•  
  •  
 

Abstract

The ophthalmologist uses various techniques to diagnose corneal abnormalities, including corneal topography and tomography devices. Currently generated color corneal elevation surface maps from topographic imaging devices are essential for detecting ocular diseases, while accurately classifying these maps to differentiate between different shapes remains an issue. This study aims to assess and compare parameters of the front and back corneal surface elevation map patterns of normal/abnormal corneas. Two hundred cases were randomly taken (100 normal and 100 abnormal) with a single normal reference image, and then an additional 25 cases were added later for the optimization process. The preprocessing of all images involved converting them from color to black and white images to highlight the elevation bowtie shapes. The correlation coefficient, as implemented in MATLAB, was used to identify similar shapes of bowties for both normal and abnormal cases and determine the elevation values at the thinnest location. Additionally, it was used to measure the elevation bowtie angle between the reference normal image (case) and all other cases for image classification. Individually, the cut-off value of maximum correlation (mean ± 2 standard deviation) was calculated to distinguish between normal/abnormal patterns for 200 cases, and the cut-off values (0.974418 ≥ correlation ≥ 0.575, 0.994548 ≥ correlation ≥ 0.537841) were used for elevation back and front, respectively. Normal angle has cut-off values of (≤ 25 and =90), while normal elevation values at the thinnest location are (≤ 15, ≤ 20) for front and back surfaces. This method is based on elevation back, achieving an accuracy of 74%, and an accuracy of 64% was achieved for elevation front. Our model’s final accuracy was calculated by combining both the elevation back and elevation front final results to achieve 75.5% accuracy. Adding more data will not further increase the accuracy output since it has reached its optimal level. The development of this method aims to assist ophthalmologists in diagnosing and confirming clinical decisions through the accurate detection of corneal disease.

Download

Share

COinS