Corneal Endothelial Cell Segmentation by Classifier-Driven Merging of Oversegmented Images

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate corneal endothelium segmentation is crucial for assessing corneal health.
• Existing automated segmentation methods struggle with low-contrast images and variable cell sizes.

Purpose of the Study:

• To develop an automatic method for segmenting corneal endothelium images.
• To improve the accuracy and precision of clinical parameter estimation from these images.

Main Methods:

• Utilized stochastic watershed segmentation to create superpixels.
• Employed support vector machines with intensity and shape information to merge superpixels into cells.
• Evaluated segmentation on in vivo specular microscopy images.

Main Results:

• Achieved 95.8% correctly merged cells and 2.0% undersegmented cells in automated segmentation.
• Demonstrated statistically significant improvements in precision for all estimated parameters compared to vendor software.
• Showed superior accuracy and precision across multiple imaging modalities and tissue types.

Conclusions:

• The proposed method offers a robust solution for corneal endothelium segmentation.
• This technique enhances the reliability of corneal health assessment through improved parameter estimation.
• The approach shows potential for broader applications in ocular tissue analysis.