Detection of retinal OCT images using DETR deep learning technique

This thesis presents a study using a deep learning-based method, DETR (Detection Transformer), for the identification of optical coherence tomography (OCT) images. OCT images, which provide high-resolution cross-sections of the retinal layers, play a crucial role in the early diagnosis of eye diseases. Considering the limitations of traditional methods, deep learning techniques offer promising results in this field. In this study, 435 randomly selected images from the dataset, including both retina disease cases and control healthy retina images, were used. Thus, the model's performance was evaluated with both diseased and healthy samples.To assess the accuracy of the trained model, a series of experiments were conducted on images divided into training and test datasets. The results indicate that the DETR model is able to detect retinal deformations with 98% precision and recall. On the other hand, the YOLO model detects retinal deformations with 91% precision and 98% recall. This study demonstrates that the deep learning-based DETR model is an effective tool for the automatic identification of OCT images and its applicability in clinical settings. The results also provide a foundation for future research on larger datasets and various eye diseases. Keywords: DETR (Detection Transformer), Optical Coherence Tomography (OCT), Retinal Deformation, Deep Learning, Image Identification, Eye Disease,YOLO