Development of a Fundus Image-Based Deep Learning Diagnostic Tool for Various Retinal Diseases

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence (AI) is increasingly utilized in ophthalmology.
• Retinal images, like fundus photographs, are key for developing AI diagnostic tools.
• Numerous retinal pathologies necessitate accurate and timely diagnosis.

Purpose of the Study:

• To develop and evaluate an AI model for the differential diagnosis of retinal diseases using fundus images.
• To compare the performance of different convolutional neural network (CNN) architectures and dense layers for this diagnostic task.

Main Methods:

• Three CNN models (ResNet50, VGG19, Inception v3) were employed to classify retinal images.
• The study compared the performance of various dense (fully connected) layers.
• The AI model was trained to differentiate between eight retinal diseases and normal controls.

Main Results:

• The ResNet50 model with a dense layer of 128 nodes achieved a prediction accuracy of 87.42% for diagnosing nine classes.
• The developed AI tool demonstrated potential to augment ophthalmologists' diagnostic capabilities.
• The fundus image-based AI model proved applicable in the medical diagnosis workflow for retinal diseases.

Conclusions:

• Fundus image-based AI models are effective for the differential diagnosis of retinal diseases.
• AI tools can support and enhance the diagnostic performance of ophthalmologists.
• This AI approach shows promise for improving the efficiency and accuracy of retinal disease diagnosis.