Blood Vessel Segmentation for Retinal Fundus Disease Detection

Multiple eye diseases can be easily detected from the orientation of the fundus and the blood vessels on the fundus of the eye. Most of the already present methodologies use methods like Gaussian filter, channel extraction and other preprocessing techniques for blood vessel segmentation. Which does not really solve the problem of giving faster and better results. Our approach involves data augmentation on the obtained samples as the number of samples is less and augmentation of these samples would result in better performance and efficiency of our model. We use an Encoder-Decoder Model, which is coded for segmenting the blood vessels from the samples. Hence providing more clear and accurate segmentation. Deep Learning Models like VGG-16, VGG-19, and ResNet50 can be applied to the blended images for predicting if the patient suffers from Retinal Disease or not. Since we use deep learning techniques and complex models, highly accurate results are obtained. This could be a very efficient alternative to reduce the efforts of an ophthalmologist and a major step towards automation in the field of ophthalmology.