Papers › LACFormer: Toward accurate and efficient polyp segmentation
LACFormer: Toward accurate and efficient polyp segmentation
Quan Van Nguyen, Mai Nguyen, Thanh Tung Nguyen, Huy Trịnh Quang, Toan Pham Van
Polyp segmentation is an essential task in medical image analysis for early detection of colorectal cancer. Deep learning models, particularly encoder-decoder architectures, have been successful in polyp segmentation. However, these models often struggle to capture long-range dependencies and exhibit limited performance on small polyps. In this paper, we propose LACFormer, a novel hierarchical Transformer-CNN model incorporating the Laplacian pyramid for polyp segmentation. The proposed model combines the strengths of Transformers and CNNs along with Laplacian images to overcome the limitations of previous models. Specifically, the hierarchical Transformer backbone captures long-range dependencies and hierarchically processes the features to generate multi-scale representations. These representations are then fused with a novel CNN decoder, which enhances feature representations and refines the segmentation masks. Besides, many novel modules for effective polyp segmentation are also proposed. We evaluated our model on five popular benchmark datasets for polyp segmentation, including Kvasir, CVC-Clinic DB, CVC-ColonDB, CVC-T, and ETIS-Larib. Experimental results show that LACFormer outperforms state-of-the-art models, achieving a Dice similarity coefficient (DSC) of 0.927 and a mean intersection-over-union (mIoU) of 0.878 on CVC-ClinicDB, a DSC of 0.831 and mIoU of 0.753 on CVC-ColonDB and a DSC of 0.824 and mIoU of 0.753 on ETIS-Larib.
Code
Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.
Code Syntology ran Syntology
Not run by Syntology. Nothing on this page verifies that the listed code works.
Tasks
Results from the paper archive 2025-07-28
No leaderboard rows for this paper in the archive.
Methods
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections