A multidomain bio-inspired feature extraction and selection model for diabetic retinopathy severity classification:...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of blindness worldwide.
• Early detection and severity estimation are vital for preventing vision loss in diabetes mellitus patients.
• Current methods require improvement for accurate and timely DR diagnosis.

Purpose of the Study:

• To propose an augmented bioinspired multidomain feature extraction and selection model for diabetic retinopathy severity estimation.
• To develop an ensemble learning process for enhanced DR classification.
• To improve the accuracy and efficiency of DR severity assessment.

Main Methods:

• Segmentation of key retinal features (optic disc, macula, blood vessels, exudates, hemorrhages) using adaptive thresholding.
• Extraction of multidomain features including frequency, entropy, cosine, Gabor, and wavelet components.
• Optimal feature selection using a Modified Moth Flame Optimization algorithm.
• Ensemble learning with algorithms like Naive Bayes, KNN, SVM, MLP, Random Forests, and Logistic Regression for DR severity identification.

Main Results:

• The proposed model successfully segmented retinal features and extracted multidomain components.
• The Modified Moth Flame Optimization algorithm effectively selected optimal features.
• The ensemble learning model achieved a high accuracy of 96.5% in identifying DR severity levels.
• The approach outperformed conventional methods in experiments on public datasets.

Conclusions:

• The developed augmented bioinspired model demonstrates superior performance in diabetic retinopathy severity estimation.
• The combination of advanced feature extraction, selection, and ensemble learning offers a promising tool for early DR detection.
• This method has the potential to significantly aid in preventing blindness caused by diabetic retinopathy.