New deep learning network may improve diagnosis of diabetic retinopathy
AVA-Net, a deep learning network, differentiated the changes in blood flow in healthy eyes compared with eyes of patients with diabetes with and without mild diabetic retinopathy.
VA-Net,a deep learning network, differentiated the changes in blood flow in healthy eyes compared with eyes of patients with diabetes with and without mild diabetic retinopathy. (Image Credit: Adobe Stock/pdusit)
A new computer program, AVA-Net,a deep learning network developed by scientists from the University of Illinois at Chicago, differentiated the changes in blood flow in healthy eyes compared with eyes of patients with diabetes with and without mild diabetic retinopathy (DR) using optical coherence tomography angiography (OCTA) images.1
Differential artery-vein analysis using OCTA images seems promising for early detection of ocular diseases. However, the currently available methods for this analysis are limited without quantitative information of vascular perfusion intensity. Mansour Abtahi, PhD, and colleagues from the University of Illinois at Chicago, reported their results indeveloping and validating a method for quantitative artery-vein analysis of the intensity of vascular perfusion.
AVA-Net was developed for automated artery-vein area (AVA) segmentation in OCTA. The investigators tested 7 new OCTA features that might help detect early diabetic retinopathy: including arterial area (AA), venous area (VA), AVA ratio (AVAR), total perfusion intensity density (T-PID), arterial PID (A-PID), venous PID (V-PID), and arterial-venous PID ratio (AV-PIDR). Each was evaluated for quantitative evaluation of OCTA images from healthy controls, diabetic patients without DR, and mild DR.
The investigators found that a few of the features tested, AA, VA, and AVAr, were able to show significant differences between the control subjects and the patients with mild DR.
T-PID and A-PID, the vascular perfusion parameters, were able to differentiate mild DR from control group.
AV-PIDR showed significant differences among the controls, the patients with diabetes but no DR,and the patients with mild DR. Analysis showed that using the combination of A-PID and AV-PIDR revealed significant differences in all 3 groups.
Dr. Abtahi and colleagues commented, “AVA-Net, which is available on GitHub for open access, enables quantitative AV analysis of the AV area and vascular perfusion intensity. Comparative analysis revealed AV-PIDR as the most sensitive feature for OCTA detection of early DR. Ensemble AV feature analysis, e.g., the combination of A-PID and AV-PIDR, can further improve the performance for early DR assessment.”
