角膜振動特徵於青光眼之輔助診斷

Abstract

隨著青光眼成為高齡化社會中的主要視覺健康問題，更準確的早期診斷成為當前重要的研究課題。本研究旨在探討罹患青光眼對角膜應力表現的影響，以及角膜參數與青光眼判讀的相關性，以提高早期診斷的準確性和效率。研究使用了由Corvis® ST眼壓機拍攝的角膜影像，這些數據來自青光眼患者和正常受試者，並且收集自多個醫療機構的眼科門診。首先我們對角膜影像進行初步的處理，將影像二值化後對角膜上緣進行曲線擬合。隨後，將曲線進行Legendre展開並依照模態特徵建立參數，比較青光眼患者與正常受試者之間的差異，並使用統計方法找出最具顯著性的參數，最後利用機器學習作分類驗證。結果顯示，青光眼患者的角膜參數與正常受試者存在顯著差異(p值小於0.05)。使用機器學習模型分類的準確率達到82%以上。另外，開角型和閉鎖型青光眼患者之間在靠近鼻側和顳側的角膜參數也存在顯著差異(p值亦小於0.05)。使用機器學習模型分類的準確率則可達到70%以上。本研究證實了角膜參數在青光眼輔助診斷中的潛力。通過結合影像處理技術和機器學習分析，我們能夠提高青光眼早期診斷的準確性和效率。未來的研究可以進一步優化這些方法，並探討更多角膜參數的應用，以實現更全面的診斷系統。這對於應對高齡化社會中日益增長的青光眼患者具有重要意義。

As glaucoma becomes a major visual health issue in an aging society, more accurate early diagnosis has become a crucial research topic. This study aims to investigate the impact of glaucoma on corneal stress performance and the correlation between corneal parameters and glaucoma diagnosis, to improve the accuracy and efficiency of early diagnosis. The study utilized corneal imaging data captured by the Corvis® ST tonometer, collected from glaucoma patients and normal subjects from multiple ophthalmology clinics. First, we performed initial processing of the corneal images by binarizing the images and fitting curves to the upper edge of the cornea. Then, we expanded the curves using Legendre polynomials and established parameters based on modal characteristics to compare differences between glaucoma patients and normal subjects. We used statistical methods to identify the most significant parameters and finally used machine learning for classification validation. The results showed significant differences in corneal parameters between glaucoma patients and normal subjects (p-value < 0.05). The accuracy of the machine learning model classification reached over 82%. Additionally, significant differences were found in corneal parameters near the nasal and temporal sides between open-angle and closed-angle glaucoma patients (p-value also < 0.05). The classification accuracy of the machine learning model for these parameters reached over 70%. This study confirmed the potential of corneal parameters in assisting the diagnosis of glaucoma. By combining image processing techniques and machine learning analysis, we can improve the accuracy and efficiency of early glaucoma diagnosis. Future research can further optimize these methods and explore the application of additional corneal parameters to achieve a more comprehensive diagnostic system. This is of significant importance in addressing the growing number of glaucoma patients in an aging society.