Identification of Retinal Diseases Using Light Convolutional Neural Networks and Intrinsic Mode Function Technique

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Fundus imaging is vital for early retinal disease detection.
• Automated interpretation faces challenges from image variations and noise.
• Accurate diagnosis relies on robust image analysis techniques.

Purpose of the Study:

• To develop a novel hybrid model for enhanced fundus image classification.
• To improve the accuracy and efficiency of automated retinal disease diagnosis.
• To address limitations in current automated fundus image analysis.

Main Methods:

• A hybrid model integrating Intrinsic Mode Function (IMF) filtering (from Empirical Mode Decomposition) with Light Convolutional Neural Networks (LightCNN).
• IMF filtering preprocesses fundus images to reduce noise and preserve retinal patterns.
• LightCNN performs lightweight feature extraction and classification on the refined image components.

Main Results:

• The IMF + LightCNN model achieved 99.4% accuracy, 99.1% precision, 98.87% recall, and 98.31 F1-score on DIARETDB datasets.
• Demonstrated superior performance compared to conventional Convolutional Neural Networks (CNN) and ResNet-based models.
• Indicated significant improvements in diagnostic accuracy and computational efficiency.

Conclusions:

• Integrating advanced signal processing (IMF filtering) with lightweight deep learning (LightCNN) enhances diagnostic accuracy.
• The hybrid framework offers a promising approach for reliable, real-time clinical screening of retinal diseases.
• This method improves both diagnostic performance and computational efficiency for fundus image analysis.