Enhanced RNFL Thickness Estimation with Cost Function Approach from Fundus Images via TSNIT Graph Mapping.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal nerve fiber layer (RNFL) thickness estimation is crucial for diagnosing and monitoring glaucoma.
• Current methods often rely on optical coherence tomography (OCT) data, which can be inaccessible.
• Transforming fundus images into OCT-equivalent RNFL maps offers a potential alternative.

Purpose of the Study:

• To develop a more efficient method for estimating RNFL thickness using only fundus images.
• To create OCT-equivalent RNFL maps from fundus images for improved glaucoma assessment.
• To validate the accuracy of the proposed method against established ground truth data.

Main Methods:

• A novel autoencoder model was developed incorporating a cost function that considers global and local RNFL thickness.
• The model integrates fundus images with TSNIT graphs for accurate RNFL thickness map generation.
• A dataset of 443 fundus images from glaucoma, glaucoma-suspected, and healthy individuals was analyzed.

Main Results:

• The study achieved high micro-average peak signal-to-noise ratio (PSNR) of 24.92 dB and structural similarity index (SSIM) of 91.14%.
• Consistent PSNR and SSIM values were observed across glaucoma (26.26 dB, 92.30%), glaucoma-suspected (25.07 dB, 91.91%), and no glaucoma (24.41 dB, 90.73%) groups.
• The method demonstrated improved RNFL thickness estimation and TSNIT graph accuracy solely from fundus images.

Conclusions:

• The proposed method effectively transforms fundus images into OCT-equivalent RNFL maps.
• This innovative approach enhances the accuracy of RNFL thickness estimation without requiring raw OCT data.
• The findings suggest a promising tool for glaucoma diagnosis and management using readily available fundus images.