Detecting Retinal Damage From OCT Images

Description

This study explores the application of convolutional neural networks (CNNs) and transfer learning to classify retinal OCT images for diagnosing retinal disorders. We compare a custom CNN model with two pretrained models, MobileNet and VGG16, to assess their effectiveness in classification. The project applies image enhancement techniques and evaluates model performance using confusion matrices.

Our results show that the VGG16-based CNN model outperforms others, offering superior classification accuracy for retinal diseases. By incorporating techniques like data augmentation, fine-tuning, and hyperparameter optimization, the model's performance can be further improved, reducing overfitting and enhancing diagnostic capabilities.

Conclusion

The findings suggest that CNNs, particularly the VGG16-based model, have significant potential in automating the diagnosis of retinal diseases using OCT images. These models can improve diagnostic accuracy and provide reliable solutions for early detection in ophthalmology.

Through strategies like data augmentation and hyperparameter tuning, the performance of these models can be further enhanced, making them more robust and generalizable. This work underscores the importance of applying deep learning techniques to medical imaging, offering promising prospects for better patient outcomes in the future.