SCA-Net: A Scale- and Contrast-Aware Network for Subtle and Low-Contrast Polyp Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate polyp segmentation is crucial for early colorectal cancer detection.
• Existing methods face challenges with subtle polyps, weak boundaries, and cross-dataset generalization.

Purpose of the Study:

• To propose SCA-Net, a scale- and contrast-aware network for enhanced polyp segmentation.
• To improve semantic representation, scale adaptability, and boundary sensitivity in polyp segmentation.

Main Methods:

• Developed SCA-Net, a unified encoder-decoder framework.
• Introduced a semantic module group (SMG) with cross-scale global aggregator (CSGA) and gated semantic injection (GSI).
• Implemented a size-adaptive dynamic router (SADR) and a Laplacian-guided synergistic refiner (LGSR).

Main Results:

• SCA-Net demonstrated competitive performance on seen datasets.
• Achieved significant gains on challenging unseen benchmarks, including ETIS-LaribPolypDB.
• Attained 86.0% Dice score on ETIS-LaribPolypDB with a PVTv2-B4 backbone.

Conclusions:

• SCA-Net effectively addresses limitations in existing polyp segmentation methods.
• The proposed network enhances scale adaptability and boundary refinement for improved accuracy.
• SCA-Net shows promise for advancing early colorectal cancer detection through improved segmentation.