Classification of diabetic retinopathy algorithm based on a novel dual-path multi-module model.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a diabetes complication affecting retinal blood vessels.
• Progressive DR can lead to significant vision impairment and blindness.
• Early detection and intervention are critical for managing DR and preserving sight.

Purpose of the Study:

• To introduce a novel dual-path multi-module network algorithm for accurate diabetic retinopathy classification.
• To enhance early diagnosis and timely intervention for diabetic retinopathy.

Main Methods:

• Utilized color correcting and multi-scale fusion for retinal lesion feature enhancement.
• Employed a multi-path multiplexing structure with varying convolutional kernel sizes for local data optimization.
• Integrated a multi-feature fusion module to boost classification model accuracy.

Main Results:

• Achieved high classification accuracies of 98.9%, 99.3%, and 98.3% on public and hospital datasets.
• Demonstrated the algorithm's effectiveness in automatic diabetic retinopathy diagnosis.
• Validated the algorithm's practicality and advancement in the field.

Conclusions:

• The proposed algorithm shows significant promise for automatic DR diagnosis.
• This technology offers strong potential for clinical application in early DR screening and treatment.
• The findings support improved patient outcomes by facilitating timely medical intervention.