Deep tessellated retinal image detection using Convolutional Neural Networks.

Area of Science:

• Ophthalmology
• Medical Image Analysis
• Computer Vision

Background:

• Tessellation in fundus images is a significant feature associated with age-related and myopic maculopathy.
• The presence of tessellation can interfere with accurate retinal vessel segmentation, a critical step in retinal image analysis.
• Automated detection of tessellated images is essential for robust retinal image processing.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) model for the automated detection of tessellated fundus images.
• To assess the model's performance in classifying retinal images as either tessellated or non-tessellated.

Main Methods:

• A pre-processed fundus image is used as input to the proposed CNN model.
• The model utilizes a transfer learning technique with a pre-trained GoogLeNet architecture.
• A comprehensive database of 12,000 color retinal images was curated for model evaluation.

Main Results:

• The developed CNN model demonstrated high classification performance in detecting tessellated images.
• The best performing tessellation classifier achieved an accuracy of 97.73%.
• An Area Under the Curve (AUC) value of 0.9659 was obtained, indicating excellent discrimination.

Conclusions:

• The proposed CNN model is effective for the automated detection of tessellated fundus images.
• This approach can significantly aid in the analysis of conditions like age-related and myopic maculopathy.
• The findings highlight the utility of deep learning for improving retinal image analysis accuracy.