Optimized glaucoma detection using HCCNN with PSO-driven hyperparameter tuning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness worldwide.
• Early detection and accurate diagnosis are crucial for effective management.
• Automated systems can aid clinicians in identifying glaucoma from fundus images.

Purpose of the Study:

• To develop an effective glaucoma detection system using a Hybrid Centric Convolutional Neural Network (HCCNN).
• To enhance classification accuracy and reduce computational complexity via Particle Swarm Optimization (PSO).
• To segment optic disc (OD) and optic cup (OC) regions for glaucoma severity assessment using a modified U-Net.

Main Methods:

• Feature extraction from fundus images using the proposed HCCNN model.
• Hyperparameter optimization (dropout rate, learning rate, dense layer neurons) with PSO for improved performance.
• Channel separation and segmentation of OD and OC regions using modified U-Net for severity analysis.

Main Results:

• The PSO-HCCNN model achieved 94% and 97% classification accuracy on the DRISHTI-GS and RIM-ONE datasets, respectively.
• Demonstrated significant improvements in accuracy, sensitivity, specificity, and AUC for early glaucoma detection.
• Segmentation performance evaluated using Dice coefficient and Jaccard index.

Conclusions:

• Integrating PSO with HCCNN optimizes parameters, leading to a robust and precise glaucoma classification model.
• Accurate OD and OC segmentation aids in determining glaucoma severity.
• The proposed method shows potential for early and accurate glaucoma diagnosis in clinical practice.