Abstract Colorectal cancer is among the most common malignancies and can develop from high-risk colon polyps. Colonoscopy is an effective screening tool to detect and remove polyps, especially in the case of precancerous lesions. However, the missing rate in clinical practice is relatively high due to many factors. The procedure could benefit greatly from automatic polyp segmentation models, which provide valuable insights for colon polyp detection improvement. How- ever, precise segmentation is still challenging due to the variation of polyps in size, shape, texture, and color. This paper proposes a novel neural network architecture called AG-CUResNeSt, which enhances Coupled UNets using the robust ResNeSt backbone and attention gates. The network is capable of ef- fectively combining multi-level features to yield accurate polyp segmentation. Experimental results on five popular benchmark datasets show that our pro- posed method achieves state-of-the-art accuracy compared to existing methods.
Download here Google Driver
Put it into ./checkpoints
Download here Google Driver
Put it into .datasets
In configs.py, choose any block and uncomment it
Run the following command
run.shJust enjoy it!
Other comparison results Google Driver
Author
Chung Tran Quang: Hanoi University of Science and Technology
Sang Dinh Viet: Hanoi University of Science and Technology
Contact
If you have any question, please contact me via email: bktranquangchung@gmail