Prediction of late stage AMD based on deep learning models using multimodal AREDS data

Abstract

Purpose :
Purposes: Age-related macular degeneration (AMD) is a principal cause of blindness influenced by both genetic and environmental factors. AMD severity is primarily measured through color fundus photographs. Recently, machine learning techniques utilizing these images have emerged for AMD prediction. In our previous research, using both genetic and image data, we successfully predicted AMD progression. In this study, we expanded our methodology by incorporating more predictors, such as age, smoking status, education, medical history (including cataract surgery, laser photocoagulation, visual acuity, diabetes, and hypertension), and AMD treatment.

Methods : Methods: We employed a modified deep convolutional neural network (CNN) to predict the progression to late AMD using 31,262 fundus images, 52 AMD-related genetic markers[CE([1] e, and 9 demographic/clinical variables. This data was sourced from 1,351 subjects [CE([2] in the Age-Related Eye Disease Study (AREDS) observed over 12 years.

Results : Results: Our findings indicated that the combination of fundus images with demographic/clinical and genetic data predicted late AMD progression with an average area under the curve (AUC) of 0.9 (95%CI: 0.88-0.91). In comparison, integrating only fundus images with genotypes resulted in an AUC of 0.85 (95%CI: 0.83-0.86), and using solely fundus images yielded an AUC of 0.81 (95%CI: 0.80-0.83).

Conclusions : Conclusion: Inclusion of clinical data in deep learning models significantly enhances late stage AMD prediction accuracy.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.