The visual prognosis of patients suffering from retinal branch vein occlusion complicated by macular oedema is assessed in relation to the degress of capillary fallout. Extensive retinal capillary fallout associated with macular oedema was found to confer a poor visual prognosis in a series of 89 patients foNowed for an average time of 14 months. If the degree of capillary fallout associated with marular oedema was minimal, the visual prognosis was considerably improved. The most common cause of reduced central vision in retinal branch vein occlusions is macular oedema. Macular oedema complicates approximately 60% of temporal branch vein occlusions and becomes chronic in two-thuds of these cases.' ' Retinal branch vein occlusion is a common cause of retinal vascular disease and because macular oedema is a frequent complication, prognostic factors in these cases have been closely studied? From these investigations, suggested factors with prognostic signficance have been the site of occlusion, extent of the macular oedema and disruption of the foveal arcade. The retinal vasculature has been described as responding in two ways to venous occlusion.' The blood vessels may become dilated and leak as a result of disruption of the blood retinal barrier. Alternatively, an ischaemic change may occur and this is maiifested by capillary closure. It has been appreciated that these responses may co-exist but most investigators have discussed them separately, emphasising abnormal permeabdity alone as the cause of macular oedema. Method Eighty-nine consecutive patients suffering from retinal branch vein occlusion were examined by one of the authors and underwent retinal photography with fluorescein angiography. Results Forty-eight patients (53%) had detectable macular oedema on fluorescein angiography. The duration of symptoms before examination varied from six weeks to 28 months but most patients were seen two to three months after the onset of symptoms. The patients were followed for an average of 14 months with minimum follow-up time of five months. There was no significant sex difference (female 47'70, male 5370) and the average age of presentation was 63 years. The right eye was affected in 5570, the left in 41% and in 4% the retinal branch vein occlusions were bilateral. The site of occlusion is shown in Table 1 and the branch vein occlusions complicated by macular oedema had a similar distribution to the overall series. Of the 48 patients with macular oedema,

SummaryRetinal branch vein occlusion is a common cause of retinal vascular disease and because macular oedema is a frequent complication, prognostic factors in these cases have been closely studied?From these investigations, suggested factors with prognostic signficance have been the site of occlusion, extent of the macular oedema and disruption of the foveal arcade.The retinal vasculature has been described as responding in two ways to venous occlusion.' The blood vessels may become dilated and leak as a result of disruption of the blood retinal barrier. Alternatively, an ischaemic change may occur and this is maiifested by capillary closure. It has been appreciated that these responses may co-exist but most investigators have discussed them separately, emphasising abnormal permeabdity alone as the cause of macular oedema. MethodEighty-nine consecutive patients suffering from retinal branch vein occlusion were examined by one of the authors and underwent retinal photography with fluorescein angiography. ResultsForty-eight patients (53%) had detectable macular oedema on fluorescein angiography. The duration of symptoms before examination varied from six weeks to 28 months but most patients were seen two to three months after the onset of symptoms. The patients were followed for an average of 14 months with minimum follow-up time of five months. There was no significant sex difference (female 47'70, male 5370) and the average age of presentation was 63 years. The right eye was affected in 5570, the left in 41% and in 4% the retinal branch vein occlusions were bilateral. The site of occlusion is shown in Table 1 and the branch vein occlusions complicated by macular oedema had a similar distribution to the overall series. Of the 48 patients with macular oedema,

A total of 16.8% of the eyes observed in this study had either disc haemorrhages or vascular changes. The underlying trend of vascular and haemorrhagic changes in glaucoma are demonstrated in this sample, which is in general agreement with previous studies. The high percentage of optic disc haemorrhages in low tension glaucoma is highlighted. The presence of microaneurysms and nerve fibre layer haemorrhages is interesting but of unknown significance.

This report describes the application of this technique to the eye, where it provides, as in the brain, a rapid and accurate method of measuring changes in arterial calibre. Materials and methodsThe system consists of a standard Zeiss fundus camera through which the fundus is visualised by means of a low-light-level television camera (Ikegami). This camera contains a silicon intensifier target (SIT) tube. The image of the retinal arterioles is passed from the fundus camera through an image splitting eyepiece (Vickers), and the split image received by the television camera is displayed on a monitor. The shearing screw of the eyepiece which controls the degree of image splitting is connected to a 10-turn potentiometer and in turn to a digital readout of resistance. In this way frequent measurements of shear (which is directly proportional to vessel calibre) were obtained in arbitrary units. The fundus is illuminated with the standard observation light of the fundus camera on position 2.Pupils were dilated with 2 drops of tropicamide 1% instilled 20 and 10 minutes before measurement. The accuracy of the method (intraobserver error) was first assessed by making 50 measurements of shear on a first-order arteriole near the optic disc in a human volunteer. One observer determined the end point by observing the point between overlap and separation of the optically split image on the television monitor. A second observer recorded each measurement from the digital meter. Interobserver error was assessed by 3 trained observers making 10 measurements of a single retinal arteriole in a second volunteer.A further 7 volunteers were used in 11 experiments in which 5 measurements were made at resting intraocular pressure followed by another 5 measurements after increasing the intraocular pressure to 30-50 mmHg. Initial attempts to increase and maintain the intraocular pressures with a suction cup were not reliable. With this method the rise in pressure was unstable while the suction resulted in distortion of the globe, which degraded the quality of the image. The method finally employed used a Mackay-Marg tonometer mounted on a jig which could be adjusted by a micrometer screw to compress the temporal sclera of the eye and so increase the intraocular pressure.4 The tonometer was connected to an analogue meter from which continuous readings of increase in intraocular pressure were monitored, and the intraocular pressure was kept at a sustained level. Arterioles were selected at random, but all were near the disc and either first-or secondorder retinal arterioles. Results INTRAOBSERVER ERRORThe results of 50 measurements made on a secondorder arteriole by one observer gave a mean of 76 291

The red-free negatives of 53 right eyes (30 normal eyes and 23 glaucomatous eyes) and 51 left eyes (32 normal eyes and 19 glaucomatous eyes) were analysed using two different image densitometry techniques. The first technique measured the density from rectangular sample areas, while the second measured density from sector-shaped sample areas which more closely follow the course of the nerve fibres in the retina.Indices which measured the deviation of the data from a clinically determined normal 'gold standard'were calculated, and were used to determine the optimum sensitivity and specificity in separating normal from glaucomatous eyes. There is a significant difference between the data from the normal and glaucomatous groups of eyes, when measured from the sector sample areas. The relative efficacy of this technique is also shown by the improved values of sensitivity (from 42%-70% to 70%-91 Vo), although specificity remained fairly constant (from 66%-83% to 620/0-88%).Key words: Densitometry, glaucoma, image analysis, predictive value, red-free photography, retinal nerve fibre layer.Analysis of red-free photographs has become an established method of observing glaucomatous damage to the retinal nerve fibre layer (RNFL). A number of studies have used semi-subjective techniques to grade diffuse and wedge defects in thl RNFL.'.' These have shown an association be tween the degree of damage and the severity o glaucoma. However, a number of reports havc indicated a need for an objective method of gradine RNFL defects.*-" Densitometry has been used as a tool to measurc the RNFL1*-15 and sources of variance have beer identified.16 We have previously published reports on densitometry data from rectangular sample areas in the temporal retina,16 and have been able tc separate normal from abnormal eyes. This paper reports on changes to our data collection techniques that reflect better the course of the nerve fibres in the retina. Two data sampling and analysis techniques are presented, and their relative efficacy is determined by comparing the results against a clinical judgement of whether the subject is normal or has glaucomatous damage. Methods SubjectsNormal and abnormal (glaucomatous) subjects were selected from over 220 patients who have been studied for a period of five years. A subject with no visual field defects, a normal intraocular pressure ( < 22 mmHg), and a normal optic disc (C:D ratio less than 0.6) was classified as normal; a subject with visual field loss and abnormal optic disc cupping (C:D ratio > 0.6) was classified as glaucomatous (or abnormal). These abnormal subjects were chosen to cover a range of severity as would be seen in a glaucoma population. Full details of the classification From the Lions Eye Institute,