Attention-based deep learning framework to recognize diabetes disease from cellular retinal images.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a significant cause of visual impairment in diabetic patients.
• High blood sugar levels damage retinal blood vessels, leading to DR.
• Early detection and intervention are crucial to prevent vision loss.

Purpose of the Study:

• To develop and evaluate an attention-based hybrid deep learning model for early detection of diabetic retinopathy.
• To improve the accuracy and robustness of automated DR detection systems.
• To differentiate between various stages of diabetic retinopathy.

Main Methods:

• Utilized DenseNet121 architecture for convolutional feature extraction.
• Enhanced feature vectors using a channel and spatial attention model.
• Implemented both binary (DR present/absent) and multiclass (DR severity scale 0-4) classification.
• Incorporated data augmentation techniques to enhance model generalization.

Main Results:

• The proposed model achieved 98.57% accuracy for multiclass classification.
• The model achieved 99.01% accuracy for binary classification.
• Data augmentation demonstrated a positive impact on model robustness and generalization.

Conclusions:

• The attention-based deep learning model shows high accuracy in detecting diabetic retinopathy from retinal images.
• This automated approach holds promise for early diagnosis and management of DR.
• The model's performance suggests its potential for clinical application in screening diabetic patients.