Retinal neovascularization as self-organized criticality on ultra-widefield fluorescein angiography imaging of diabetic retinopathy

Background/objectives: Do the distributions of surface area of non-perfusion (NP) and neovascularization (NV) on ultra-widefield fluorescein angiography (UWF FA) in patients with diabetic retinopathy (DR) differ significantly?

Subjects/methods: Inclusion criteria were patients who had a UWF FA taken for DR at the Kellogg Eye Center from January 2009 to May 2018. Exclusion criteria included previous panretinal photocoagulation and significant media opacity (e.g., vitreous haemorrhage or significant cataract). UWF FAs were manually segmented for surface areas of NP and NV. The total areas per patient were organized in a histogram, and logarithmically binned to test against power law and exponential distributions. Then, a computational model was constructed in Python 3.7 to suggest a mechanistic explanation for the observed distributions.

Results: Analysis of areas of NV across 189 images demonstrated a superior fit by the least squares method to a power law distribution (p = 0.014) with an R² fit of 0.9672. Areas of NP over 794 images demonstrated a superior fit with an exponential distribution instead (p = 0.011). When the far periphery was excluded, the R² fit for the exponential distribution was 0.9618. A computational model following the principles of self-organized criticality (SOC), akin to earthquake and forest fire models, matched these datasets.

Conclusions: These distributions inform what useful statistics may be applied to study of these imaging characteristics. Further, the difference in event distribution between NV and NP suggests that the two phenomena are mechanistically distinct. NV may follow SOC, propagating as a catastrophic event in an unpredictable manner.