This is a retrospective, cross-sectional study performed on 105 patients diagnosed with DR by the Department of Ophthalmology of Shenzhen People's Hospital between January 2020 and May 2023. All the research and measurements were conducted in compliance with the tenets of the Declaration of Helsinki. The study was approved by the ethics committee of the hospital and informed consent was obtained from all individuals after a detailed discussion of the nature and possible consequences of the study procedures.

All participants underwent standardized multimodal ophthalmic evaluations including: slit-lamp biomicroscopy examination, Fundus Fluorescein Angiography (Topcon, TRC 50IA; Tokyo, Japan), OCTA (Zeiss, CIRRUS AngioPlex, Dublin, CA). Demographic and clinical factors including age, sex, history of hypertension, and serum lipid profile, glucose metabolism, leukocytes number, duration of diabetes and the grade of diabetic retinopathy were also reviewed. Inclusion criteria were as follows: (1) age above 18 years; (2) type 1 or 2 diabetes mellitus; and (3) the grade of DR was based on the international clinical diabetic retinopathy disease severity scale and DME was based on international clinical diabetic macular edema disease severity scale proposed by American Academy of Ophthalmology in 2019. Exclusion criteria were: (1) presence of any other retinal disorder such as retinal vein occlusion, retinal detachment, uveitis or any other maculopathy; (2) other ocular condition that compromises media opacities (i.e., vitreous hemorrhage or mature cataract); (3) previous treatment with any intraocular surgery, laser photocoagulation or intravitreal injection of anti-VEGF or corticosteroids drug; (4) clearly identify patients taking lipid-lowering drugs such as fenofibrate, statins et al.; and (5) poor quality OCTA images defined by the signal strength index of <5/10.

Macular-centered 6 × 6 mm scan grids (350 × 350 A-scan density) were acquired using Cirrus 5000 Angioplex. Optical microangiography algorithms generated three-dimensional vascular maps through semi-automated segmentation (AngioPlex Metrix v10.0, an integrated analytical software suite for Cirrus 5000 Angioplex and available through two primary channels: directly via the OCT imaging system or through the ZEISS FORUM platform), which is comprise of structural OCT B-scans with flow overlays, superficial/deep vascular plexus en face projections and depth-resolved capillary density maps. Vascular layers were segmented using validated anatomical landmarks: the superficial vascular plexus was defined as extending from the internal limiting membrane to 9 μm above the inner plexiform layer-inner nuclear layer (IPL-INL) junction, and the deep vascular complex encompassed the region from 9 μm above the IPL-INL junction to 9 μm below the outer plexiform layer-outer nuclear layer (OPL-ONL) junction. Corresponding structural en face OCT images were analyzed using identical segmentation slabs to ensure anatomical correspondence. The walls of cystoid spaces are typically formed by adjacent parenchyma or thin, highly reflective lines resulting from optical property differences at the boundaries following hyporeflective fluid accumulation. The border of a serous retinal detachment is defined as the demarcation zone between the detached neurosensory retina and RPE, with underlying hyporeflective subretinal fluid causing an abrupt optical reflectivity transition. All automated segmentations underwent rigorous quality assessment: (1) B-scans with flow overlay were systematically reviewed to identify HRMs or hyperreflective spots exhibiting decorrelation signals; (2) segmentation accuracy was verified through dynamic B-scan navigation using orthogonal (horizontal/vertical) reference lines; and (3) manual corrections were applied when segmentation errors were detected, particularly at layer boundary transitions.

Four distinct categories of pathological microvascular alterations detectable by OCT angiography (OCTA) were analyzed and their morphological classifications were established based on well-characterized criteria from prior validated studies (16, 17). Specifically: (1) microaneurysms were defined as focal capillary dilations (25–100 μm diameter) appearing as hyperreflective dots on OCTA, corresponding to the deep-red punctate lesions observed clinically via ophthalmoscopy; (2) intraretinal microvascular abnormalities (IRMAs) were identified as aberrant, tortuous microvascular structures exhibiting abnormal arteriolar-venular connections within the retinal layers, distinguished by their characteristic branching pattern without vitreal protrusion; (3) retinal neovascular membranes were characterized by the presence of abnormal blood flow signals above the inner limiting membrane, indicating pathological angiogenesis; and (4) non-perfusion area manifested as discrete regions of capillary dropout with complete absence of detectable vascular flow on OCTA imaging.

Two independent masked retinal specialists (L.Z. and H.S.) performed qualitative and quantitative assessments of HRMs using registered B-scan and en face OCT/OCTA images. A senior retinal specialist (M.M.Y.) served as adjudicator for discordant cases to ensure consensus. Each HRMs was evaluated based on the following standardized criteria: (1) anatomical location [inner retina defined as layers between inner limiting membrane and inner nuclear layer, outer retina defined as layers below the inner boundary of the outer nuclear layer (18)], (2) size (≥30 μm or ≤ 30 μm), (3) presence or absence of back shadowing, (4) reflectivity [two levels of hyperreflectivity were identified: moderate reflectivity similar to normal NFL and high reflectivity similar to RPE-Bruch membrane (18)], (5) visible or invisible on structural en face images, and (6) decorrelation signal (9). For systematic analysis, we employed a standardized localization method: (1) each HRMs was sequentially identified across 350 linear B-scans. (2) Precursor cursor placement on B-scan images enabled: exact HRMs localization in the z-axis and automated co-registration with corresponding en face OCTA maps. (3) This spatial mapping protocol ensured precise correlation between cross-sectional structural features (B-scan) and microvascular characteristics (en face OCTA).

Inter-rater reliability was quantified via intraclass correlation coefficients (ICCs) with 95% confidence intervals. To examine the association between HRMs subtypes and retinal vascular pathology, we performed the following analyses: (1) the Kruskal–Wallis test was employed to compare HRMs counts across the four predefined categories of abnormal vascular morphology; (2) Post-hoc pairwise comparisons with Dunn–Bonferroni correction were conducted for significant findings. Generalized estimating equations (GEE) with exchangeable correlation matrices were implemented to address clustered data structures in two analytical contexts: (1) comparative analysis of HRMs burden between DME (+) and DME (–) cohorts, (2) systematic evaluation of HRMs characteristics against seven metabolic covariates including diabetes duration, hypertension status, DR severity (graded per AAO criteria), glucose homeostasis markers [fasting blood glucose (FBG), HbA1c], lipid profile parameters (TG, TC, HDL, LDL, and APO-A/B), systemic inflammation (leukocyte count), and renal function (urinary albumin-creatinine ratio, UACR). (3) Correlation of HRMs with visual function. All analyses were performed using SPSS v27 (IBM) with statistical significance set at P < 0.05 (two-tailed). Effect sizes with 95% confidence intervals are reported for significant associations.

The study cohort comprised 105 treatment-naïve diabetic retinopathy (DR) patients (69 males, 36 females; mean age 55.0 ± 8.1 years) undergoing standardized OCTA imaging, yielding 205 evaluable eyes (104 right, 101 left). Table 1 summarizes the demographic characteristics. DR severity stratification revealed: 21.5% (44/205) mild non-proliferative DR (NPDR), 11.2% (23/205) moderate NPDR, 36.6% (75/205) severe NPDR, and 30.7% (63/205) proliferative DR (PDR). Diabetic macular edema (DME) was present in 52.2% (107/205) of eyes, with morphological subtypes distributed as: cystoid (21.0%, 23/107), mixed pattern (23.9%, 25/107), diffuse (6.3%, 7/107), and serous (1.0%, 1/107).

Six distinct HRMs subtypes were systematically classified through multimodal image analysis (Figure 1), defined by anatomical localization, morphometrics, and signal characteristics: (1) inner retinal hyperreflective spots (IRHFs, inner retina, size ≤ 30 μm, reflectivity similar to nerve fiber layer or RPE, invisible on en face images, absence of back shadowing, Figure 1A green arrow); (2) outer retinal hyperreflective spots (ORHFs, outer retina, size ≤ 30 μm, reflectivity similar to RPE or nerve fiber layer, invisible on en face images, absence of back shadowing, Figure 1A white arrows); (3) inner retinal hard exudates (IRHE, inner retina, size >30 μm, hyperreflective, visible on en face images, presence of back shadowing, Figure 1B yellow circle and arrows); (4) outer retinal hard exudates (ORHE, similar to IRHE but located in outer retina and invisible on superficial en face images, Figures 1C, 2E, F, yellow arrows); (5) decorrelation-positive HRMs (size>30 μm, hyperreflective, invisible on en face images, presence of back shadowing and non-vascular decorrelation signal, Figure 1D blue arrows); (6) cotton-wool spot (size>30 μm, located in the RNFL, hyperreflective, absence of back shadowing, Figure 1E red circle and arrows).

Quantitative analysis revealed distinct topographic distributions of hyperreflective materials (HRMs) across four retinal vascular abnormalities visualized by OCTA, with excellent intergrader reliability (Supplementary Table S2). The majority of IRHFs (Table 2, 75.4%, P < 0.0001, ICC = 0.991) and ORHFs (Table 2, 89.5%, P < 0.0001, ICC = 0.999) colocalized with intraretinal microvascular abnormalities (IRMA), characterized by segmentally tortuous capillaries connecting arterioles and venules in both superficial (Figures 3A1–A3, red rectangle in Figure 3A1 indicating IRMA, green arrows indicating IRHFs) and deep vascular plexuses (Figures 3B1–B3, areas between black arrow heads in Figure 3B1 indicating IRMA, white arrows indicating ORHFs). Non-perfusion areas contained 19.0% of IRHFs (Table 2, P < 0.0001, ICC = 0.987) and 8.7% of ORHFs (Table 2, P < 0.0001, ICC = 0.997; Figures 2A, C, white arrows); while neovascular membranes showed minimal HRMs association (3.3% IRHFs and 1.4% decorrelation-positive HRMs; Figures 2D–F, red arrow heads in 3D indicating neovascular area, green arrows indicating IRHFs and blue arrows indicating decorrelation-positive HRMs). En face imaging demonstrated that 90.8% of superficial inner retinal hard exudates (Figure 3C2, yellow circle; Figure 3C3, yellow arrows) and 94.0% of deep outer retinal hard exudates (Figure 3D2, yellow circle; Figure 3D3, yellow arrows) precisely corresponded to IRMA locations (Figures 3C1, 3D1, red rectangle and black arrow heads showed IRMA area; Table 2, both P < 0.0001, ICC = 0.999), compared to 7.2% and 4.0%, respectively in non-perfusion areas (Figure 2B, yellow arrows). Decorrelation-positive HRMs showed strong spatial correlation with IRMA (Table 2, 77.6%, P < 0.0001, ICC 0.930; Figure 3A1 and blue arrows in Figure 3A3) and microaneurysms (Table 2, 21.0%, P = 0.003, ICC = 0.954; Figures 2G, H and blue arrows in Figure 2I), while cotton-wool spots exhibited exclusive localization to non-perfusion regions (Figures 2B, C, red arrows and circle).

To investigate the potential role of HRMs in diabetic macular edema (DME), we stratified patients with DR into two groups: those with DME and those without DME. Quantitative analysis revealed significantly higher HRMs counts in DME eyes across all vascular abnormalities (all P < 0.05, Table 3). Specifically, inner retinal hyperreflective spots were markedly increased in DME eyes within non-perfusion areas [median 8 (0–72) vs. 0 (0–18), P < 0.001], IRMA regions [median 40 (0–140) vs. 0 (0–42), P < 0.001], and neovascular areas [median 2 (0–39) vs. 0 (0–9), P = 0.002]. Similarly, outer retinal hyperreflective spots showed significant elevation in non-perfusion zones [median 2 (0–39) vs. 0 (0–12), P = 0.002], IRMA regions [median 54 (0–216) vs. 0 (0–34), P < 0.001], and microaneurysms [median 1 (0–78) vs. 0 (0–6), P < 0.001]. Hard exudates demonstrated comparable increases, with inner retinal hard exudates (IRHE) and outer retinal hard exudates (ORHE) being significantly more prevalent in both non-perfusion [IRHE: median 2 (0–20) vs. 0 (0–8), P = 0.009; ORHE: median 2 (0–25) vs. 0 (0–5), P = 0.013] and IRMA regions [IRHE: median 12 (0–108) vs. 0 (0–50), P < 0.001; ORHE: median 24 (0–215) vs. 0 (0–54), P < 0.001]. Decorrelation-positive HRMs and cotton-wool spots in non-perfusion areas were also significantly more frequent in DME eyes. Spatial distribution analysis revealed that 46.3% of IRHFs, 51.8% of ORHFs, 56.1% of IRHE, 57.3% of ORHE and 34.8% decorrelation-positive HRMs localized to cystoid space walls, while 11.1%−16.0% of these HRMs subtypes were observed at the outer border of serous retinal detachments (Supplementary Table S1), suggesting preferential accumulation at fluid-tissue interfaces. Among these HRMs, only IRHFs showed a moderate correlation with worse LogMAR acuity (β = 0.1, 95% CI: 0.0–0.2, P < 0.001, Supplementary Table S3), indicating a potential functional on visual function.

Our generalized estimating equation (GEE) analysis presented in Table 4 identified significant associations between systemic metabolic parameters and the counts of hyperreflective materials in patients with DR. The duration of diabetes exhibited an inverse correlation with inner retinal hard exudates (IRHE, β = −0.1, 95% CI: −0.2–0.0, P = 0.037). Furthermore, the severity of DR demonstrated positive correlations with both inner retinal hyperreflective spots (IRHFs; β = 0.7, 95% CI: 0.4–1.0, P < 0.001), outer retinal hyperreflective spots (ORHFs; β = 0.7, 95% CI: 0.2–1.1, P = 0.002) as well as cotton-wool spots (β = −0.6, 95% CI: −0.2–1.0, P = 0.002). Glycemic control markers exhibited divergent relationships: HbA1c correlated negatively with IRHFs (β = −0.1, 95% CI: −0.2–0.1, P < 0.001) and ORHFs (β = −0.1, 95% CI: −0.2–0.1, P = 0.01), whereas fasting blood glucose (FBG) showed positive correlations with IRHFs (β = 0.1, 95% CI: 0.0–0.2, P < 0.001), ORHFs (β = 0.1, 95% CI: 0.0–0.2, P = 0.007), and cotton-wool spots (β = −0.1, 95% CI: 0.0–0.2, P = 0.008). Lipid metabolism parameters demonstrated particularly strong associations, with LDL cholesterol showing positive correlations for ORHFs (β = 1.0, 95% CI: 0.1–1.9, P = 0.028), IRHE (β = 1.4, 95% CI: 0.5–2.4, P = 0.004), ORHE (β = 1.2, 95% CI: 0.2–2.3, P = 0.02), and cotton-wool spots (β = 1.4, 95% CI: 0.5–2.4, P = 0.004). Conversely, apolipoproteins showed inverse relationships: APO a with IRHFs (β = −2.5, 95% CI: −4.3–0.6, P = 0.009) and APO b with ORHFs (β = −3.3, 95% CI: −6.1–0.5, P = 0.022), IRHE (β = −4.6, 95% CI: −7.8–1.4, P = 0.004), and cotton-wool spots (β = −4.6, 95% CI: −7.8–1.5, P = 0.01). Declining renal function, as measured by urinary albumin-creatinine ratio (UCR), correlated with increased IRHFs (β = 0.1, 95% CI: 0.0–0.2, P = 0.015) and ORHFs (β = 0.1, 95% CI: 0.0–0.2, P = 0.023). No significant associations were observed between HRMs counts and age, leukocyte count, or hypertension status. In general, these findings indicate the crucial role of lipid metabolism in the accumulation of hyperreflective materials.