Detection of microscopic glaucoma through fundus images using deep transfer learning approach

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma is a leading cause of irreversible blindness due to optic nerve damage.
• Early detection is crucial but challenging as glaucoma often presents asymptomatically.
• Machine learning and deep learning offer promising avenues for automated glaucoma diagnosis.

Purpose of the Study:

• To develop and evaluate a novel hybrid deep learning model for early glaucoma detection.
• To assess the performance of the proposed model using fundus images from multiple datasets.
• To compare the proposed model's efficacy against existing methods in terms of accuracy and efficiency.

Main Methods:

• A combined DenseNet and DarkNet architecture was utilized for feature extraction and classification.
• The model was trained and validated on three distinct fundus image datasets: HRF, RIM1, and ACRIMA.
• Performance metrics including accuracy, sensitivity, and specificity were calculated for each dataset.

Main Results:

• The fused DenseNet and DarkNet model achieved high performance across all datasets.
• Exceptional results were observed on the HRF database with 99.7% accuracy, 98.9% sensitivity, and 100% specificity.
• The model demonstrated robust performance on the ACRIMA database (99% accuracy, 100% sensitivity, 99% specificity) and competitive results on RIM1.

Conclusions:

• The proposed hybrid DenseNet and DarkNet model is a robust and efficient tool for glaucoma detection from fundus images.
• This AI-driven approach shows significant potential for early and accurate diagnosis, aiding in the prevention of vision loss.
• The method offers advantages in computational time and complexity compared to existing literature.