Diabetic Retinopathy Prediction Based on Wavelet Decomposition and Modified Capsule Network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Diabetic retinopathy (DR) is a leading cause of diabetes-related vision loss due to retinal blood vessel damage.
• Early detection and staging of DR are critical for effective patient management and sight preservation.
• Current diagnostic methods for DR are time-consuming and require expert interpretation.

Purpose of the Study:

• To develop an automated method for detecting diabetic retinopathy severity levels.
• To improve the accuracy and efficiency of DR diagnosis through advanced computational techniques.

Main Methods:

• A novel deep hybrid model combining pyramid hierarchy of discrete wavelet transform, modified capsule networks, and modified inception blocks was developed.
• The model integrates inception blocks with capsule networks for enhanced feature extraction from retinal fundus images.
• The approach was validated using the APTOS dataset.

Main Results:

• The proposed method achieved a high training accuracy of 97.71%.
• A testing accuracy score of 86.54% was obtained, demonstrating strong performance on unseen data.
• These results are competitive with state-of-the-art methods on the same dataset.

Conclusions:

• The developed deep hybrid model offers a promising automated solution for diabetic retinopathy severity detection.
• This approach has the potential to aid ophthalmologists and improve early diagnosis and patient monitoring.
• Further research can explore clinical integration and validation of this advanced diagnostic tool.