Modular Architecture of Retinal Layers in Diabetic Patients Without Retinopathy

Purpose Diagnosing diabetic retinopathy (DR) in the pre-clinical stage is crucial to reversing DR. This study aimed to compare the retinal thickness changes between healthy controls (HCs) and diabetics without retinopathy (DWORs). For the first time, we would like to introduce the concept of network modularity analysis in studying retinal networks to demonstrate disrupted retinal layer organization as evidence of subclinical retinopathy. Methods This was a cross-sectional study on 156 eyes of HCs and 78 eyes of DWORs. Retinal layer thickness was measured on Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany). Average thickness values from the outer ring of the ETDRS grid (Avg_O) and the inner ring (Avg_I) were calculated for each layer. Mean retinal thicknesses for each layer between the two groups were compared using the t-test. Age-related thickness changes were compared between the groups using Fisher's r-to-z transform. Group-based structural covariance networks were estimated for both DWORs and HCs. Optimal community architecture was estimated using Louvain's modularity. Results Inner retinal layers, namely RNFL_C (HC: 10.16 ± 2.48 µm versus DWOR: 10.85 ± 2.23 µm; p=0.023) and INL_Avg_I (HC: 39.9 ± 3.7 µm versus DWOR: 40.9 ± 3.16 µm; p=0.035), were thicker in the DWOR group compared to the HC group. Outer retinal layers, namely OR_C (HC: 89.9 ± 3.8 µm versus DWOR: 88.7 ± 3.6 µm; p=0.017) and OR_Avg_I (HC: 81.4 ± 3.16 µm versus DWOR: 80.5 ± 2.28 µm; p=0.02), were thinner in the DWOR group compared to the HC group. The central sub-field showed an age-related thickening in retinal nerve fiber layer (RNFL) (r=0.117, p=0.04), GCL (r=0.078, p=0.17), inner plexiform layer (r=0.137, p=0.01), inner nuclear layer (INL) (r=0.29, p≤0.001), outer plexiform layer (r=0.256, p<0.001), and outer nuclear layer (r=0.197, p=0.001) layers in the HC group, which was not seen in the DWOR group. There was an abnormal increase in modularity among DWORs compared to HCs (Qhc=0.47, Qdowr=0.51, p=1.6x10^-8). In the DWOR group, we noted a disruption in the community architecture and minimal inter-community interactions compared to HCs. Conclusion RNFL and INL are thicker in DWORs compared to HCs. Outer retinal layers are thinner in DWORs compared to HCs. On modularity analysis, we noted a disruption in the community architecture in the DWOR group compared to the HC group.