AI-Enhanced Drusen Detection in OCT Scans: An Efficient Approach to Assist Clinical Decision-Making and Improve Patient Outcomes

Abstract

Purpose : This study investigates the application of machine learning and deep learning algorithms to accurately identify drusen: extracellular deposits of lipids, proteins, and cellular debris found within the layers of the retina. This study focuses on developing an artificial intelligence (AI) model to differentiate drusen from non-drusen cases within Optical Coherence Tomography (OCT) scans.

Methods : We developed a model able to differentiate between healthy eye scans and those with drusen by utilizing a dataset of 4,820 images from the Kaggle online platform, including 2,329 drusen OCT scans and 2,491 normal OCT scans. To optimize model performance and ensure its generalizability, the dataset was randomly split into training (80%), validation (10%), and testing (10%) sets. Utilizing Google’s Collaboration platform, the model was trained in 1 hour and 48 minutes.

Results : The AI model demonstrated exceptional performance metrics with accuracy, precision, recall (sensitivity), specificity, and an F1-score of 0.98. The model demonstrated an Area Under the Curve (AUC) value of 0.998, illustrating its efficacy in identifying drusen within OCT scans.

Conclusions : The findings highlight the potential of machine learning algorithms in improving the efficiency and accuracy of drusen detection. Additional research and integration into clinical practice are necessary to ensure the reliability and generalizability of these models. This study illustrates the potential of artificial intelligence as a diagnostic tool to improve patient outcomes through early identification of drusen.

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