A Robust Machine Learning Model for Diabetic Retinopathy Classification.

Area of Science:

• Artificial Intelligence (AI)
• Machine Learning (ML)
• Image Processing

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss, necessitating effective diagnostic tools.
• Analyzing fundus eye images is crucial for early DR detection and management.
• Current diagnostic methods can be enhanced by automated systems leveraging AI.

Purpose of the Study:

• To develop and validate an AI-based model for classifying diabetic retinopathy severity using fundus images.
• To explore the efficacy of image processing techniques, including contrast manipulation and entropy analysis, for feature extraction.
• To identify the optimal machine learning classifier for DR diagnosis.

Main Methods:

• Fundus images were preprocessed using gamma correction for contrast manipulation.
• Texture analysis was performed using Shannon and fuzzy entropies to extract ten novel features.
• Ensemble learning with the PyCaret library was employed, evaluating fifteen classifiers.
• Gradient Boosting Classifier (GBC) was selected as the best-performing model.

Main Results:

• The GBC model achieved high performance in classifying DR as normal or severe.
• Achieved accuracy of 0.929, F1 score of 0.902, and Area Under the Curve (AUC) of 0.941.
• Model validation was confirmed using a bootstrap statistical technique.

Conclusions:

• The proposed method effectively extracts features from preprocessed fundus images for DR classification.
• Controlled contrast manipulation and entropy-based texture analysis are valuable for DR diagnosis.
• The AI-driven approach offers a promising tool for automated DR screening and diagnosis.