OculusNet: Detection of retinal diseases using a tailored web-deployed neural network and saliency maps for...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinal diseases are a major cause of global blindness.
• Manual interpretation of optical coherence tomography (OCT) images is labor-intensive and requires specialized expertise.
• Early detection of retinal disorders is crucial for effective treatment.

Purpose of the Study:

• To introduce OculusNet, an efficient and explainable deep learning (DL) model for detecting retinal diseases using OCT images.
• To improve diagnostic accuracy and efficiency in ophthalmology.
• To provide an accessible tool for retinal disease detection.

Main Methods:

• Developed OculusNet, a deep learning model tailored for complex OCT image patterns.
• Utilized Saliency Map visualization (Explainable AI - XAI) for model interpretability.
• Compared OculusNet against pre-trained models (VGG19, MobileNetV2, VGG16, DenseNet-121) using transfer learning.
• Deployed the model on a web interface for user accessibility.

Main Results:

• OculusNet achieved a test accuracy of 95.48% and a validation accuracy of 98.59%.
• The model outperformed all compared pre-trained models.
• Matthews Correlation Coefficient and Cohen's Kappa Coefficient validated the model's reliability and generalizability for clinical use.

Conclusions:

• OculusNet demonstrates high accuracy and efficiency in detecting retinal diseases from OCT images.
• The explainable AI component enhances trust and understanding of the diagnostic process.
• The web-based deployment offers potential for widespread integration into clinical settings, improving patient care.