POLYP SEGMENTATION IN COLONOSCOPY IMAGES

Colorectal cancer is the third most common diagnosed cancer. Colonoscopy is the gold standard to screen colon for polyp detection and localization. Early detection and removal of adenoma may increase the survival probability. However, there are still misses due to the complex structure of the colon and human mistakes [5]. To minimize the misses, a computer-aided diagnosis (CAD) method needs to be developed to segment and localize the polyps in colonoscopy images. In this project, Convolutional Neural Network (CNN) with transfer learning was proposed as the method for automatic polyp segmentation. Publicly available databases, CVC-ClinicDB, CVC-ColonDB, and ETIS-LaribPolypDB have been used to train and test the model. A CNN named VGG-16 has been chosen as the network for transfer learning, the results obtained during the testing in terms of accuracy, sensitivity and specificity are 91.57%, 91.83% and 91.31% respectively.