Effect of U-Net Hyperparameter Optimisation in Polyp Segmentation from Colonoscopy Images

Abstract

Colorectal cancer can be detected from the location of polyps present in the colon. Colonoscopy is the standard method for polyp detection from where the images are retrieved. Deep neural networks are preferred for the segmentation of polyps from colonoscopy images. The hyperparameters directly control the performance of the neural network model. In this work, the impacts of different neural network optimizers and various activation functions are analyzed for U-net architecture. The analysis shows that the N- Adam optimizer with Sigmoid activation function yields better accuracy of 94.06% for polyp segmentation from colonoscopy images.