Comments on Xie et al.’s Study on Artificial Intelligence–Assisted Perfusion Density as Biomarker for Screening Diabetic Nephropathy

Artificial Intelligence-Assisted Perfusion Density as a Biomarker for Screening Diabetic Nephropathy. 1 The authors present a compelling case for the use of random forest classification of perfusion density from ultra-widefield swept-source optical coherence tomography angiography (SS-OCTA) as a screening tool for diabetic nephropathy (DN).The study uses a random forest model, achieving 85.8% accuracy in the type 2 diabetes mellitus (T2DM) population and 82.5% in the diabetic retinopathy (DR) population.Although these results are promising, we noted that perfusion density (PD) is already strongly correlated with DN, as demonstrated in table 3.This raises the question of whether a complex artificial intelligence (AI)-based model is necessary when simpler models can perform just as well.To investigate this, we conducted a simulationbased analysis using the published summary statistics from Xie et al.We simulated a dataset preserving the reported means and standard deviations of PD across the control, DR without DN, and DR with DN groups.Kernel density estimation was used to generate synthetic data without assuming a specific distribution, providing a flexible approximation of the underlying data structure while accounting for potential deviations from normality.