Visionary Insights: CNN-Driven Retinal Vessel Segmentation in Ophthalmic Imaging
S. Sivagami, M Gopinath, Y. J., V Srivarshinie, R Monikaa, K. Senthil
Abstract
Retinal vessel segmentation plays a vital role in the analysis of ophthalmic imaging. Binary Segmentation is one type of segmentation that segments the retinal vessel and blood vessel in retinal images. Retinal vascular geometry reflects a patient's state of health and aids in the diagnosis of certain conditions, including hypertension and diabetes. Early detection of these diseases helps to prevent blindness of patients. Machine Learning and Deep Learning are two important sectors in Artificial Intelligence. Nowadays, Deep learning algorithms are very much used in Medical Image processing. These Deep Learning Algorithms are more accurate and efficient than manual segmentation and other computer-aided diagnosis methods. Especially Convolutional Neural Networks (CNNs) in Deep Learning Architecture are used for the segmentation process. CNN architecture can handle the complex characteristics of retinal images. Traditional Data augmentation techniques are used to enlarge the training and testing samples. Matched filter methods are used to remove the noise and Morphological operations are implemented to detect the abnormal regions in the retinal image. Our proposed model achieves an accuracy of 97.87%, precision of 96.8%, Intersection over Union of 97.1%, and specificity of 0.5% in the segmentation process. The proposed CNN architecture not only expedites the segmentation process but it makes advancements in clinical decisions in the field of ophthalmology.