Detection of Eye Pressure Disease Using ViT-Based Hybrid Learning Methods

Abstract

Glaucoma is a disease that occurs after a certain age due to damage to the optic nerves. Today, machine learning methods can be successfully applied to detect such diseases. Instead of using the image data directly, the classification process is carried to a new representation space, which positively affects the classification performance. In this study, principal component analysis (PCA), linear discriminant analysis (LDA) and vision transformation (ViT) methods are used for feature extraction. In addition, CLAHE filtering technique before ViT B16 was used in one of the proposed models. The classification process was performed with six different models using these methods alone or in combination, and the results are presented comparatively. Fine Tune ViT-PCA-SVM and Fine Tune ViT-LDA-SVM models achieved 92% classification success. As a result, the combination of ViT, which is a deep learning method, and PCA or LDA, which are machine learning methods, as feature extraction methods increased the classification success.