Automated Detection of Tight Junction Damage in Corneal Endothelium using Machine Learning

The cornea is a critical component of the human eye, essential for forming clear images on the retina. It comprises multiple layers, each contributing to the transparency and health of the tissue. The corneal endothelium, a key structure within the cornea, plays an important role to maintain clarity through its tight junctions, primarily constituted by Zonula Occludens-1 (ZO-1). It is a cytoplasmic protein located in the outer layers of endothelial cells and is crucial for the integrity of these tight junctions. This work aims to detect damage to ZO-1 in endothelial cells by applying Gabor filters to ZO-1 images, followed by feature extraction and the application of various machine learning techniques. The machine learning algorithms used include Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree, Naïve Bayes, and Stochastic Gradient Descent (SGD). Our findings show that the KNN algorithm outperforms the others, achieving the highest accuracy in identifying ZO-1 damage, while SGD exhibited the lowest performance. Consequently, KNN appears to be the most effective classifier for detecting ZO-1 damage, offering valuable support for doctors and medical professionals in making quick, accurate decisions regarding corneal endothelial health.