Automatic Detection of Glaucomatous Optic Nerve Head from Optical Coherence Tomography Retinal Images

Purpose: To develop an automatic method for detection of glaucoma from Spectral domain optical coherence tomography (OCT) images by image processing algorithms instead of traditional manual time consuming and labor intensive detection method for glaucomatous optic nerve head (ONH) objects. Methods: This study included 299 glaucoma patients, and 576 healthy participants with good quality OCT Bscans images (768 × 496) taken with the Spectralis OCT-Heidelberg Engineering, Germany. The images were classified into normal or glaucomatous types by 2 glaucoma specialists. Randomly, 656 B-Scans were selected for training data and 219 for test data. A deep convolutional neural network (CNN), as the most successful and widely used deep learning model was trained with the training data and evaluated with the test data. Results: The method is developed with least complicated algorithms and the results show considerable improvement in accuracy of detection the glaucoma over similar methods. The automated classification results were compared to manual results from two glaucoma specialists. The validated accuracy against test data for the CNN was 95%. Conclusion: OCT analysis of the ONH is useful for early glaucoma detection. This method having an acceptable result can be effective in automatic diagnosis of glaucoma and the proposed machine learning system has proved to be good identifiers for different type of Optic disk with high accuracy.