Scopus Indexed Publications

Paper Details


Title
Attention-Guided U-Net for Robust and Accurate Polyp Segmentation in Colonoscopy Images

Author
, Tamim Mahmud,

Email

Abstract

Colorectal cancer (CRC) remains one of the leading causes of cancer-related mortality worldwide. Early detection and removal of precancerous polyps can significantly reduce both the incidence and mortality of CRC. Colonoscopy is considered the gold standard for CRC screening; however, its accuracy heavily depends on operator expertise and is susceptible to human errors, fatigue, and inter-observer variability, which can lead to missed or mischaracterized polyps. In this work, we propose an Attention-Guided U-Net model for accurate polyp segmentation in colonoscopy images. Attention gates are incorporated into the skip connections of U-Net to help the network focus on target regions while suppressing background noise. For fair comparison, a vanilla U-Net serves as the baseline, and DeepLabV3++ is used as a state-of-the-art reference. Experiments on publicly available colonoscopy datasets demonstrate that the proposed Attention-Guided U-Net outperforms the baseline models, achieving a validation Dice score of 93.30% and an Intersection over Union (IoU) of 87.53%. High precision and recall further confirm the robustness of the model. Qualitative results show improved boundary delineation, particularly for small, flat, and low-contrast polyps. These results indicate that attention-based feature selection effectively enhances segmentation performance, suggesting the proposed framework as a reliable computer-aided diagnosis (CAD) tool for CRC screening.


Keywords

Journal or Conference Name
2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence and Networking, QPAIN 2026

Publication Year
2026

Indexing
scopus