Semi-automated Method for the Glaucoma Monitoring

The current trend of computer vision and image processing systems in biomedical field is the application of the Computational Intelligence (CI) approaches, which include the use of tools as machine learning and soft computing. The CI approaches bring a new solution to automatic feature extraction for a particular task. Based on that techniques, we have proposed in this work a semi-automated method for the glaucoma monitoring through retinal images. Glaucoma is a disease caused by neuro-degeneration of the optic nerve leading to blindness. It can be assessed by monitoring Intraocular Pressure (IOP), by the visual field and the aspect of the optic disc (ratio cup/disc). Glaucoma increases the rate of cup/disc (CDR), which affects the loss of peripheral vision. In this work, a segmentation method of cups and discs regions is proposed in a semi-supervised pixel-based classification paradigm to automate the cup/disc ratio calculation for the concrete medical supervision of the glaucoma disease. The idea is to canvas the medical expert for labeling the regions of interest (ROI) (three retinal images) and automate the segmentation by intelligent region growing based on machine learning. A comparative study of semi-supervised and supervised methods is carried out in this proposal, by mono approaches (decision tree and SETRED) and multi-classifiers (Random Forest and co-Forest). Our proposition is evaluated on real images of normal and glaucoma cases. The obtained results are very promising and demonstrate the efficacy and potency of segmentation by the multi-classifier systems in semi-automatic segmentation.