Computer Vision and Machine Learning for Glaucoma Detection
MetadataShow full metadata
Deep learning based on computer vision and machine learning is an emerging technology in both the medical imaging industry and academia. Despite the existence of some commercial glaucoma detection systems such as retinal imaging, OCT scans, and ocular tonometry, we are at the beginning of a long research pathway toward the future generation of intelligent glaucoma detection systems. Early glaucoma diagnosis prevents permanent structural optic nerve damage and consequent irreversible vision impairment. Longitudinal studies have described both baselines structural and functional factors that predict the development of glaucomatous change in ocular hypertensive and glaucoma suspects. Although there is neither a gold standard for disease diagnosis nor progression, photographic assessment of the optic nerve head remains a mainstay in the diagnosis and management of glaucoma suspects and glaucoma patients. This thesis discusses several image processing techniques comprising disparity map, superpixel and noise removal for pre-processing. A stack of traditional classifiers was utilized as a hybrid model based on the ensemble method to generalise and boost the performance of the proposed model to detect glaucoma through the thickness of the retina. A method was needed aiming at both detecting pathologic changes characteristic of glaucomatous optic neuropathy in optic disc images, and classification of images into categories of glaucomatous/suspect or normal optic discs. Therefore, different machine learning algorithms were used including transfer learning, deep convolutional neural networks, and deep multilayer neural networks that extract features automatically based on clinically relevant optic disc features. Meanwhile, biomarkers were demonstrated with the proposed deep learning model to interpret which parts of the retina had been affected by glaucoma. Finally, this research proposes methods based on evolving deep pre-trained learning architecture, stereo matching with the usage of disparity maps, hybrid models with statistical analysis to retinal nerve fibre layer (RNFL) classification, and visualization of biomarkers with deep learning to detect glaucoma in early stages based on fundus images. Besides, in Appendix A; we discuss a hypothesis of glaucoma detection through detecting specified pattern with signal processing and video processing to achieve glaucoma detection at its early stages. Thus, we are going to specify the OKN pattern of eye movement to detect glaucoma at its initial stage.