Retinal Microvessel Segmentation using Variants of UNet and Domain Adaptation

The use of deep learning in medical image segmentation has gained popularity, and one commonly utilised method is U-Net. U-Net is favoured for its ability to accurately identify small objects within images and its adaptable network structure. U-Net has received significant attention in academic literature, with over 2500 citations. Researchers have been continuously enhancing its architecture to improve its performance. This paper presents a proposal that focuses on various UNet variations for medical image segmentation. It explores their structure, innovative features, and overall efficiency. Our proposal uses multiple variants of U-Net and rotation-based domain adaptation. The paper also reviews the application of domain adaptation on one dataset to get better results, which will be helpful for future research