Revolutionizing AMD detection Bi model CNNs and hybrid feature selection for automated grading

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Age-related macular degeneration (AMD) is a leading cause of vision loss in elderly populations.
• Automated grading of AMD from fundus images is crucial for early detection and timely intervention.

Purpose of the Study:

• To develop a comprehensive framework for enhancing fundus image quality and improving automated AMD grading accuracy.
• To introduce a novel Bi-Model Convolutional Neural Network (CNN) architecture for precise AMD grading.

Main Methods:

• Image preprocessing using Contrast Limited Adaptive Histogram Equalization (CLAHE) and Gamma correction.
• Hybrid feature selection combining handcrafted and deep learning features.
• A Bi-Model CNN architecture integrating global and local features from fundus images and patches.

Main Results:

• The proposed Bi-CNN + Feature Fusion model achieved an accuracy of 99.5%.
• The model demonstrated high performance with precision (0.995), recall (0.995), and F1-score (0.995).
• A Cohen's Kappa of 0.990 indicated near-perfect agreement between predicted and actual AMD grades.

Conclusions:

• The developed framework significantly enhances fundus image quality and automated AMD grading accuracy.
• The Bi-Model CNN architecture effectively utilizes global and local features for precise AMD detection.
• This approach promises more effective early identification of AMD, improving patient outcomes.