An Intelligent Model for Blood Vessel Segmentation in Diagnosing DR Using CNN

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a diabetes complication causing vision loss.
• Early DR detection is crucial to prevent blindness.
• Ophthalmologists diagnose DR through dilated eye exams observing retinal deviations.

Purpose of the Study:

• To develop an automated system for DR detection and severity classification from fundus images.
• To enhance the accuracy of DR diagnosis through advanced image processing and machine learning techniques.

Main Methods:

• Retinal image segmentation and blood vessel extraction using morphological processing, thresholding, edge detection, and adaptive histogram equalization.
• Development of a convolutional neural network (CNN) architecture for automatic DR severity classification.
• Training the CNN on diverse datasets (DRIVE, DIARETDB0, DIARETDB1_v1) and hospital-collected images using a GPU.

Main Results:

• The proposed CNN achieved high performance metrics.
• Sensitivity: 98%
• Specificity: 93%
• Accuracy: 96.9% on a database of 854 images.

Conclusions:

• The developed CNN model demonstrates significant potential for accurate and automated diabetic retinopathy diagnosis.
• This approach can aid in early detection, improving patient outcomes and preventing vision loss.