Localization of Region of Interest from Retinal Fundus Image for Early Detection of Glaucoma

Glaucoma is one of the serious ocular conditions that lead to permanent vision loss if not detected at the early stages of the disease. With the emergence of machine-learning techniques, computer-aided diagnosis is gaining prominence in the early detection of glaucoma. To an extent, this helps ophthalmologists in screening the eye to identify the progression of glaucoma. Convolutional neural networks (CNN) have gained remarkable significance in medical image analysis, and thus a deep CNN model has been explored in this paper to identify the changes brought in by glaucoma. A large number of glaucomatous and non-glaucomatous fundus images collected from a renowned eye institute have been used in this work. An appropriate region of interest is selected from these images as a part of pre-processing done using OpenCV and fed to the deep CNN architecture. The model is trained and tested with these pre-processed images, and the results obtained indicate that this technique has better accuracy and sensitivity.