A Topological-Aware Automatic Grading Model Corneal Epithelial Damage Evaluation from Full Corneal Fluorescence Staining Images
Zi-Kai Ren,Jun Feng,Lei Tian,Kai-Ni Wang,Jing-Yi Wang,Yuan-Chao Shu,Yi-Ran Hao,Ying Jie,Guang-Quan Zhou
DOI: https://doi.org/10.1016/j.compbiomed.2024.109451
IF: 7.7
2024-01-01
Computers in Biology and Medicine
Abstract:Corneal Fluorescence Staining (CFS) imaging is commonly employed for assessing corneal epithelial damage. Automating the grading of CFS images can minimize subjectivity in clinical evaluations and enhance diagnostic efficiency. Existing methods typically depend on the texture and morphological information extracted from whole CFS images, often neglecting the spatial and distribution information between stained regions. This oversight hinders the accurate evaluation of corneal epithelial injury states. This study proposes a three-stage automatic corneal epithelial damage assessment model for full CFS images, optimizing grading by considering topological features among detected stained regions, which are crucial for accurately interpreting the spatial properties of objects within an image. Accurate corneal localization, robust to variations in contrast, is first achieved by integrating CFS images' intensity and phase information, subsequently by a multi-scale morphological top-hat operator concerning their prior shape to detect the stained regions. Finally, a multi-scale graph structure is constructed based on the detected stained areas, and distance-weighted topological features, along with textural and morphological features, are extracted into an automatic grading model based on an ensemble model. Experiments on an in-house dataset of CFS images annotated with six categories of Ocular Surface Staining (OSS) scores reveal that incorporating topological features achieves the highest Accuracy (0.7589), F1 score (0.7449), and AUC (0.9335). Moreover, topological features outperformed other individual features. These findings underscore the effectiveness of our proposed model in CFS grading, indicating its potential for assessing corneal epithelial damage. Additionally, the valuable insights provided by topological features into the spatial distribution patterns of staining suggest promising applications for enhancing disease classification and supporting more informed clinical decision-making in managing dry eye conditions.