CTS-Net: A Segmentation Network for Glaucoma Optical Coherence Tomography Retinal Layer Images.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) is crucial for non-invasive glaucoma diagnosis.
• Analyzing retinal layer thickness and shape aids in early glaucoma detection.
• Accurate retinal layer segmentation enhances diagnostic efficiency for ophthalmologists.

Purpose of the Study:

• To propose a novel deep learning method, CTS-Net, for precise retinal layer segmentation in OCT images.
• To improve the accuracy of boundary segmentation by focusing on edge regions.
• To validate the model's performance and generalization ability on a glaucoma retina dataset.

Main Methods:

• Development of a CSWin Transformer-based neural network (CTS-Net) for pixel-level retinal layer segmentation.
• Introduction of a boundary-aware Dice loss function (BADice Loss) to enhance edge feature learning.
• Application and testing of the CTS-Net model on a publicly available glaucoma retina dataset.

Main Results:

• CTS-Net achieved high accuracy with Mean Absolute Distance (MAD) of 1.79 pixels, Root Mean Square Error (RMSE) of 2.15 pixels, and Dice-Similarity Coefficient (DSC) of 92.79%.
• The model demonstrated superior performance compared to existing methods.
• Cross-validation experiments confirmed the model's generalization ability with minimal variation in performance metrics.

Conclusions:

• CTS-Net effectively performs pixel-level retinal layer segmentation, yielding smooth boundaries.
• The proposed BADice Loss function improves boundary segmentation accuracy.
• The model shows significant potential for enhancing glaucoma diagnosis through accurate OCT image analysis.