The paper presents a segmentation method of the artery and vein in the fundus retinal image using the deep learning architecture U-Net. In the segmentation of the artery and vein, the architecture is hyper tuned using the optimizers at a learning rate of 0.001 and 0.01. The segmentation of artery and vein is performed on the retinal fundus imaging, that can be helpful for the early assessment of traumatic conditions in patients’ diagnosis with the head injury. The traumatic conditions are related to the intracranial pressure (ICP) which is correlated with artery-vein diameter, optic disc cup ratio, and retinal tortuosity. It has been concluded that these parameters containing information for multiple measures with and without papilledema symptoms can be a breakthrough for the traumatic disorder. The performance of the segmented artery and vein is evaluated for three different optimizers by evaluating the optimal accuracy, sensitivity and specificity. The optimal accuracy for the NADAM optimizer at a learning rate of 0.001 and batch size of 8 is 93.70%. Further, sensitivity and specificity were computed for the DRIVE database which found to be 92.35 and 93.41 respectively.