Utilizing AI to Detect Diabetic Macular Edema in OCT Scans: Performance and Dataset Insights

Abstract

Purpose :
Diabetic Macular Edema (DME), a common complication of diabetes affecting the retina, is a leading cause of vision impairment worldwide. With an estimated global prevalence affecting over 21 million individuals, the urgent need for swift and accurate diagnostics is evident. This study focuses on developing an artificial intelligence (AI) model to discern DME from non-DME cases within Optical Coherence Tomography (OCT) scans. The primary aim is to create a precise and efficient tool that aids ophthalmologists in swiftly identifying DME, allowing for timely interventions and improved clinical diagnostics and patient care.

Methods :
We trained our AI model using an available image dataset obtained from Kaggle.com. The dataset consisted of 3988 images with 991 being DME OCT scans and 2997 normal OCT scans. To ensure training we randomly divided the dataset into three sets; training (80%) validation (10%) and testing (10%). Leveraging Google’s Collaboration platform we achieved training, within a short span of 1 hour and 48 minutes. This approach also allowed us to minimize costs and maintain a carbon footprint.

Results :
Our AI model demonstrated performance metrics achieving approximately 99% accuracy, precision, recall (sensitivity) specificity and F1 score. Additionally, the model exhibited an Area Under the Curve (AUC) value of 0.997 highlighting its strong ability to accurately identify DME within OCT scans. This performance exceeds the benchmarks by a significant margin and shows great potential for use in clinical applications.

Conclusions :
This research emphasizes how effective an AI powered method can be in identifying DME (Diabetic Macular Edema) within OCT scans. The model's impressive accuracy, precision and outstanding AUC highlight its potential as a diagnostic tool for healthcare professionals. By utilizing this technology we can greatly impact detection strategies and proactive management of retinal conditions, in clinical settings leading to improved patient outcomes.

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