Digital image processing software for diagnosing diabetic retinopathy from fundus photograph

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss in diabetic patients.
• Early detection and screening are crucial for preventing severe vision impairment.

Purpose of the Study:

• To develop automated software for screening and diagnosing diabetic retinopathy (DR) using fundus images.
• To evaluate the software's accuracy in detecting DR and classifying its severity.

Main Methods:

• Developed automated software using MATLAB R2015a with Image Processing and GUI Toolboxes.
• Extracted clinically significant features for DR detection and severity classification.
• Validated software accuracy against ophthalmologist diagnoses on 400 fundus images.

Main Results:

• The software achieved 98% sensitivity, 67% specificity, and 96.25% accuracy in DR detection.
• Classification accuracy for non-proliferative vs. proliferative DR was 66.58%.
• Average processing time per image was 7 seconds.

Conclusions:

• Automated DR screening software developed using MATLAB demonstrates high accuracy (96.25%) for DR detection.
• The software can serve as a valuable tool for DR screening in underserved rural areas.
• Further improvements in classification accuracy are warranted.