Study of Loss Functions on Retinal Vessel Segmentation using UNET Architecture

Using deep learning in image segmentation is gaining popularity with many industries adopting it and many researchers working to improve its performance. One of the areas where image segmentation is used is in medical image segmentation where a region of interest is segmented from a medical image. One of the key aspects that determine the performance of a deep learning model used to perform segmentation task is its loss functions used in training the model. This paper aims to compare the performances of a unet model in segmenting the vessels of a fundus image using different popularly used loss functions. Combinations of some of the loss functions are also used to train the model and their performances are also studied.