Cross-Instrument Optical Coherence Tomography-Angiography (OCTA)-Based Prediction of Age-Related Macular Degeneration (AMD) Disease Activity Using Artificial Intelligence

Abstract

Purpose : This study explores the effectiveness of predicting AMD disease activity through deep neural networks (DNN) using a cross-instrument training dataset derived from Heidelberg Spectralis and Optovue OCTA images.

Methods : A retrospective cross-sectional study was conducted, utilizing 2D vascular en-face OCTA images from Heidelberg Spectralis (697 samples: 477 for training, 120 for validation, 100 for testing) and Optovue (580 samples: 416 training, 104 validation, 60 for testing). The OCTA scans were categorized based on clinical diagnosis and information on the fluid in adjacent B-scan OCT into 4 main categories: normal, dry AMD, active wet AMD, and wet AMD in remission with no signs of activity. Various experiments were conducted to investigate cross-instrument OCTA disease classification prediction, utilizing Heidelberg OCTA data, Optovue data separately, and the combined dataset for training the DNN. Testing was performed on Heidelberg (100 samples) and Optovue (60 samples) test sets.

Results : Training the OCTA data solely on Heidelberg yielded a 72% overall prediction accuracy for the disease stage on the Heidelberg test set and 62% accuracy on the Optovue test set. Training on Optovue data only resulted in 32% accuracy on Heidelberg and 76% accuracy on Optovue test sets. Combining training data from both instruments (Heidelberg+Optovue) achieved 76% overall accuracy on Heidelberg and 75% on Optovue test sets for the overall disease stage prediction.

Conclusions : AI can utilize 2D en-face vascular OCTA scans to predict the disease stage of AMD. Cross-instrument classifier training demonstrated high classification prediction accuracy for both Heidelberg and Optovue OCTA data.

This abstract was presented at the 2024 ARVO Imaging in the Eye Conference, held in Seattle, WA, May 4, 2024.