Polyp Segmentation in Colonoscopy Images using U-Net and Cyclic Learning Rate

Colonoscopy is an important procedure in the diagnosis of colorectal cancer. The use of computer aided systems has become important to support clinicians performing colonoscopy and to prevent polyps from escaping the clinician's attention. Image segmentation studies using deep learning achieves successful results and can play a crucial role on diagnosis procedure of colorectal cancer. We trained Unet architecture for polyp segmentation and determined the learning rate, one of the most important training parameters, using Cyclic Learning Rate policy. The results show that the success rate is increased in the segmentation task performed Unet with Cyclic Learning Rate policy. In this study, we have contributed to more accurate detection of polyp diagnosis, which can be a precursor to cancer, by using the UNET architecture with an effective learning rate strategy.