Investigating Rabbit Persistent Retinal Neovascularization (PRNV) Model Leveraging Artificial Intelligence (AI)-assisted OCT & FA Assessment

Abstract

Purpose : The merangiotic nature of rabbit retina made it difficult to quantitatively compare different eyes. We developed an imaging pipeline for optical coherence tomography (OCT) and fluorescein angiography (FA) to readout retina changes. The method was applied to the PRNV model created using intravitreal injection (IVT) of DL-alpha-aminoadipic acid (DLAAA).

Methods : 27 Dutch belt rabbits received 50µl 80mM IVT DLAAA were imaged using FA and OCT up to 13 weeks before and 9 weeks post treatment (Eylea®, 125ug/eye). FA images were taken using Spectralis up to 10 minutes. Vascular and leakage areas were recognized by in-house deep learning-based AI algorithm. OCT images were obtained using Envisu SD-OCT for volume scans (9*9mm, 2000B*500A-scans) temporal and inferior to optic nerve head (ONH), the location of retinal degeneration. B-scans were segmented into 3 layers by AI: vascular layer, retinal never fiber layer(RNFL), retina layer without RNFL. A specifically designed grid for merangiotic retina automatically defines ONH, vascular area and visual streak to calculates layer thickness.

Results : AI-based FA image segmentation robustly detects PRNV areas in FA images. En face images based on segmentation of OCT vascular layer provides OCT-A like images to access PRNV; en face of RNFL layer produces allows assessment of ganglion cell axons loss after DLAAA. 20 of 27 eyes were selected to enrolled in treatment study: excluded 5 with retinal damage but no or minor PRNV, 2 severe retinal detachment that affects vascular area. Eylea® was able to suppress PRNV for 7 weeks then leakages resumed. There was no statistical significance of retina layer thickness before and after Eylea® treatment.

Conclusions : Quantification of retina vascular and layer structure change enabled by AI provides accurate, robust, and repeatable measurements. En face images reconstructed from OCT layer segmentation provides additional insight of regional retinal structure change in vivo, replacing the needs of OCTA and post-mortem flat mount.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Fig1. Predicted vascular and leakage area from FA image

 

Fig 2. Structure change of retina before and after DLAAA IVT in OCT reconstruction. (A) Reconstructred visual streak OCT (B) RNFL layer en face image showing ganglion cell axons loss (orange line area) (C) Vascular layer en face images showing neovascularization (blue arrowhead).