Retina images classification based on 2D empirical mode decomposition and multifractal analysis

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Geometry

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss, stemming from retinal microvascular damage due to hyperglycemia.
• Retinal changes in DR, including hemorrhages and edema, present complex morphologies often unsuited for traditional geometric analysis.
• Accurate classification of DR severity is crucial for timely intervention and patient management.

Purpose of the Study:

• To evaluate and classify retinal morphology in diabetic retinopathy using multifractal geometry.
• To develop a quantitative method for assessing DR severity based on complex retinal structures.
• To explore the utility of multifractal analysis in distinguishing diabetic retinas from healthy ones.

Main Methods:

• Employed two-dimensional empirical mode decomposition to extract high-frequency retinal features.
• Utilized multifractal geometry to analyze and classify the complex morphology of retinal images.
• Applied the random forest algorithm to select significant multifractal features for classification.

Main Results:

• The proposed multifractal geometry method achieved high classification accuracy for diabetic retinopathy.
• Achieved diagnostic performance metrics including 96% accuracy, 96% sensitivity, and 95% specificity.
• Demonstrated the effectiveness of multifractal spectrum features in correlating with DR severity.

Conclusions:

• Multifractal geometry offers a robust framework for analyzing and classifying the intricate morphology of diabetic retinopathy.
• This approach provides a promising, quantitative tool for objective assessment and diagnosis of DR.
• The findings support the potential of advanced geometric methods in ophthalmology for disease characterization.