Diabetic retinopathy detection and grading system using deep learning approach.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss in diabetic patients.
• Early detection and screening are crucial for preventing irreversible vision impairment.
• Computer-assisted diagnosis offers a promising approach for efficient DR screening.

Purpose of the Study:

• To develop and evaluate a computer-assisted system for the segmentation and classification of diabetic retinopathy.
• To accurately segment the macular region and exudates in retinal images for DR analysis.
• To classify DR severity (mild, moderate, severe) based on the presence and extent of exudates.

Main Methods:

• Employed bit-plane morphological slicing for macular region segmentation.
• Utilized U-Net deep learning for exudate segmentation in retinal images.
• Validated methodologies on the MESSIDOR and HRF datasets, with further testing on DIARETDB1, RFMD, and FIR datasets.

Main Results:

• Achieved high segmentation accuracy for the macular region (e.g., 99.8% AccI on HRF).
• Demonstrated excellent exudate segmentation performance (e.g., 99.2% AccI on MESSIDOR).
• The proposed method offers improved segmentation accuracy by locating both internal and external boundaries.

Conclusions:

• The developed method provides accurate and efficient segmentation of macular regions and exudates.
• The computer-assisted approach enables DR classification with high precision.
• This technique shows potential for improving early DR screening and diagnosis, reducing vision loss.