Novel fractal feature-based multiclass glaucoma detection and progression prediction.

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• Glaucoma is a leading cause of irreversible blindness worldwide.
• Accurate prediction of glaucoma progression is crucial for timely intervention.
• Current diagnostic methods may lack the sensitivity to detect early or subtle changes.

Purpose of the Study:

• To develop and evaluate a novel system for multiclass prediction of glaucoma progression using fractal analysis (FA).
• To compare the performance of FA with wavelet-Fourier analysis (WFA) and fast-Fourier analysis (FFA).
• To assess the capability of FA in classifying normal, non-progressive, and progressive glaucoma.

Main Methods:

• Retinal nerve fiber layer (RNFL) data from normal and glaucoma subjects were converted to pseudo 2D images.
• Fractal analysis features were extracted using box-counting and multi-fractional Brownian motion methods.
• Gaussian kernel-based multiclass classification was performed using FA, WFA, and FFA features.

Main Results:

• FA achieved a higher area under the receiver operating characteristic curve (AUROC) of 0.82 for predicting progressors versus non-progressors, compared to WFA (0.70) and FFA (0.71).
• Multiclass classification rates for FA were 0.88 (normal), 0.86 (non-progressive), and 0.82 (progressive glaucoma).
• FA demonstrated superior performance with fewer features and reduced computational complexity.

Conclusions:

• Fractal analysis provides a robust and efficient method for multiclass prediction of glaucoma progression.
• FA outperforms WFA and FFA in classifying glaucoma severity and progression.
• This novel FA-based system holds promise for improving glaucoma diagnosis and management.