Enhancing Optic Nerve Head Swelling Detection Through Deep Learning Classification of 3D OCT Images

Purpose: To leverage deep learning (DL) to process optical coherence tomography (OCT) images for the detection of optic nerve head (ONH) abnormalities, particularly optic disc swelling. Methods: The study employed a dataset of 795 pre-processed OCT images, comprising 561 images with structural abnormalities and 234 images with normal ONH structures. A DL architecture, namely a 3D ResNet model, was implemented to classify the images. Results: The 3D ResNet model achieved an accuracy of 93.7%, an F1-score of 0.87, and an AUC of 0.96 on the test dataset, demonstrating high precision in distinguishing between normal and abnormal ONH structures. Conclusion: The application of 3D ResNet for classifying ONH OCT images demonstrated promising results for the automated detection of ONH swelling. This approach may support clinical decision-making by improving diagnostic accuracy and reducing reliance on manual image interpretation.