Abstract
Purpose:
The purpose of this study was to assess projection-resolved optical coherence tomography angiography (PR-OCTA) vessel density (VD) and foveal avascular zone (FAZ) in determining severity within diabetic retinopathy (DR) and their accuracy in identifying high-risk DR patients.
Methods:
This was a retrospective study with 72 eyes of 52 DR patients, assessing the VD and FAZ area of the superficial capillary plexus (SCP) and deep vascular plexus (DVP), for both 3 × 3-mm and 6 × 6-mm scans between the DR groups (mild to moderate, severe and proliferative DR [PDR]). For accuracy, the severe and PDR groups were merged, representing the high-risk DR group for receiver operator characteristic analysis. VD of OCTA images with and without PR were compared.
Results:
In mild to moderate, severe, and PDR groups, there were 31, 21, and 20 eyes, respectively. PR-OCTA improved VD analysis only in the DVP and particularly in advanced DR stages (P = 0.042). In the 3 × 3-mm PR scans, all superficial and deep parameters were significantly different between severe and PDR groups (P ≤ 0.020), but only the mean VD of SCP and DVP was also significant between the mild to moderate and severe groups (P ≤ 0.007). In the 6 × 6-mm scans, the superficial VD, deep VD, and superficial FAZ were significantly different between the severe and PDR groups (P ≤ 0.029). The superficial VD and deep VD of the 3 × 3-mm scans were good parameters for detecting high-risk patients (area under the curve = 0.829 and 0.895, respectively).
Conclusions:
PR-OCTA improved VD analysis of DVP. The 3 × 3-mm SCP and DVP VD were the most accurate in detecting high-risk DR.
Diabetic retinopathy (DR) is a severe sight-threatening complication of diabetes mellitus and one of the leading causes of blindness worldwide.
1 The number of patients affected with this disease is expected to grow as diet and exercise habits, including increasingly sedentary lifestyle, change in developing countries.
2,3 Diabetic retinopathy is a microangiopathy that causes capillary occlusion, vascular hyperpermeability, and neovascularization in the retinal vasculature.
4 Patients at greater risk of developing severe complications, including blindness, were those with diffuse retinal ischemia, signs of severe DR, and/or neovascularization and should be considered for early photocoagulation treatment.
5
Many core concepts of DR were established with fluorescein angiography (FA), which is an important diagnostic tool for evaluating disease severity in DR.
6 Leakage, capillary nonperfusion, vascular structural abnormalities, and neovascularization are among the most common features observed in this technique. However, it is time-consuming and invasive, and although considered harmless, the dye poses risks ranging from nausea to allergic reactions, including anaphylaxis and in rare instances death.
7,8
Optical coherence tomography angiography (OCTA) through split-spectrum amplitude-decorrelation angiography (SSADA), is a noninvasive technique that allows three-dimensional mapping of the retinal and choroidal microvasculature.
9 The rapid acquisition time, the dispensing of intravenous dye, and the ability to identify capillary abnormalities without being obscured by leakage or staining are some of the advantages over FA. Another advantage is the ability OCTA has to clearly distinguish between the superficial capillary plexus (SCP) and deep vascular plexus (DVP), which is not possible on FA and can potentially affect the clinical evaluation.
10
In a study by Agemy et al.,
11 they described a quantitative analysis for retinal vascular density in DR patients with OCTA, which agreed closely with grading based on clinical features. A novel projection-resolution (PR) algorithm for OCTA improves visualization of the retinal plexuses by decreasing the projection artifacts throughout the whole volume scan.
12 Hwang et al.
12 showed a greater sensitivity in distinguishing DR from healthy control eyes with this novel OCTA algorithm.
Therefore, this volumetric retinal vessel assessment has provided novel parameters for DR progression, because it can be easily performed and repeated frequently at visits. Hence, our aim is to evaluate the performance of PR-OCTA flow parameters (i.e., vessel density [VD] and foveal avascular zone [FAZ]) in the DR stages and their accuracy in detecting high-risk patients.
A retrospective chart review of patients with DR was conducted at the Neovista Eye Center in Americana, Brazil. Patients were classified by one grader (WB) based on color fundus photography (FP), according to the International Council of Ophthalmology,
5 into three groups: mild to moderate, severe, and proliferative diabetic retinopathy (PDR). The grading was then compared with the ground truth diagnosis performed by a retina specialist (AR) with fundus examination, FP, and FA.
Due to automated segmentation errors of the superficial and deep layers that can occur in retinal anatomical distortions, all eyes with diabetic macular edema (total of 46 eyes of 28 patients) were excluded to avoid manual segmentation, which is time-consuming and beyond the scope of this study. Additionally, patients associated with greater than mild hypertensive retinopathy were excluded (9 patients; 14.7%). Furthermore, eyes with poor quality OCT images due to poor fixation or excessive motion artifacts, as well as media opacity, were excluded from the study (total of 13 eyes; 15.3%). The proliferative group had the most images with artifacts, with six eyes (46.1%) of the total excluded. The study was performed in adherence to the tenets of the Declaration of Helsinki and was approved by the Ethics Committee of Faculdade São Leopoldo Mandic (CAAE: 62362616.4.0000.5374).
Cohen's κ was used to evaluate the agreement of the grader's DR classification based on FP. To demonstrate the effect PR-OCTA had on VD in DR disease, Wilcoxon signed rank tests were used to assess whether there was a significant difference between the measurement with and without PR for the SCP and DCP of their respective scan sizes and expressed with their P value. Second, Kruskal-Wallis tests were used to assess the differences between DR stages in each of the parameters in their respective scan size. The between- and within-grader variability for the manual FAZ measurements on PR-OCTA was demonstrated with inter- and intraclass correlation coefficient and Bland-Altman plots (comparing the difference of the FAZ measurement between graders of each subject with their mean and limits of agreement) for the superficial and deep FAZ of each scan size.
A linear mixed model was used to study the effect of DR group of the OCTA parameters as described previously.
16 Superficial or deep VD was entered as a dependent variable at each resolution (superficial VD with the 3 × 3-mm scan, deep VD with the 3 × 3-mm scan, superficial VD with the 6 × 6-mm scan, deep VD with the 6 × 6-mm scan). DR stages were entered as a fixed factor, subject as a random factor, and age and corresponding FAZ as covariates. Pairwise comparisons with least significant difference (LSD) correction were performed for each VD layer and scan size between the DR stages.
A linear mixed model analysis was also conducted with superficial or deep FAZ as the dependent variable at each resolution (superficial FAZ with the 3 × 3-mm scan, deep FAZ with the 3 × 3-mm scan, superficial FAZ with the 6 × 6-mm scan, deep FAZ with the 6 × 6-mm scan). DR stages were entered as a fixed factor, subject as a random factor, and age and corresponding VD as covariates. Similarly, pairwise comparisons with LSD correction were performed for each FAZ layer and scan size between the DR stages.
Last, to determine the accuracy of these parameters in detecting high-risk DR stages, the severe and PDR groups were merged to represent these high-risk patients, and then we performed receiver operator characteristic (ROC) curves of the 3 × 3-mm and 6 × 6-mm parameters. These were represented with their area under the curve (AUC) and respective 95% confidence intervals and P values. Moreover, their cutoff values, sensitivity, and specificities were calculated. SPSS 21.0 (IBM Corp. Released 2013, IBM SPSS Statistics for Windows, Version 21.0; IBM Corp., Armonk, NY, USA) was used for all parts of the statistical analysis described above.
A total of 72 eyes of 52 patients (22 males and 30 females) with DR were included. The mean age of the patients was 57.6 ± 14.4 years and ranged from 26 to 81 years. The mean age of the mild to moderate, severe, and PDR groups were 65.8 ± 9.7, 56.4 ± 12.1, and 46.2 ± 15.0 years, respectively. Moreover, the male distribution (% of subjects in each group) was 32.3%, 61.9%, and 45.0% and the diabetes type 2 distribution was 93.5%, 95.2%, and 66.7% in the previously described groups, respectively.
The number of eyes and the distribution of parameters in the superficial and deep FAZ and VD of both scan sizes in each DR group are summarized in
Table 1. The PR-OCTA parameters are expressed by the mean values with their SD. The mean spherical equivalent was −0.21 ± 1.62 diopters (D), and the range was −5.25 to +4.25 D. There were 10 eyes (13.8%) with greater than ±2.00 D, in which 6 eyes (8.3%) were greater than ±3.00 D and 2 eyes (3.0%) were greater than ±4.00 D; their mean FAZs were 0.386 ± 0.170, 0.385 ± 0.204, and 0.556 ± 0.234 mm
2, respectively.
Table 1 PR-OCTA Parameters Between the Diabetic Retinopathy Stages in the 3 × 3-mm and 6 × 6-mm Scans
Table 1 PR-OCTA Parameters Between the Diabetic Retinopathy Stages in the 3 × 3-mm and 6 × 6-mm Scans
There was good agreement between the grader's DR classification and the patients' diagnosis (κ = 0.749; SEM = 0.065;
P < 0.001). In the 3 × 3-mm scans, the VD for the PR-OCTA of SCP and DVP was 42.7 ± 4.9% and 36.9 ± 5.0% and the VD for the OCTA without PR was 43.1 ± 4.9% and 43.7 ± 4.3%, respectively. There was a significant difference between the VD with and without PR algorithm in the deep layer (
P < 0.001) but not in the superficial layer (
P = 0.710). The mean VD distribution between the DR groups is shown in
Figure 6, highlighting a significant difference of the deep VD between the severe and PDR groups with PR-OCTA (
P = 0.042) but not with conventional OCTA (
P = 0.107). In the 6 × 6-mm scans, the VD with PR-OCTA of the aforementioned layers was 39.7 ± 2.7% and 37.8 ± 3.6% and the VD without PR was 41.1 ± 1.0% and 44.7 ± 4.0%, respectively. There was a significant difference between the conventional and PR-OCTA in the superficial and deep VD (
P = 0.026 and
P < 0.001, respectively), but no significant differences between groups were seen.
In the PR-OCTA analysis, the interclass correlation coefficient for the manual adjustments of the FAZ areas in the 3 × 3-mm scans (superficial and deep) and the 6 × 6-mm scans (superficial and deep) were 0.978, 0.935, 0.968, and 0.885, respectively. There was no bias between measurements, and the Bland-Altman plots are shown in
Supplementary Figure S1. Likewise, the intraclass correlation coefficient of the FAZ area repeated measurements were 0.987, 0.959, 0.972, and 0.901, in the respective previous groups; there was no bias, and their Bland-Altman plots are shown in
Supplementary Figure S2.
For the 3 × 3-mm analysis, in both superficial and deep layers of the VD and superficial and deep layers of the FAZ, there were significant differences in the DR stages as demonstrated in
Table 1. Further in the pairwise comparisons (
Table 2), the superficial and deep VD showed a significant difference among all three groups (
P ≤ 0.017). The superficial and deep FAZs were significantly different between the PDR and severe stages (
P ≤ 0.020) and between the PDR and mild to moderate stages (
P ≤ 0.017), but there was no significant difference between the severe and mild to moderate stages. Moreover, in our models, age and respective FAZ in the VD parameters were not a significant factor, but superficial VD was significant for the superficial FAZ parameter (
P = 0.001).
Table 2 Pairwise Comparisons With LSD Correction of the PR-OCTA Parameters Between the DR Stages in the 3 × 3-mm and 6 × 6-mm Scans
Table 2 Pairwise Comparisons With LSD Correction of the PR-OCTA Parameters Between the DR Stages in the 3 × 3-mm and 6 × 6-mm Scans
For the 6 × 6-mm scans, the pairwise comparisons in
Table 2 showed a significant difference in the superficial and deep VD between the PDR and severe stages (
P ≤ 0.029) and between the PDR and mild to moderate stages (
P ≤ 0.003). Only the superficial FAZ showed significant differences between the PDR and severe stages (
P = 0.023) and between the PDR and mild to moderate stages (
P = 0.009) but not the deep FAZ. Furthermore, the age and the respective FAZ in the VD parameters or respective VD in the FAZ were not significant factors in our models.
Furthermore, the ROC analysis (
Table 3) showed that the superficial and deep VDs in the 3 × 3-mm scans were good parameters to detect high-risk DR patients (AUC = 0.829 and 0.895, respectively;
P < 0.001). However, the superficial and deep FAZs were poor parameters to detect high-risk DR patients (AUC = 0.690 and 0.694, respectively;
P ≤ 0.006). In the 6 × 6-mm scans, the superficial VD was fair in detecting these patients (AUC = 0.702,
P = 0.005), whereas the deep VD was considered a poor parameter (AUC = 0.699,
P = 0.006). At a ≥80% specificity, the sensitivity of the superficial and deep VDs were 75% and 78% in the 3 × 3-mm scans, respectively. At a ≥80% sensitivity, both their specificities were 79%. The sensitivity and specificity with their cutoff values are shown in
Supplementary Table S1.
Table 3 ROC Analysis of the PR-OCTA Parameters to Detect High-Risk DR in the 3 × 3-mm and 6 × 6-mm Scans
Table 3 ROC Analysis of the PR-OCTA Parameters to Detect High-Risk DR in the 3 × 3-mm and 6 × 6-mm Scans
In our study, PR-OCTA has shown that in advanced stages of DR, the FAZ increases and the VD decreases significantly (
Figs. 2–
5). These findings are in agreement with previous studies
11,17,18 that performed OCTA analysis, further suggesting that these parameters in the superficial and deep plexus can be considered biomarkers for DR.
PR-OCTA greatly suppresses the pixels from the more superficial vessels that are projected into the deeper layers and better reveals the vascular abnormalities in DR.
12,13 We showed in our adjusted sample that the PR algorithm significantly altered the 3 × 3-mm VD in the DCP (
P < 0.001) and not in the SCP and improved the differentiation between PDR and severe DR in the former layer. Interestingly, in the PDR group without PR, there was a vessel dropout in the 3 × 3-mm SCP, with mean VD even lower than the DCP (
Fig. 6A), whereas in the PR-OCTA, the VD decrease in both layers was proportional as disease progressed (
Fig. 6B). This suggests that in PDR, the projection artifacts from the superficial plexus are masking the deep VD. Another consideration is that the previous SCP segmentation depth included the intermediate capillary plexus, but with the PR, there is better segmentation of the three plexi respecting their known anatomical boundaries.
12
In DR, the capillary nonperfusion and loss of retinal capillaries can lead to progressive retinal hypoxia, inducing expression of angiogenic growth factors and eventually neovascularization.
19–21 In this context, OCTA is capable of detecting early signs of DR, with FAZ increase, capillary nonperfusion, and even microaneurysms, in a noninvasive, reproducible, and repeatable way.
11,18,22,23 In our study, the 3 × 3-mm superficial and deep VDs were the best parameters because they were significantly different between all DR stages. The PDR stage caused a significant alteration of all parameters in the 3 × 3- and 6 × 6-mm scans, except for the deep FAZ in the 6 × 6-mm scans. An interesting observation in our study was that the VD of both scan sizes showed a greater difference in the DVP between the advanced stages (severe and PDR) and a greater difference in the SCP between the early stages (mild to moderate and severe).
Furthermore, high-risk DR patients are at greater risk for developing severe and irreversible ocular complications, requiring early intervention.
5 We showed that the 3 × 3-mm superficial and deep VDs were most accurate in identifying these patients (AUC = 0.829 and 0.895, respectively) and that the deep VD showed a slightly higher sensitivity (78%) compared with the superficial VD (75%) at ≥80% specificity (
Table 3;
Supplementary Table S1). Our results suggest that the VD of both layers are accurate in diagnosing these patients and can help assess their management and treatment, in particular when considering the potential for automation and artificial intelligence in ophthalmic diagnostic imaging.
24 However, caution is needed to interpret these vessel parameters, because they can vary with different vessel analysis methods and may not be interchangeable between different OCTA devices.
25 The technology and methodology are constantly improving, and future prospective studies are warranted to help establish these OCTA biomarkers in DR.
The importance of assessing vascular alterations in the DVP is attributed to the anatomic location and characteristics of this layer. There is evidence that the DCP has a significant capillary dropout in advanced stages of DR,
12,17,26,27 which can then contribute to photoreceptor loss and visual acuity decrease.
27–29 The theory is that the DCP is more vulnerable to ischemic insult, as this layer resides in a watershed zone of oxygen supply.
11 Our results support this theory and highlight the role it may play in the pathogenesis of DR and other ischemic vasculopathies.
30–32 In that regard, although conventional OCTA can mask the alterations in the deep layer, PR-OCTA has potentially improved the accuracy for detecting high-risk DR patients, as DVP seems to be more sensitive to hypoxia.
In general, the 3 × 3-mm parameters were more significant than the 6 × 6-mm parameters. The reason for this is the greater scan density and resolution generated by the former. Put in perspective, to increase scan area from 3 × 3 to 6 × 6-mm, a speed four times greater is required to maintain the same scan density and time. In the future, faster scanning speed will be essential for a larger field of view with higher resolution and less acquisition time.
33 Of notice, a novel high-definition 6 × 6-mm scan with greater scan density is available and therefore might have a better performance than the one reported in this study.
Last, the FAZ area has been most frequently studied in the literature and is considered by some a robust parameter in DR with a high reliability even between different OCTA scan sizes.
34 Although our results showed a high repeatability of the FAZ measurements between graders, they were not accurate in detecting high-risk DR and showed significant differences between scan sizes.
The limitations of our study are its retrospective nature and the moderate number of cases in each DR stage. By excluding eyes with macular edema, there was a significant reduction of the sample size. Moreover, it would be interesting to compare with healthy controls and diabetic patients without DR to assess the diagnostic accuracy as a screening tool. However, our main purpose was to assess the disease severity within DR patients. Further, the influence that axial length has on OCTA FAZ measurements has been reported.
35 However, most eyes in our study (86.2%) were within the ±2.00-D range, and only 8.3% had greater than ±3.00 D, in which the reported relative change after correction was ≤10.0% and ≤25.0%, respectively. Therefore, we believe that this had a minor impact on the FAZ analysis of our study.
Another limitation was that patients with previous laser photocoagulation treatment were not excluded. Although the treatment was reported over a year prior to the consultation, this could still potentially influence macular perfusion and should be considered in future prospective studies. Last, OCTA provides relatively small frames, and the relation between peripheral disease and OCTA macular perfusion is not yet clear in DR. We believe OCTA greatly contributes as a quantitative complimentary tool for DR but does not substitute other modalities at present.
In conclusion, our results showed that the superficial and deep VDs of the 3 × 3-mm scan were the best parameters to assess severity in DR patients. These results suggest that the measurement of VD on OCTA has the potential to be a quantitative biomarker for DR severity.
The authors thank Stephanie Cox, OD, for critical statistical guidance.
Supported by ARVO's publications grant (Rockville, MD, USA).
Disclosure: W.W. Binotti, None; A.C. Romano, Optovue (F), Zeiss Meditec (F)