Ramdass_Jermaine.pdf (2.14 MB)

Prediction of Cup-to-Disc Ratio From Optic Fundus Image

Download (2.14 MB)
posted on 24.05.2021, 07:45 by Jermaine Ramdass
A technique is proposed that can be used to predict the cup-to-disc ratio from a single optic fundus image and determine which image features have the highest contribution to a specific ophthalmologist’s measured cup-to-disc ratio. The procedure starts with image pre-processing. The main step of the procedure is feature extraction where image features related to pixel intensities are found. These features are used to train three different classifiers: neural networks, support vector machines, and sparse representation classifiers. The classifiers are tested and evaluated to see how accurately they can predict the cup-to-disc ratio. The best obtained results are in the 70-75% success range. Finally, feature ranking is performed using the methods of chi square and information gain on a combined feature vector using measured cup-to-disc ratios from each ophthalmologist to determine the importance and contribution of each feature to that ophthalmologist.





Master of Engineering


Electrical and Computer Engineering

Granting Institution

Ryerson University

LAC Thesis Type