Deep Neural Network-Based Method for Detecting Central Retinal Vein Occlusion Using Ultrawide-Field Fundus...

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare
• Machine Learning for Disease Detection

Background:

• Central Retinal Vein Occlusion (CRVO) is a significant cause of vision loss.
• Accurate and timely diagnosis of CRVO is crucial for effective treatment.
• Ultrawide-field fundus imaging provides comprehensive retinal views but requires efficient analysis tools.

Purpose of the Study:

• To evaluate the performance of deep learning (DL) and support vector machine (SVM) algorithms for CRVO detection.
• To compare the diagnostic accuracy of DL and SVM using ultrawide-field fundus images.
• To assess the feasibility of automated CRVO detection in ophthalmoscopy.

Main Methods:

• Development of a DL model using deep convolutional neural networks trained on ultrawide-field fundus images.
• Implementation of an SVM model with a radial basis function kernel.
• Comparative analysis of DL and SVM performance based on sensitivity, specificity, and AUC for CRVO diagnosis.

Main Results:

• The DL model achieved a sensitivity of 98.4%, specificity of 97.9%, and AUC of 0.989.
• The SVM model achieved a sensitivity of 84.0%, specificity of 87.5%, and AUC of 0.895.
• The DL model demonstrated significantly superior performance across all diagnostic indices compared to the SVM (P < 0.001).

Conclusions:

• Deep learning models can accurately distinguish between normal and CRVO images using ultrawide-field fundus photography.
• Automated CRVO detection via DL in ultrawide-field fundus ophthalmoscopy is feasible and highly accurate.
• This DL-based approach can enhance CRVO diagnosis, particularly in remote areas lacking specialized ophthalmic care.