Optimal U-Net Using Grey Wolf Algorithm for Retinal Vessel Segmentation

In the realm of medical imaging, retinal vessels segmentation (RVS) has a significant role for retinal diseases diagnosis, such as diabetic retinopathy, high blood pressure, arteriosclerosis, etc. In recent years, deep convolutional neural networks (DNNs) approaches have shown their effectiveness for RVS tasks, especially the u-shaped (encoder-decoder) based models. However, the manual design of these segmentation models is not trivial, and designing the optimal architecture is time-consuming, requires expert knowledge. Neural Architecture Search (NAS) is a widely used method for automatically de-signing neural network architectures within a predefined search space. In this paper, we propose the implementation strategy of three Metaheuristic optimization algorithms, such as Genetic Algorithm (GA), Grey Wolf Optimization (GWO) and Particle Swarm Optimization (PSO), to search for the optimal U-shaped (Encoder-Decoder) network and its hyperparameters for Retinal vessel segmentation (RVS). The proposed approach is evaluated on DRIVE dataset. The experimental results reveal that our GWO algorithm performs better than GA and PSO algorithms, which converge to more optimal solutions with fewer iterations.