Abstract:
The incidence rate of colorectal cancer is high worldwide, but the survival rate is
greatly improved by early identication of polyps. Colonoscopy is the gold stan dard procedure for diagnosis and removal of colorectal lesions with the potential to
evolve into cancer. Computer-aided detection systems can help gastroenterologists
find polyps more accurately, which is one of the key signs of a good colonoscopy
and a predictor of the likelihood of developing colorectal cancer. The size, shape,
and texture of the colorectal polyp is challenging factor for computer aided detection
systems. Mask RCNN is one of computer aided detection and segmentation sys tem to alleviate these challenges. In this study, we use Mask R-CNN for colorectal
polyp segmentation in colonoscopy images. We used a large dataset and optimized
the existing Mask RCNN model to improve its performance. We have used publicly
available datasets for training 3,946 images and for testing 444 images. The trained
model achieved a mean average precision of 95.75%, mean average recall of 96%
and f1 score of 95.87%. Thus, optimizing the hyperparameter of Mask R-CNN us ing a large colonoscopy image dataset improves its polyp segmentation performance
in colonoscopy images. The proposed method improves the early diagnosis of col orectal polyps and helps to save lives
