AI-Enhanced CNV Diagnosis in OCT Scans: A Sustainable and Efficient Approach

Abstract

Purpose : The main objective of this study is to address the need for identifying Choroidal Neovascularization (CNV) in Optical Coherence Tomography (OCT) scans, which presents a significant challenge in the field of ophthalmology. CNV refers to the growth of blood vessels beneath the retina, this can have a severe impact on ocular health. Ophthalmologists often face difficulties and time constraints when manually diagnosing CNV. Therefore this study introduces an intelligence (AI) model that has been trained on a diverse dataset to diagnose CNV. Notably Google's Collaboration platform was utilized for developing a cost environmentally friendly model.

Methods : The AI model underwent training using a dataset consisting of 2392 images, including 1194 OCT scans with CNV and 1198 normal OCT scans. To ensure reliability, the data was randomly divided into three sets; training (80%) validation (10%) and testing (10%). Leveraging an available image dataset from Kaggle.com, the model efficiently trained on Google's servers for just 1 hour and 48 minutes.

Results : The AI model exhibited capability in distinguishing between OCT scans with and without CNV across a diverse dataset. The strategic distribution of 2392 images, among the training, validation and testing sets contributes to the reliability of this model. The evaluation metrics used to measure the models performance, such as accuracy (98.5%) precision (97.09%) recall (100%) specificity (97%) F1 score (98.63%) and an Area Under the Curve (AUC) of 0.993 demonstrate that the model is proficient.

Conclusions : This research contributes to the field of eye health by introducing an AI model that can accurately identify CNV in OCT scans. The use of Google's Collaboration platform enabled a time environmentally conscious development process for the model. The careful distribution of datasets and transparency in details enhance the models adaptability and reproducibility. With performance metrics this model shows potential to revolutionize the diagnosis and management of CNV, offering benefits such as early detection, improved efficiency, and resource optimization in ophthalmic care.

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

 

Figure 1. Confusion Matrix for Model Performance

 

Figure 2. Precision-recall (95% Confidence)