Quantitative optical coherence tomography angiography biomarkers of the choriocapillaris for objective detection of early diabetic retinopathy

Abstract

PURPOSE:

To evaluate quantitative optical coherence tomography (OCT) angiography (OCTA) biomarkers from the choriocapillaris (CC) for detecting early microvascular changes associated with diabetic retinopathy (DR).

MATERIALS AND METHODS:

In this retrospective study, 191 macular OCTA images were analyzed from 78 healthy eyes, 64 eyes from diabetic individuals without clinical signs of DR (NoDR), and 49 eyes with mild nonproliferative DR (NPDR). Five CC biomarkers were extracted from 6 mm × 6 mm enface OCTA images: flow deficit density (FDD), FD number (FDN), mean FD size (MFDS), perfusion intensity density (PID), and normalized blood flow index (NBFI). Flow maps were binarized using Phansalkar local thresholding, and statistical comparisons were performed using one-way analysis of variance and two-sample t-tests.

RESULTS:

All five biomarkers demonstrated significant differences across study groups (P < 0.001). FDD and MFDS were significantly elevated in both NoDR and mild NPDR eyes compared to controls, indicating increased nonperfusion and enlargement of flow voids. FDN decreased with disease severity, indicating spatial consolidation of capillary loss. PID and NBFI, which reflect flow signal intensity, also declined in diabetic eyes, suggesting a reduction in overall CC perfusion consistent with early vascular compromise.

CONCLUSION:

Quantitative OCTA biomarkers of the CC reveal early microvascular changes in diabetic eyes. Among them, FDN and MFDS demonstrated the highest sensitivity to early disease progression. These findings support the use of CC-derived OCTA features as potential imaging biomarkers for detecting and monitoring early diabetic microvascular dysfunction.

Introduction

Diabetic retinopathy (DR) is the most common microvascular complication of diabetes and remains one of the leading causes of preventable vision loss worldwide, affecting about 28.5% of diabetic adults in the United States alone.[1,2,3] In 2010, the global prevalence of diabetes was estimated at 285 million, with projections indicating a 69% increase by 2030.[4] As diabetes prevalence continues to rise, the incidence and clinical burden of DR are expected also to increase. Notably, the early stages of DR are typically asymptomatic and often go undetected until irreversible retinal damage or vision loss has already occurred.[5,6] Consequently, the early detection and diagnosis of DR are critical for preventing disease progression and preserving vision.

Conventional imaging modalities commonly used for DR screening, such as color fundus photography and fluorescein angiography, have notable limitations. Fundus photography provides a two-dimensional view of the retina and lacks the sensitivity to detect subtle or early microvascular abnormalities, particularly those occurring in deeper vascular layers.[7,8] Fluorescein angiography, though more effective in visualizing retinal vasculature, is invasive, requiring intravenous dye injection that carries potential risks of adverse reactions and limits its applicability for routine screening.[9] Optical coherence tomography (OCT) angiography (OCTA) has become an established, noninvasive imaging technique capable of capturing high-resolution, depth-resolved images of the retinal and choroidal microvasculature without the need for exogenous contrast agents.[10] By providing capillary-level visualization of vascular networks, OCTA enables the detection of subtle microvascular alterations, offering considerable promise for identifying early, preclinical changes in diabetic eyes.

Several studies have employed OCTA to detect and classify DR, primarily by assessing microvascular alterations in the superficial and deep capillary plexuses.[11,12,13] While such studies have advanced our understanding of retinal microvascular compromise due to DR, they have largely overlooked the potential contributions of the choriocapillaris (CC). The CC is a densely vascularized layer of the choroid that supplies oxygen and nutrients to the outer retina, including the metabolically demanding photoreceptors.[14] Emerging evidence suggests that perfusion deficits in the CC may precede clinically observable changes in the inner retinal vasculature, positioning it as a potentially sensitive site for early DR detection.[15,16,17] To address this gap, this study focused on quantifying OCTA biomarkers from the CC for the detection of early DR.

Materials and Methods

Study population

In this retrospective study, we evaluated the potential of quantitative OCTA biomarkers derived from the CC for the early detection of DR. The Institutional Review Board of the University of Illinois Chicago (UIC) approved the study protocol, and it adhered to the tenets of the Declaration of Helsinki (Protocol#: 2016-0752). The requirement of informed consent was waived due to the retrospective nature of the study. Participants were recruited from the UIC retinal clinic. A total of 191 OCTA images were included in this study, comprising 78 eyes from 47 healthy controls, 64 eyes from 40 diabetic individuals without clinical signs of retinopathy (NoDR), and 49 eyes from 36 patients diagnosed with mild nonproliferative DR (NPDR) according to the Early Treatment Diabetic Retinopathy Study (ETDRS) scale. Eyes with moderate NPDR, severe NPDR, and proliferative DR were excluded from this study to focus on early-stage microvascular changes. Clinical and demographic information of the study participants are shown in Table 1. Individuals under the age of 18 were excluded. Additional exclusion criteria included the presence of macular edema, prior vitrectomy, significant media opacities, age-related macular degeneration, or any other concurrent retinal pathology. All participants underwent a comprehensive ophthalmic examination. Classification into NoDR or mild NPDR groups was performed by a single retina specialist based on the ETDRS severity scale.

Table 1.

Demographics of study participants

	Controls	NoDR	Mild NPDR
Number of subjects (n)	47	40	36
Age (years)	49.21±18.35	60.45±13.15	62.30±12.85
Age (range)	23–87	33–85	24–78
Gender, n (%)
Male	29 (61.70)	17 (42.50)	18 (50.00)
Female	18 (38.30)	23 (57.50)	18 (50.00)
Number of images	78	64	49

NoDR=No diabetic retinopathy, NPDR=Nonproliferative diabetic retinopathy

Optical coherence tomography angiography image acquisition and preprocessing

Macular OCTA scans were acquired using the AngioVue spectral-domain OCT system (Optovue, Fremont, CA), with a 6 mm × 6 mm scanning protocol centered on the fovea. All scans were manually reviewed for quality, and only images with a signal quality index >6 were included. Motion artifacts were evaluated using criteria described in the review by Anvari et al., and images showing blink lines in more than two quadrants, or moderate to severe quilting, displacement, vessel doubling, or stretch artifact were excluded from the analysis.[18] Enface images of the CC were exported using the system’s built-in ReVue software (version 2018.1.0.43). The software automatically segmented the CC slab as the region extending from 10 μm above to 30 μm below the Bruch’s membrane. Figure 1 shows the segmentation boundaries for the superficial vascular plexus, the deep capillary plexus, and the CC. Figure 2 shows representative CC OCTA images of a control subject, a NoDR subject, and a mild NPDR subject. Exported OCTA images were subsequently processed in MATLAB (MathWorks, Natick, MA), where a custom script was used to extract quantitative biomarkers from the images.

Figure 1.

Optical coherence tomography angiography segmentation of the retinal vascular plexuses and corresponding enface projections. (A) Cross-sectional B-scan showing the segmentation boundaries used to define each vascular plexus. (B1) Enface projection of the superficial vascular plexus, segmented from the inner limiting membrane to 10 μm above the inner plexiform layer (IPL) (red line–green line). (B2) Enface projection of the deep capillary plexus, segmented from 10 μm above the IPL to 10 μm below the outer plexiform layer (green line to blue line). (B3) Enface projection of the choriocapillaris, segmented from 10 μm above the Bruch’s membrane (BRM) to 30 μm below the BRM (orange lines)

Figure 2.

Representative optical coherence tomography angiography (OCTA) images of the choriocapillaris from a control subject (A1 and A2), a NoDR subject (B1 and B2), and a Mild Nonproliferative DR subject (C1 and C2). Row 1 shows the original OCTA image, and Row 2 shows the corresponding pseudo-color intensity maps

Before calculating the OCTA biomarkers, the images were preprocessed to isolate regions of flow and nonflow within the CC. Local thresholding was applied using the Phansalkar method, an adaptive binarization technique widely utilized for OCTA image analysis of the CC. A radius of 15 pixels was chosen for the Phansalkar binarization based on previously published studies demonstrating stable and reliable segmentation of CC FDs in OCTA images.[19,20,21,22,23] By adjusting the threshold based on local intensity variations, this method enables robust segmentation of FDs in the presence of spatial heterogeneity. The resulting binarized flow maps served as the foundation for subsequent quantitative feature extraction.

Quantitative optical coherence tomography angiography biomarkers

In this study, five quantitative OCTA features were analyzed: flow deficit density (FDD), FD number (FDN), mean FD size (MFDS), perfusion intensity density (PID), and normalized blood flow index (NBFI). These OCTA biomarkers have been previously established to quantify retinal vascular changes.[16,17,23,24,25] In this study, we extended these OCTA feature analyses for characterizing CC abnormalities caused by DR. Figure 3 shows the various image processing steps taken to calculate the various quantitative biomarkers.

Figure 3.

Image processing steps in the quantification of optical coherence tomography angiography (OCTA) biomarkers. (a) Representative original OCTA image. (b) Original image with a pseudo-color intensity map. (c) Binarized OCTA map. (d) Pseudo-color perfusion map. (e) Pseudo-color intensity map of background noise removed from OCTA image

Flow deficit density

This is defined as the proportion of the OCTA image lacking detectable flow signal on the binarized flow map [Figure 3c]. It serves as a global indicator of nonperfusion, reflecting the overall extent of capillary dropout or flow impairment within the imaged region. This can be calculated as:

Where C (x, y) represents the pixels representing flow signal in the binarized flow map [Figure 3c] and A (x, y) represents all the pixels in the OCTA image [Figure 3a]. FDD, therefore, represents the proportion of the image area that lacks detectable flow signal.

Flow deficit number

This is defined as the total number of discrete FDs identified within the binarized flow map [Figure 3c]. FDs were identified as contiguous pixel clusters representing areas with absent flow signal, and the total count of these clusters was reported as the FDN. To reduce the inclusion of noise and physiologic voids within normal intercapillary spacing, only FDs with an equivalent diameter larger than 24 μm were counted, based on the normal intercapillary spacing reported by Zhang et al.[26]

Mean flow deficit size

This is defined as the average area of individual FDs identified in the binarized OCTA image [Figure 3c]. It provides a measure of the typical size of nonperfusion regions in the OCTA image. This was calculated as:

Where A_i is the area of the i^th FD and FDN is the total number of FDs identified in the OCTA image.

Perfusion intensity density

This feature is defined as the mean pixel intensity of the nonbinarized OCTA image [Figure 3b]. It reflects the average perfusion signal across the entire OCTA image and serves as a global measure of CC flow intensity. It was calculated as:

PID = μ (Original OCTA Image)

Normalized blood flow index

This is defined as the ratio of the mean pixel intensity of the perfusion map [Figure 3d] to the standard deviation of the pixel intensities of the noise map [Figure 3e]. This approach normalizes the average flow signal by the background variation, thereby reducing the influence of inherent noise in the OCTA image. The perfusion map was generated by masking the original OCTA image using the binarized flow map [Figure 3c], thereby isolating regions corresponding to flow signal. The noise map was constructed from the areas not identified as containing flow signal.

Statistical analysis

All statistical analyses were performed using R Software, version 4.2.0 (R Core Team, Vienna, Austria). The Shapiro–Wilk test was used to assess the normality of each quantitative OCTA feature. For features that followed a normal distribution, one-way analysis of variance (ANOVA) was used for multiple group comparisons, followed by pairwise comparisons using the unpaired Student’s t-test. For features that did not meet the normality assumption, the Kruskal–Wallis test was applied for multiple group comparisons, followed by pairwise comparisons using the Mann–Whitney U-test. A P < 0.05 was considered statistically significant in this study. All statistical tests were two-tailed.

Results

Comparative analysis of the quantitative OCTA biomarkers in control, NoDR, and mild NPDR groups is summarized in Table 2. One-way ANOVA disclosed statistically significant (P < 0.001) differences across all five OCTA metrics, including FDD, FDN, MFDS, PID, and NBFI. FDD was significantly increased in diabetic eyes (NoDR + mild NPDR) compared to controls (P < 0.001), while no significant difference was observed between the NoDR and mild NPDR groups (P = 0.767). FDN and MFDS exhibited opposing trends: FDN decreased, while MFDS increased with disease severity. Both metrics significantly differentiated all three groups (P < 0.05). PID and NBFI decreased with increasing severity, indicating a reduction in blood flow as DR worsens. PID was sensitive in distinguishing diabetic subjects from controls (P < 0.001), but not between NoDR and mild NPDR (P = 0.143). In contrast, NBFI did not differ between controls and NoDR (P = 0.414) but was able to distinguish individuals with no clinical signs of DR (control + NoDR) from those with mild NPDR (P < 0.05).

Table 2.

Comparisons of choriocapillaris quantitative features between controls, no diabetic retinopathy, and mild nonproliferative diabetic retinopathy groups

Feature	Control (I)	NoDR (II)	Mild (III)	P
				I versus II	I versus III	II versus III	ANOVA
FDD (%)	36.86±2.46	38.38±2.09	38.50±2.12	<0.001	<0.001	0.767	<0.001*
FDN (×102)	28.47±4.15	26.57±4.75	23.93±4.39	0.010	<0.001	0.005	<0.001†
MFDS (µm2)	48.25±12.14	54.32±13.50	60.49±14.90	0.002	<0.001	0.022	<0.001†
PID	0.432±0.029	0.400±0.028	0.391±0.032	<0.001	<0.001	0.143	<0.001*
NBFI	3.04±0.08	3.03±0.08	2.98±0.09	0.414	<0.001	0.004	<0.001*

*Multiple group comparisons performed using one-way ANOVA, corresponding individual comparisons were performed using Student’s t-test, ^†Multiple group comparisons performed using Kruskal–Wallis one-way ANOVA, corresponding individual comparisons were performed using Mann–Whitney’s U-test. Statistical significance P<0.05. Data are expressed as mean±SD. FDD=Flow deficit density, FDN=Flow deficit number, MFDS=Mean flow deficit size, PID=Perfusion intensity density, NBFI=Normalized blood flow index, NoDR=No diabetic retinopathy, SD=Standard deviation, ANOVA=Analysis of variance

Discussion

This study evaluated five quantitative OCTA biomarkers derived from the CC to assess their ability to detect early microvascular changes associated with DR. All five biomarkers demonstrated statistically significant differences across study groups, indicating sensitivity to early-stage vascular changes. These findings support the growing recognition of CC dysfunction as an early feature of diabetic microvascular compromise. A study by Le et al. demonstrated that outer retinal disruption was detectable in diabetic eyes without clinical signs of retinopathy and worsened with disease progression.[27] These findings support histologic and metabolic evidence that photoreceptor dysfunction may occur early in diabetes.[28] Parravano et al. further advanced this understanding by linking outer retinal changes directly to CC hypoperfusion, suggesting that reduced CC perfusion compromises photoreceptor metabolic integrity.[16] Importantly, their study found a proportional relationship between the degree of CC FD and the severity of outer retinal dysfunction, confirming that microvascular compromise in the CC may contribute to early retinal neurodegeneration. These observations align closely with the findings of this study, in which CC biomarkers showed significant changes across the early stages of DR.

FDD is arguably the most widely used quantitative OCTA biomarker for assessing CC integrity. It is inversely related to blood vessel density, providing a global estimate of the proportion of the imaged region lacking flow signal. Elevated FDD values reflect a reduction in functional capillary coverage and are commonly interpreted as evidence of capillary dropout. From Table 2, FDD was significantly elevated in both diabetic groups, NoDR and Mild NPDR, compared to controls. This early increase in FDs may reflect diffuse capillary dropout in the CC, consistent with the notion that microvascular compromise precedes observable structural retinal changes. Hyperglycemia-induced oxidative stress, basement membrane thickening, and pericyte loss are known to disrupt choroidal microvasculature,[24] ultimately reducing capillary perfusion even in the absence of visible fundus abnormalities.

Like FDD, MFDS was elevated in diabetic eyes, indicating that individual nonperfused regions enlarged with worsening capillary damage. This trend suggests that as perfusion declines, small flow voids expand through the loss of adjacent capillaries. In contrast, FDN decreased with disease severity, indicating that while nonperfusion became more extensive and individual deficits grew larger, the number of discrete flow voids declined. This pattern supports the idea that as adjacent capillaries are lost, smaller deficits begin to merge, forming fewer but larger regions of nonperfusion. Notably, both FDN and MFDS were able to differentiate all three groups, highlighting their potential utility in detecting not only the presence of microvascular dysfunction but also its progression in the disease process.

Studies have shown that the decorrelation intensity of OCTA images is correlated with blood flow.[29,30] Therefore, an index of overall perfusion can be derived by calculating the average pixel intensity of the nonbinarized image. In this study, two perfusion-based biomarkers, PID and NBFI, were evaluated. While PID represents the mean flow signal across the CC, NBFI normalizes this signal relative to background noise, accounting for the effect of inherent noise in OCTA images. From Table 2, both PID and NBFI were significantly reduced in mild NPDR eyes compared to controls, consistent with decreased CC perfusion in early DR. While PID also distinguished diabetic eyes from controls, it did not significantly differentiate between NoDR and mild NPDR, suggesting limited sensitivity to subtle progression once perfusion begins to decline. In contrast, NBFI was significantly reduced in mild NPDR eyes compared to both control and NoDR groups but showed no difference between controls and NoDR. This suggests that while NBFI may not be sensitive to the earliest perfusion changes, it is more effective in detecting microvascular deterioration once retinopathy begins to manifest.

CC perfusion is an essential parameter in understanding the early microvascular changes associated with DR. This study evaluated five quantitative OCTA biomarkers to characterize perfusion abnormalities in the CC and their relationship to disease severity. Among these, FDN and MFDS demonstrated the highest sensitivity in distinguishing all three groups. The ability of FDN and MFDS to distinguish progressive changes across the early stages of DR highlights their potential as sensitive markers for subclinical disease monitoring. Furthermore, incorporating both structural and intensity-based biomarkers may enhance detection accuracy and offer a more comprehensive view of CC dysfunction, which could ultimately support earlier diagnosis and risk stratification.

This study has certain limitations that should be acknowledged when interpreting the findings. The relatively modest sample size and use of a single OCTA device may limit the generalizability of the results across broader populations and imaging devices. In addition, projection artifact removal was not applied. However, the use of Optovue’s CC slab definition and manual exclusion of scans with visible artifacts helped reduce their impact. Systemic variables known to influence ocular blood flow, such as blood pressure, glycemic control, and diabetes duration, were not included due to limited clinical data availability. In addition, patients with macular edema were excluded to avoid confounding artifacts in OCTA signal analysis; as a result, the current findings may not extend to eyes with edema. Given that different OCTA systems may introduce varying levels of image noise and signal sensitivity, future studies incorporating larger, more heterogeneous cohorts and data from multiple imaging platforms are needed to further validate the reproducibility and clinical applicability of the biomarkers evaluated in this study.

Conclusion

This study highlights the utility of quantitative OCTA biomarkers from the CC in detecting early microvascular changes associated with DR. Among the evaluated biomarkers, FDN and MFDS were most effective in differentiating control, NoDR, and mild NPDR groups. Intensity-based metrics such as PID and NBFI also detected reductions in perfusion in diabetic eyes, reinforcing the relevance of intensity-based measures. These findings support the potential role of CC biomarkers in identifying subclinical vascular alterations and contribute to the growing body of evidence emphasizing the importance of the CC in the early pathogenesis of DR.

Data availability statement

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Conflicts of interest

The authors declare that there are no conflicts of interests in this paper.

Funding Statement

This study was financially supported by the National Eye Institute (R01 EY023522, R01 EY029673, R01 EY030101, R01 EY030842, P30EY001792); Research to Prevent Blindness; Richard and Loan Hill Endowment.