Classification of Diabetic Retinopathy Disease Levels by Extracting Spectral Features Using Wavelet CNN.

Area of Science:

• Ophthalmology and Medical Imaging

Background:

• Diabetic retinopathy (DR) is a major cause of blindness due to blood vessel damage.
• Current diagnosis relies on manual fundus image analysis, which is error-prone and time-consuming.
• Computer-assisted methods, particularly Convolutional Neural Networks (CNNs), show promise for automated DR detection.

Purpose of the Study:

• To explore the utility of spectral features, beyond spatial features, for more accurate DR severity grading.
• To introduce and evaluate a novel model combining Wavelet CNN and Support Vector Machine (SVM) for DR classification.

Main Methods:

• The study utilized the EyePACS dataset for experiments.
• A hybrid model integrating Wavelet CNN for feature extraction and SVM for classification was developed.
• Performance was assessed using precision, recall, F1-score, accuracy, and AUC score.

Main Results:

• The proposed Wavelet CNN-SVM model demonstrated enhanced performance in classifying clinically significant DR grades.
• The model achieved superior results compared to existing state-of-the-art techniques on the EyePACS dataset.
• Incorporating spectral features contributed to improved diagnostic accuracy.

Conclusions:

• The Wavelet CNN-SVM model offers a promising approach for automated and accurate diabetic retinopathy grading.
• Leveraging spectral features alongside spatial information can significantly enhance DR detection systems.
• This method has the potential to assist clinicians, reduce diagnostic errors, and improve patient outcomes in managing DR.