Detection of Diabetic Retinopathy Using Bichannel Convolutional Neural Network

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) screening is crucial for preventing vision loss.
• Deep learning (DL) models, particularly Convolutional Neural Networks (CNNs), show promise for automated DR detection from fundus photographs.
• Existing methods using luminance entropy can improve DR detection, but further enhancements are being explored.

Purpose of the Study:

• To propose a novel method for enhancing the detection of referable diabetic retinopathy using deep learning.
• To investigate the utility of the green color component of fundus images for generating entropy images.
• To develop a bichannel CNN model that integrates multiple image features for improved DR screening.

Main Methods:

• Entropy images were computed using the green component of fundus photographs.
• Image enhancement was performed using unsharp masking (UM) as a preprocessing step.
• A bichannel CNN was designed to incorporate features from both gray-level entropy images and UM-preprocessed green component entropy images.

Main Results:

• The proposed method utilizing the green component entropy and unsharp masking shows potential for improving DR detection.
• The bichannel CNN architecture effectively integrates diverse image features, leading to enhanced performance in identifying referable DR.
• This approach offers a more robust and accurate automated screening tool for diabetic retinopathy.

Conclusions:

• The integration of green component entropy images, preprocessed with unsharp masking, significantly enhances deep learning-based diabetic retinopathy detection.
• The proposed bichannel CNN model demonstrates superior performance in identifying referable DR compared to existing methods.
• This technique represents a valuable advancement in the automated screening and diagnosis of diabetic retinopathy, potentially improving patient outcomes.