AI-Driven Macular Hole Detection in OCT Scans: Efficient Training, Robust Performance, and Global Significance

Abstract

Purpose : Macular Hole (MH), a vision-threatening pathology affecting the macula, necessitates prompt and accurate diagnosis for effective intervention. With an estimated global prevalence affecting approximately 7 in 100,000 individuals, the significance of precise and timely diagnostic tools is evident. This study focuses on developing an artificial intelligence (AI) model to distinguish MH from normal cases within Optical Coherence Tomography (OCT) scans. The primary aim is to create a precise and efficient tool that aids ophthalmologists in swiftly identifying MH, allowing for timely interventions and improved clinical diagnostics and patient care.

Methods : Utilizing a publicly available image dataset from Kaggle.com, comprising 1,000 images, including 500 MH OCT scans and 500 normal OCT scans, our AI model underwent rigorous training. The dataset was partitioned randomly into three sets: training (80%), validation (10%), and testing (10%). Leveraging Google's Collaboration platform, the model was trained efficiently in a mere 1 hour and 48 minutes, ensuring cost-free processing and maintaining a carbon-neutral footprint.

Results : The AI model showcased exceptional performance metrics, boasting an accuracy, precision, recall (sensitivity), specificity, and F1-score, all surpassing 99%. Additionally, the model demonstrated an outstanding Area Under the Curve (AUC) value of 1.0, signifying its robust discriminatory power in identifying MH within OCT scans. These results significantly outperformed existing benchmarks and hold promise for practical application in clinical settings.

Conclusions : With MH impacting individuals globally, this study highlights the effectiveness of an AI-driven approach in accurately identifying MH within OCT scans. The model's exceptional accuracy, precision, and impressive AUC underscore its potential as a valuable diagnostic tool for ophthalmologists, enabling rapid and accurate detection of macular pathologies. Leveraging this technology could profoundly impact early intervention strategies and proactive management of retinal pathologies, ultimately leading to improved patient outcomes in clinical practice.

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