Abstract
Purpose :
This study introduces a multi-class model proficient in discerning various pathologies from OCT scans such as: Diabetic Macular Edema (DME), Choroidal Neovascularization (CNV), and Drusen. This investigation aims to assist ophthalmologists in accurately diagnosing common retinal conditions like DME, CNV, and Drusen, which require timely identification for effective treatment due to their distinct features and clinical importance. This research aims to demonstrate that multi-class models, an alternative to current single-class models, have the potential to be utilized in conjunction with OCT scans to aid ophthalmologists in improving diagnostic accuracy and enhancing patient care in ophthalmology.
Methods :
The AI (Artificial Intelligence) model was trained utilizing Google's Collaboration platform, this process took 1 hour and 58 minutes. Leveraging Google's servers ensured a cost-free and carbon-neutral training process. The model was trained on a diverse dataset of 1860 OCT images with either CNV (407 images), DME (485 images), drusen (468 images), or normal (500 images) pathologies. Publicly available images from Kaggle.com were used to train the model.
Results :
The model demonstrated exceptional performance metrics, achieving an Area Under the Curve (AUC) value of .9759, an accuracy of 98.72%, a precision of 98.04%, a recall (sensitivity) of 99.78%, a specificity of 98.02%, and an F1-score of 98.91%. These results indicate the model's efficacy in accurately classifying Drusen, CNV, DME, and normal pathologies within OCT scans with high precision and sensitivity.
Conclusions :
This investigation demonstrates the potential of multi-classification models to be used as a valuable tool for ophthalmologists in clinical practice. By utilizing Google's Collaboration platform for sustainable and cost-effective model training, this research underscores the importance of adopting environmentally friendly approaches in healthcare. The high accuracy, precision, AUC, and recall achieved by the model suggest its utility as a reliable diagnostic aid, facilitating timely and accurate diagnoses of diverse retinal pathologies. Continued advancements in AI-driven healthcare solutions hold promise for the technology to be employed for multiple different use-cases simultaneously.
This abstract was presented at the 2024 ARVO Imaging in the Eye Conference, held in Seattle, WA, May 4, 2024.