Automatic Diagnosis of Glaucoma from Retinal Images Using Deep Learning Approach

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma, characterized by elevated intraocular pressure and optic nerve damage, can lead to irreversible blindness.
• Early detection is vital, yet glaucoma is often diagnosed late in the elderly, necessitating improved diagnostic methods.
• Current manual glaucoma assessment is time-consuming, costly, and requires specialized skills, with experimental techniques yet to offer a definitive solution.

Purpose of the Study:

• To develop and validate an automated deep learning-based method for accurate early-stage glaucoma detection.
• To identify subtle patterns in retinal images that may be overlooked by human clinicians.

Main Methods:

• Utilized gray channels of fundus images for analysis.
• Employed data augmentation to create a diverse dataset for training.
• Trained a Convolutional Neural Network (CNN) model using the ResNet-50 architecture.

Main Results:

• Achieved high accuracy in detecting early-stage glaucoma across multiple datasets (G1020, RIM-ONE, ORIGA, DRISHTI-GS).
• On the G1020 dataset, the model demonstrated 98.48% accuracy, 99.30% sensitivity, 96.52% specificity, 97% AUC, and 98% F1-score.
• The deep learning approach successfully identified patterns often missed in manual assessments.

Conclusions:

• The proposed deep learning model offers a highly accurate and automated solution for early glaucoma detection.
• This method has the potential to significantly aid clinicians in timely diagnosis and intervention, preventing vision loss.