Classification of retinal diseases based on oct data by extended lateral inhibition method

This thesis aims to develop an innovative deep learning model that integrates the principle of lateral inhibition for the automatic classification of retinal diseases based on Optical Coherence Tomography (OCT) images. Retinal diseases are a major cause of vision loss worldwide, making early and accurate diagnosis critically important. Manual analysis of OCT images using traditional methods is time-consuming and prone to errors, which highlights the need for machine learning-assisted analytical techniques. In recent years, deep learning approaches have achieved high accuracy rates in the classification of retinal diseases. However, the potential of biologically inspired approaches in this domain remains insufficiently explored. In this study, inspired by the lateral inhibition mechanism found in biological neural networks, a model was developed that integrates this principle into deep learning architectures to achieve more effective classification of retinal diseases. The results obtained using popular CNN architectures such as DenseNet121, InceptionV3, ResNet50, and Xception demonstrate that the inclusion of lateral inhibition significantly enhances overall model performance. Comparative analyses revealed that models incorporating lateral inhibition achieved higher F1-scores and accuracy, particularly in the CNV and DME classes. The findings indicate that integrating the principle of lateral inhibition into deep learning models leads to a substantial improvement in the classification of OCT images. It is anticipated that this approach will contribute to the early and accurate diagnosis of retinal diseases in clinical settings. Future studies are expected to examine the impact of lateral inhibition across different medical imaging modalities and larger datasets, suggesting that this method could be widely adopted in clinical applications.