Automated Measurement of Pterygium Progression Rate Using Anterior Segment Image Segmentation

Abstract

Purpose : The progression of pterygium towards the corneal center affects corneal refraction and increases higher-order aberrations. Accurate and convenient automatic measurement of this progression is essential. This study aimed to assess the effectiveness of an automated system using semantic segmentation of the pterygium and cornea for measuring the progression rate.

Methods : This study used anterior segment images from Tsukazaki Hospital (363 eyes, 443 images for training and validation) and Miyata Eye Hospital (142 eyes, 142 images for testing). Half of the training and validation images were from normal eyes, and the other half from pterygium-affected eyes. Annotations of the cornea and pterygium regions were validated by a corneal expert. A detection model was developed using U-Net and DenseNet-121. Corneal diameters and limbal regions were obtained from the segmented cornea, and the end of the pterygium was identified from the segmented pterygium. The progression rate was calculated as the ratio of the end of the pterygium to the corneal diameter. The evaluation included Pearson correlation coefficient, Mean Absolute Error (MAE), and 95% limits of agreement in the Bland-Altman plot, comparing the progression rates from annotations and segmentation results.

Results : The automated system's estimated progression rates showed a high correlation with expert annotations (correlation coefficient 0.879). The MAE was 2.75%, with 95% limits of agreement in the Bland-Altman plot ranging from -0.16 to 0.12.

Conclusions : The results demonstrate the high accuracy of the system in measuring pterygium progression rate, indicating its potential for clinical application. Future efforts will be directed towards further refining the system and exploring its viability as a practical diagnostic support tool.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.