MCA-Net: multi-cascade attention network for polyp segmentation

Accurate polyp segmentation is crucial for diagnosing colorectal cancer, but it remains challenging due to shape, texture, and scale variations, as well as difficulties in determining boundaries. Existing methods incorporating attention mechanisms have improved accuracy but lack effective fusion of...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications 2024-03, Vol.83 (11), p.33713-33730
Main Authors: Liu, Yitong, Shen, Xuanjing, Lyu, Yingda, Wang, Xue
Format: Article
Language:English
Subjects:
Algorithms
Background noise
Colorectal cancer
Computer Communication Networks
Computer Science
Data Structures and Information Theory
Datasets
Modules
Multimedia Information Systems
Multiscale analysis
Polyps
Segmentation
Special Purpose and Application-Based Systems
Track 2: Medical Applications of Multimedia
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Accurate polyp segmentation is crucial for diagnosing colorectal cancer, but it remains challenging due to shape, texture, and scale variations, as well as difficulties in determining boundaries. Existing methods incorporating attention mechanisms have improved accuracy but lack effective fusion of different level features. Besides, extracting boundary information from surrounding mucosa and early polyps poses challenges. To tackle these issues, a multi-cascade attention-based network (MCA-Net) is proposed. Three components are introduced, including an axial receptive module (ARM), a multi-cascade feature aggregation module (MFA), and an edge fusion module (EFM). The ARM enhances multi-scale analysis by incorporating receptive fields and axial attention, providing the algorithm with a better knowledge of the features. Along with the integration of multi-cascade supervision, MFA selectively refines the information, effectively fusing relevant cues from different levels, suppressing background noise, and highlighting essential polyp features. The EFM focuses on capturing object boundary details, resulting in well-defined and accurate segmentation. Experiment results on five polyp datasets show that our MCA-Net outperforms state-of-the-art (SOTA) methods. Specifically, our MCA-Net achieves an 8.2% improvement in mean Dice compared to the state-of-the-art on the ETIS dataset.
ISSN:1573-7721
1380-7501
1573-7721
DOI:10.1007/s11042-023-16805-9