Evaluation of the corneal topography based on deep learning

Area of Science:

• Ophthalmology
• Computer Science
• Medical Imaging

Background:

• Corneal topography evaluation is crucial for orthokeratology (OK) lens fitting.
• Traditional methods can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate a deep learning-based method for corneal topography evaluation.
• To assess the accuracy of this method in segmenting key corneal zones and calculating relevant indicators for OK lens fitting.

Main Methods:

• Retrospective analysis of clinical data from 1,302 myopic subjects.
• Utilized U-Net neural networks for segmenting pupil and treatment zones.
• Employed image processing algorithms to calculate decentration and effective defocusing contact range.

Main Results:

• Achieved high precision and recall for treatment zone (0.9587, 0.9459) and pupil segmentation (0.9771, 0.9712).
• The deep learning method demonstrated over 98% accuracy in classifying corneal topography compared to optometrist assessments.

Conclusions:

• The developed deep learning algorithm provides an effective and accurate approach for corneal topography evaluation.
• This method serves as a valuable auxiliary tool for optometrists in selecting optimal OK lens parameters.