Automatic arteriosclerotic retinopathy grading using four-channel with image merging

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Research

Background:

• Arteriosclerosis retinopathy is a serious complication of hypertension, posing a significant threat to human health.
• Current manual assessment of arteriosclerosis retinopathy is labor-intensive, time-consuming, and costly.
• There is a critical need for automated methods for accurate and efficient grading of arteriosclerosis retinopathy.

Purpose of the Study:

• To develop a novel automated method for grading arteriosclerosis retinopathy using convolutional neural networks (CNNs).
• To enhance feature extraction for fundus blood vessel analysis through image merging and contour enhancement.
• To improve the accuracy and efficiency of arteriosclerosis retinopathy detection and classification.

Main Methods:

• A novel feature extraction scheme involving image merging for contour enhancement was developed.
• Adaptive threshold processing was used to generate a new contour channel, merged with the original fundus image.
• A pre-trained CNN with transfer learning and Kaiming initialization was employed, incorporating ArcLoss for improved classification.

Main Results:

• The proposed method achieved an accuracy of 65.354% for arteriosclerosis retinopathy grading.
• This accuracy represents an improvement of nearly 4% compared to existing methods.
• The method obtained a Kappa score of 0.508, indicating good agreement in grading.

Conclusions:

• The developed CNN-based method demonstrates superior performance in arteriosclerosis retinopathy grading.
• The approach offers a valuable and efficient tool for automated detection and grading of the condition.
• This automated system has the potential to aid clinicians in managing hypertensive retinopathy.