is always more variable, but even so this indicates that a more finely grained central field such as a 10-2 might have been a better choice of test.

To evaluate the patterns in the probability, the same questions posed in the last chapter apply, with the same reliance on knowledge of anatomy and disease. These questions are not recapitulated here. However, the appearance of artifact on automated perimetry will differ from that in manual perimetry. The following examples should be considered:

1.“High responder”: Some subjects are determined to see every possible light (Fig. 13). In the parlance of signal detection theory (see appendix), they will have a criterion bias that is shifted toward low signal values. Anything that suggests the hint of a light will trigger a response. Thus, these patients have elevated FP rates, their MD is a positive value, and they have locations with improbably high thresholds. They will often also show FLs, because their drive to see will lead them to look for targets with eye movements.

2.“Low responder”: Other subjects feel that they should be cautious and not respond unless they are certain that a target was present (Fig. 14). This is not uncommon among patients doing automated perimetry for the first time, especially if poorly instructed. The result is a criterion bias opposite to that of high responders. Interestingly, this shift tends to affect the peripheral more than the central field, so that there is often the appearance of a concentric constriction. This can lead to problems if the patient is being investigated for just such a defect. The examination should be repeated with more detailed instructions about the need to respond to very faint lights. It can also be helpful to supplement the test with Goldmann perimetry, which is less prone to such problems.

3.Lens holder artifact: This is more variable than in Goldmann perimetry, because the lens holder is adjustable. The typical pattern is a severe reduction in the peripheral lower field (Fig.15). (The holder does not have an upper part, so the upper field is spared, unless one errs by using a trial lens with a large rim.) This can mimic an

inferior arcuate defect, but two points can help to differentiate the two. First, lens rim artifact may not respect the nasal horizontal meridian (see the examples, Fig. 15). Second, arcuate defects will “point” to the blind spot, whereas rim artifacts will often circle around it. This will be particularly evident with the 30-2 program, which provides a 10° margin between the blind spot and the edge of the tested field. However, at times it can be hard to tell the difference (Fig. 16), and one should repeat perimetry without a lens holder or with more careful attention to it.

4.Lid artifact: This sometimes crops up on automated perimetry, just as it does with manual perimetry (Fig. 17). The appearance is not much different, a shade over the superior field with variable slant. Perimetry can be repeated with the lids taped up (it is too much to ask patients to voluntarily elevate their lids for the whole time it takes to do automated perimetry), or a quick check with Goldmann perimetry can confirm or deny the defect.

2.2. SEQUENTIAL ANALYSES OF FIELDS

As with Goldmann perimetry, one of the key issues is detecting change over time. The difficulty again is differentiating true change in pathology from the variable performance of a patient with a stable, unchanging defect. Even normal subjects will not produce identical perimetric results from two different sessions. This is known as long-term fluctuation, or inter-test variability. Unlike SF, there is no simple way of assessing the degree of longterm fluctuation in a given patient, although studies show that it is quite variable in normal subjects. The advantage of an automated collection of numerical data is that it is possible to apply statistical analysis to the problem of detecting or depicting change.

Local changes or changes in the pattern of the field are best captured on the overview printout (Fig. 18). This plots the gray scale and pictorial probability maps for each session in one row, with subsequent sessions added one below the other. Each also has a listing of the reliability indices, global indices, pupil size, and acuity. The progression over time is more easily appreciated, but no statistical analysis is provided. As with Goldmann perimetry, a new defect superimposed on an old defect can be easy to spot (see Case 25). On the other hand, progression in the severity of a pre-existing defect is more difficult to judge with confidence (see Case 43). Again, repeating the assessment several times to check for persistence or continued decline across a number of fields is the best strategy if doubt lingers about the stability of the field loss.

The change analysis program of the Humphrey automated perimeter is a convenient way of looking at the progression of some overall summary variables (Fig. 19). This printout first displays a box plot for each date of testing. This simply masses all the threshold values obtained in a test, irrespective of location, and determines the median (center stripe in the box), the 15th and 85th percentiles (the top and bottom of the box, meaning that the box thus contains 70% of all data points), and the least and most sensitive thresholds (indicated by the extent of the arms). Next are four plots for the four global indices. Points obtained from tests with poor reliability indices are flagged with an “x” rather than an “o”. These plots show the 5 and 1% limits of normative data. If five or more fields are used in the change program, it will also indicate whether the slope of the linear regression is significantly different from zero.