Grading of Fundus Photographs and Fluorescein Angiography

Grading protocols for fundus photographs and FA were adapted from the Central Vein Occlusion Study (CVOS) Group, the Branch Vein Occlusion Study (BVOS) Group, the Standard Care vs Corticosteroid for Retinal Vein Occlusion (SCORE), and the Early Treatment of Diabetic Retinopathy Study (ETDRS). Grading was designed to provide qualitative and quantitative assessment of fundus photography and angiographic endpoints. For each modality, image quality was classified as “adequate” when photographic features could be read with confidence, “fair” when all or nearly all photographic features could be read but confidence in grading was only fair, and “poor” when no photographic features could be read with confidence owing to poor image quality.

Color fundus photographs were evaluated for presence and area of intraretinal hemorrhage (IRH) as well as presence of other features of vein occlusion, including dilated retinal veins, ghost vessels, intraretinal cystoid spaces, hard exudates, epiretinal membrane, subretinal fibrosis, photocoagulation scars, cotton-wool spots, pigmentary changes, collateral vessels, neovascularization elsewhere (NVE), preretinal or vitreous hemorrhage, and vitreous hemorrhage, as well as disc swelling or disc neovascularization (NVD) ( Supplemental Figure 1 , available at AJO.com ).

For FA, the area of fluorescein leakage, area of capillary nonperfusion, and area of capillary changes (dilation of capillaries or arteriolar abnormalities such as pruning) were quantified. Other features evaluated on FA included FAZ size and outline, presence of collaterals, and presence of NVD (originating and residing within 1 disc diameter of the disc margin) and/or NVE. A single early- to mid-phase image centered on the macula was selected for analysis of areas of ischemia and FAZ. A late-phase image was chosen for analysis of leakage and neovascularization. The area of IRH, fluorescein leakage, capillary nonperfusion, and capillary changes were calculated in disc areas (DA) based on the ETDRS grid centered on the macula. The grid had a foveal central subfield, a middle ring, and an outer ring (each corresponding to a circle with a 500-, 1500-, and 3000-μm radius) divided in a total of 9 subfields corresponding to 11.1 Macular Photocoagulation Study DA (1 DA equivalent to 2.54 mm ² of the retina). The grid consisted of white lines and was overlaid on the image electronically ( Supplemental Figure 2 , available at AJO.com ).

Grading of Optical Coherence Tomography Angiography

All OCTA images were acquired with the RTVue XR Avanti (Optovue Inc, Fremont, California, USA) in a manner previously described by Spaide and associates. The instrument captured 2 consecutive B-scans at a fixed location (each containing 304 A-scans at a rate of 70 000 scans per second) before proceeding to another location. Approximately 3 seconds are required to acquire a 3-dimensional vascular cube.

All OCTA scans were reviewed, but only the best-quality 3 × 3-mm and 8 × 8-mm scans centered on the macula and the best 4.5 × 4.5-mm scan centered on the disc (when available) were selected for analysis. Grading was done on the full-thickness 3-dimensional image composed of the superficial, deep, and outer retina vascular layers superimposed. Images were graded using the same protocol described for FA. When there were doubts about the nature of a lesion (for example, for NVE) the OCT co-registered B-scans were evaluated. For the 3 × 3-mm scans only the foveal central subfield and the middle ring (1500 μm radius) were used. For each of the grading features, the superficial capillary plexus (automatically segmented and representing a projection of the vasculature between the retinal nerve fiber layer and the ganglion cell layer) and the deep capillary plexus (automatically segmented and representing the vasculature from the border of the inner plexiform layer and the inner nuclear layer to the border of the inner nuclear layer and the outer plexiform layer) were compared to the full-thickness scan and it was determined if their presence, grade, or total area was different between the 2 layers ( Supplemental Figure 3 , available at AJO.com ). In cases of possible automated software segmentation error, manual adjustment of the boundaries was not performed. The disc scan was graded for presence of ischemia (in the peripapillary retina quadrants) and for the presence of collaterals, NVE, and NVD ( Supplemental Figure 3 ). The grader (J.N.C.) was masked (ie, fundus photographs, FAs, and OCTA images were graded independently). In cases of doubts in the grading, a senior grader (P.A.K.) evaluated the images and any disagreements were resolved by open adjudication. After grading and classification of the images, FA and OCTA 8 × 8-mm and disc scans were compared to evaluate what additional features could be seen in either scan.

Statistical Analysis

Descriptive statistics (percentages, means, and standard deviation) were computed for demographic and clinical variables. Fundus photographs, FA, and OCTA features were categorized and are expressed as percentages. Agreement between the continuous variables graded on FA and OCTA was tested with the intraclass correlation coefficient and are presented as intraclass correlation coefficient (ICC) and 95% confidence intervals (95% CI). Cohen’s unweighted and weighted kappa coefficients were calculated for nominal and ordinal categorical variables, respectively. Logistic regression models were used to test the best-corrected visual acuity (BCVA) low threshold for poor quality of OCTA scans. To compare means, the Mann-Whitney U test was used and to compare categorical variables, the Pearson χ ² . For all tests, a P < .05 was considered significant. Statistical analyses were performed using SPSS software version 21 (SPSS, Inc, Chicago, Illinois, USA) and MedCalc version 15.8 (Medcalc Software, Ostend, Belgium).

Disease Features on Color Fundus Photography and Fluorescein Angiography

Color fundus photographs (centered on the posterior pole) were available for all of the patients included in the study. In 64 eyes (79.0%) the color photograph was obtained on the same day as OCTA and within 4 weeks in 10 eyes (12.3%). The remaining 7 eyes were excluded from this analysis. IRH were present in half of the study eyes (50.0%) and, in 16.7% of gradable images, total size of IRH was larger than 1 DA. A summary of RVO features graded using color fundus photography is presented in Supplemental Table 1 (Supplemental Material available at AJO.com ).

FA was available for 58 eyes (71.6%) and 29 (50.0%) met the inclusion criteria. From the 29 FAs encompassed in our analysis, 23 (79.3%) were done on the same day as OCTA and 6 (20.7%) were done within 4 weeks, 3 of the 29 FAs had an anti-VEGF intravitreal injection between the 2 scans (10.3%). The majority of the patients were seen in the acute phase (≤3 months) and most features could be graded with confidence, independently of the machine used for the fluorescein angiogram. A summary of the FA results can be seen in Supplemental Table 2 (Supplemental Material available at AJO.com ).

Disease Features on Optical Coherence Tomography Angiography

OCT angiography was acquired in all study eyes (N = 81). However, scanning area and number of scans varied: 79 out of 81 (97.5%) had a 3 × 3-mm scan centered on the macula, 71 (87.7%) had an 8 × 8-mm scan centered on the macula, and 49 (60.5%) had a 4.5 × 4.5-mm scan centered on the disc. Forty-six out of 81 eyes (56.8%) had all 3 scans. Most patients had the OCT angiogram in the chronic phase of the disease. When quality of the scan was adequate or fair, grading could be done with confidence for most features, except for the FAZ radius. In grading the disc scan, confidence for presence of NVD and NVE was high, but it was low for collaterals, as their presence could be affirmed with confidence in 26.1% of cases when they were suspected. When we compared some graded features between the superficial and deep vascular layers, we estimated that differences between the 2 layers were lowest for the presence of collaterals (in the 8 × 8-mm scan) and highest for FAZ radius and area of capillary changes (when graded in the 3 × 3-mm scan). ME was frequently associated with differences in the graded features between the 2 vascular layers, particularly for capillary nonperfusion (χ ² = 13.54; P < .001). Features graded on OCTA can be seen in Table 2 for the 3 × 3-mm and for the 8 × 8-mm scans and in Table 3 for the 4.5 × 4.5-mm scans centered on the disc. An example of images with grading grids can be seen in Figure 1 .

Multimodal analysis with optical coherence tomography (OCT) angiography, color fundus photography, and fluorescein angiography of a 65-year-old woman with a branch retinal vein occlusion of the right eye, exemplifying good correspondence between these imaging techniques. (Top) The upper image corresponds to a montage with OCT angiography images (8 × 8-mm centered on the macula, 6 × 6-mm of the periphery, and 4.5 × 4.5-mm disc scans). (Middle) Images correspond to the OCT angiography 3 × 3-mm macular scan (left), the 8 × 8-mm macular scan (center left), and the 4.5 × 4.5-mm disc scan (center) used for grading. The preretinal nature of the disc neovascularization on the disc OCT angiography scan (center) could be identified in the superotemporal quadrant if the automatically segmented vitreous scan (center right) or the co-registered B-scans (right) were analyzed. (Bottom) Row corresponds to the color fundus photograph (left) and mid-phase (center) and late-phase fluorescein angiography (right) used for grading.

Assessment of Image Quality and Artifacts

When compared to the other imaging modalities, quality of scans was a major issue in OCTA. For fundus photographs, images were classified as adequate for grading in 78.4%, fair in 18.9% and nongradable in 2.7% vs 54.2%, 27.8%, and 18.0% for 3 × 3-mm scans and 66.1%, 23.1%, and 10.8% for 8 × 8-mm scans, respectively. In FA, images were adequate for grading in 82.8%, fair in 13.8%, and nongradable in 3.4% vs 41.4%, 31.0%, and 27.6% for the corresponding 3 × 3-mm scans and 46.2%, 38.5%, and 15.4% for the 8 × 8-mm scans, respectively. The main causes for quality problems in OCTA were motion artifacts (particularly for the 3 × 3-mm) and opacities in ocular media. However, even if 16 out of the 79 3 × 3-mm scans had insufficient quality for grading (20.3%), presence or absence of ischemia could be evaluated in 6 of the 16 scans (37.5%). In the 8 × 8-mm scans, 9 out of 71 (12.7%) had insufficient quality for grading, but ischemia could be evaluated in 5 of these scans (55.6%) ( Figure 2 ). Presence of ischemia, NVD, and NVE could also be inferred in the only nongradable FA. The main variable associated with poor quality of OCTA was worst BCVA (odds ratio = 27.5 [95% CI 5.8–130.1; P < .001] for the 3 × 3-mm and odds ratio = 12.1 [95% CI 2.9–50.3; P < .01], using BCVA as a continuous variable). Factors associated with worst BCVA were also associated with worst-quality scans (ischemic type, previous laser, relative afferent pupillary defect [RAPD], and rubeosis iridis, but not ME or previous injections [data not presented]). We also tested different BCVA cutoffs for worst-quality OCTA 3 × 3-mm scans (0.7, 1.0, 1.3, and 2.0 logarithm of the minimal angle of resolution [logMAR]) and found BCVA worse than 1.3 logMAR as the better predictor for nongradable images (odds ratio = 44.25 [95% CI 9.7–202.1; P < .001]).

A 49-year-old patient with central retinal vein occlusion, macular ischemia, and hand movement visual acuity. Owing to motion artifacts, a low signal, and segmentation errors, the optical coherence tomography angiography 8 × 8-mm scan was considered nongradable (Right). However, the ischemic nature of the central retinal vein occlusion could be correctly assessed and was confirmed by fluorescein angiography (Center). A color fundus photograph is added for comparison (Left).

Agreement Between Fluorescein Angiography and Optical Coherence Tomography Angiography Features of Retinal Vein Occlusion

Agreement between FA and OCTA was tested when both examinations were available and gradable ( Table 4 ). Twenty-one eyes had a gradable 3 × 3-mm and 21 a gradable 8 × 8-mm scan. One eye was excluded from both analyses owing to insufficient quality of FA and OCTA. There was good agreement between both imaging techniques for area of ischemia (ICC 0.825 [95% CI 0.616–0.925] for the 3 × 3-mm and 0.891 [95% CI 0.752–0.954] for the 8 × 8-mm), values that increased when OCTA image quality was adequate. Agreement between area of capillary changes was still significant, but poorer. Figure 3 shows the agreement for capillary nonperfusion and capillary changes by Bland-Altman plot. For the FAZ grading, agreement was good for evaluated parameters between the FA and the 3 × 3-mm scan, but poor between the FA and the 8 × 8-mm scan. For the presence of collateral vessels (in BRVO/HRVO) agreement was poor between FA and OCTA. In NVE, agreement was substantial between the FA and the 8 × 8-mm scan ( Figure 1 ). Seventeen patients had both an FA and an OCTA of the disc. Agreement between NVD in the 2 imaging techniques was good (15 out of 17 eyes [88.2%] were graded the same way), but because only 1 eye had definite neovascularization, kappa value was 0.433 (95% CI 0–1.000).

Table 4

Agreement Between Graded Features on Fluorescein Angiography and Optical Coherence Tomography Angiography (3 × 3-mm Scans [N = 21 Eyes] and 8 × 8-mm Scans [N = 21 Eyes] Centered on the Macula) in Retinal Vein Occlusion Patients

	3 × 3-mm Scan				8 × 8-mm Scan
	FA	OCTA	Coefficient	95% CI	FA	OCTA	Coefficient	95% CI
Area of ischemia (DA), mean ± SD
All examinations	0.32 ± 0.37 a	0.33 ± 0.38 a	0.825 b	0.616–0.925	1.55 ± 2.16 a	1.44 ± 1.66 a	0.891 b	0.752–0.954
Only OCTA 3 × 3-mm adequate quality	0.33 ± 0.36 a	0.34 ± 0.42 a	0.896 b	0.676–0.969	1.29 ± 1.86 a	1.30 ± 1.70 a	0.985 b	0.973–0.998
Area of capillary changes (DA), mean ± SD
All examinations	0.24 ± 0.27 a	0.34 ± 0.33 a	0.712 b	0.413–0.874	1.22 ± 1.39 a	0.91 ± 1.02 a	0.787 b	0.514–0.910
Only OCTA 3 × 3-mm adequate quality	0.21 ± 0.22 a	0.24 ± 0.29 a	0.812 b	0.494–0.944	1.04 ± 1.44 a	0.93 ± 1.21 a	0.926 b	0.760–0.979
Foveal avascular zone radius (all examinations), no.			1.000 c , d	1.000–1.000			0.156 c , d	0–0.508
<300 μm	1	1			2	1
300 μm	1	1			1	1
300–500 μm	18	17			17	16
>500 μm	0	0			0	1
Cannot grade	1	2			1	2
Foveal avascular zone radius (only adequate quality), no.			1.000 c	1.000–1.000			−0.065 c	−0.156 to 0.027
<300 μm	1	1			1	0
300 μm	1	1			0	1
300–500 μm	9	9			10	10
>500 μm	0	0			0	0
Foveal outline (all examinations), no.			0.799 c	0.612–0.986			0.600 c	0.367–0.834
Outline normal	9	11			8	13
Outline questionable	5	3			5	2
Outline destroyed <1/2	7	7			6	6
Outline destroyed <1/2, but with remnants	0	0			2	0
Foveal outline (only adequate quality), no.			0.904 c	0.722–1.000			0.732 c	0.419–1.000
Outline normal	6	7			6	7
Outline questionable	1	0			1	1
Outline destroyed <1/2	4	4			3	3
Outline destroyed <1/2, but with remnants	0	0			1	0
Collaterals (for BRVO/HRVO), no			0.242 c	0–0.554			0.211 c	0–0.619
Absent	3	8			2	4
Present	5	0			5	2
Questionable	2	2			2	3
Neovascularization elsewhere, no. e							0.667 c	0.303–1.000
Absent					16	17
Present					4	2
Questionable					0	1

Only gold members can continue reading. Log In or Register to continue

Share this:

• Click to share on Twitter (Opens in new window)
• Click to share on Facebook (Opens in new window)

Related

Related posts:

1. The Ahmed Baerveldt Comparison (ABC) Study: Long-Term Results, Successes, Failures, and Complications
2. Peripheral 360 Degree Retinotomy, Anterior Flap Retinectomy, and Radial Retinotomy in the Management of Complex Retinal Detachment
3. Comparison of Intraocular Pressure Measurements Obtained by Rebound, Noncontact, and Goldmann Applanation Tonometry in Children
4. Safety of Spectacles for Children’s Vision: A Cluster-Randomized Controlled Trial

Stay updated, free articles. Join our Telegram channel

Join