Segmentation of corneal endothelium images using a U-Net-based convolutional neural network

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Corneal endothelial cell analysis is crucial for diagnosing eye health.
• Manual or semi-automatic cell segmentation is time-consuming and prone to errors.
• Existing automatic segmentation methods lack perfection.

Purpose of the Study:

• To develop an automated method for segmenting corneal endothelial cells using a U-Net convolutional neural network.
• To accurately identify cell borders for improved morphometric analysis.
• To enhance the diagnostic capabilities for corneal endothelium health.

Main Methods:

• A U-Net-based convolutional neural network was trained to detect pixel-level cell borders.
• The network's output edge probability map was binarized and skeletonized.
• The method was validated on 30 corneal endothelial images.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.92.
• Obtained an average DICE score of 0.86 for edge segmentation.
• Demonstrated a mean absolute percentage error of 4.5% for cell counting.
• Morphometric parameter errors ranged from 5.2% (density) to 11.93% (cell size variation).

Conclusions:

• The proposed U-Net model provides accurate and efficient automatic segmentation of corneal endothelial cells.
• The method significantly reduces manual correction needs for cell edge identification.
• This approach enables more precise morphometric analysis, aiding in corneal health diagnostics.