Segmentation technique and dynamic ensemble selection to enhance glaucoma severity detection.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Glaucoma severity classification is crucial for effective treatment.
• Traditional methods rely on parameters like cup-to-disc ratio (CDR) and peripapillary atrophy (PPA).
• Accurate segmentation and classification are essential for reliable glaucoma diagnosis.

Purpose of the Study:

• To develop and evaluate an advanced glaucoma classification model.
• To improve the accuracy of glaucoma severity detection using combined ocular parameters.
• To leverage dynamic ensemble selection (DES) for robust classification, especially with imbalanced datasets.

Main Methods:

• Active contour snake (ACS) for optic disc (OD) and optic cup (OC) segmentation to calculate CDR.
• Otsu segmentation and thresholding for peripapillary atrophy (PPA) identification.
• Grey-level co-occurrence matrix (GLCM) for feature extraction and dynamic ensemble selection (DES) for classification.

Main Results:

• The proposed model achieved a mean accuracy of 0.96.
• The DES-based model outperformed the deep learning U-Net model (0.90 accuracy).
• Evaluation was conducted on three datasets comprising 250 retinal fundus images using 5-fold cross-validation.

Conclusions:

• The proposed glaucoma classification model demonstrates superior performance compared to U-Net.
• The integration of ACS, GLCM, and DES offers a robust approach for glaucoma severity assessment.
• This method holds promise for accurate glaucoma detection and guiding subsequent treatment decisions.