Automatic Polyp Segmentation Using Modified Recurrent Residual Unet Network

Colorectal cancer is a dangerous disease with high mortality rate. To increase the likelihood of successful treatment, early detection polyps useful solution. The Unet-architecture network model showing success in medical image segmentation including analysis from colonoscopy images. Traditional Unet and Unet-based models are often huge, requiring training deployment high-performance system. Designing compact size performance would be an important goal. In this study, we proposed to modify Residual Recurrent architecture improve while ensuring performance. has flexibility changing number filters convolution units. By taking advantage strengths residual recurrent structures terms reuse convolutional functions, new model, therefore, was not only smaller but also superior compared traditional others. evaluations were performed on three public Colonoscopy datasets: CVC-ClinicDB, ETIS-LaribPolypDB, CVC-ColonDB. Dice score CVC-ClinicDB reached 94.59%, ETIS-LaribPolypDB 92.73% 93.31% CVC-ColonDB dataset. experimental results obtained datasets better than those recent related studies. introduced nevertheless outstanding performance, it extremely productive for developing applications low-performance devices.