MAUNet: Polyp segmentation network based on multiscale feature fusion of attention U‐shaped network structure DOI

Jianwu Long,

Jian Lin,

Jiayin Liu

et al.

International Journal of Imaging Systems and Technology, Journal Year: 2024, Volume and Issue: 34(3)

Published: May 1, 2024

Abstract Colorectal cancer is a prevalent malignant tumor affecting the digestive tract. Although colonoscopy remains most effective method for colon examination, it may occasionally fail to detect polyps. In an effort enhance detection rate of intestinal polyps during colonoscopy, we propose MAUNet, polyp segmentation network based on multi‐scale feature fusion Attention U‐shaped structure. Our model incorporates advanced components, including Receptive Field Block, Reverse and Residual Refinement Module, mirroring analytical process performed by medical imaging professionals. We evaluated performance MAUNet five challenging datasets conducted comparative analysis against state‐of‐the‐art models using six evaluation metrics. The experimental results demonstrate that achieves varying levels improvement across datasets. Particularly Kvasir dataset, Mean Dice IOU metrics reached 91.6% 84.3%, respectively, confirming model's outstanding in segmentation.

Language: Английский

MDER-Net: A Multi-Scale Detail-Enhanced Reverse Attention Network for Semantic Segmentation of Bladder Tumors in Cystoscopy Images DOI Creative Commons

Chao Nie,

Chao Xu, Zhengping Li

et al.

Mathematics, Journal Year: 2024, Volume and Issue: 12(9), P. 1281 - 1281

Published: April 24, 2024

White light cystoscopy is the gold standard for diagnosis of bladder cancer. Automatic and accurate tumor detection essential to improve surgical resection cancer reduce recurrence. At present, Transformer-based medical image segmentation algorithms face challenges in restoring fine-grained detail information local boundary features have limited adaptability multi-scale lesions. To address these issues, we propose a new detail-enhanced reverse attention network, MDER-Net, robust segmentation. Firstly, efficient channel module (MECA) process four different levels extracted by PVT v2 encoder adapt changes tumors; secondly, use dense aggregation (DA) aggregate advanced semantic feature information; then, similarity (SAM) used fuse high-level low-level features, complementing each other position finally, (DERA) capture non-salient gradually explore supplementing addition, space (ECSA) that enhances context improves performance suppressing redundant features. Extensive experiments on dataset BtAMU, established this article, five publicly available polyp datasets show MDER-Net outperforms eight state-of-the-art (SOTA) methods terms effectiveness, robustness, generalization ability.

Language: Английский

Citations

1
Gca-pvt-net: group convolutional attention and PVT dual-branch network for oracle bone drill chisel segmentation DOI Creative Commons
Guoqi Liu, Yiping Yang,

Xueshan Li

et al.

Heritage Science, Journal Year: 2024, Volume and Issue: 12(1)

Published: July 29, 2024

Abstract Oracle bones (Obs) are a significant carrier of the shang dynasty civilization, primarily consisting tortoise shells and animal bones, through study which we can gain deeper understanding political, economic, religious, cultural aspects dynasty. The oracle bone drill chisel (Obdc) is considered an essential non-textual material. segmentation Obdc assists archaeologists determine approximate age Obs, possesses considerable research value. However, breakage thousands years underground buried blurring edges area burned by Obdc, different shapes, inconsistent number have brought challenges to accurate Obdc. In this article, propose group convolutional attention pvt dual-branch network (GCA-PVT-Net) for segmentation. To our knowledge, paper first automatic It hybrid Convolutional neural (CNN) Transformer framework. work offers following contributions: (1) images labeled based on delineation criteria (DC) shapes create dataset. (2) A module (CAM) proposed as both encoder decoder. feature extraction process, effectively integrates global local information, ensures better modeling long-term correlations in while preserving details. (3) channel aggregation (CFAM) designed enhance effective integration features, enabling fusion across various branches at levels. (4) edge deep supervision strategy applied smooth jagged predicted decoder’s end. Extensive experiments dataset show that GCA-PVT-Net outperforms other state-of-the-art (SOTA) methods. comparative experimental results accuracy model reach top 1.

Language: Английский

Citations

1
SwinSAM: Fine-grained polyp segmentation in colonoscopy images via segment anything model integrated with a Swin Transformer decoder DOI
Zhoushan Feng, Yuliang Zhang, Yanhong Chen

et al.

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 100, P. 107055 - 107055

Published: Nov. 15, 2024

Language: Английский

Citations

1
PDCA-Net: Parallel dual-channel attention network for polyp segmentation DOI
Gang Chen, Minmin Zhang,

Junmin Zhu

et al.

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 101, P. 107190 - 107190

Published: Nov. 27, 2024

Language: Английский

Citations

1
MAUNet: Polyp segmentation network based on multiscale feature fusion of attention U‐shaped network structure DOI

Jianwu Long,

Jian Lin,

Jiayin Liu

et al.

International Journal of Imaging Systems and Technology, Journal Year: 2024, Volume and Issue: 34(3)

Published: May 1, 2024

Abstract Colorectal cancer is a prevalent malignant tumor affecting the digestive tract. Although colonoscopy remains most effective method for colon examination, it may occasionally fail to detect polyps. In an effort enhance detection rate of intestinal polyps during colonoscopy, we propose MAUNet, polyp segmentation network based on multi‐scale feature fusion Attention U‐shaped structure. Our model incorporates advanced components, including Receptive Field Block, Reverse and Residual Refinement Module, mirroring analytical process performed by medical imaging professionals. We evaluated performance MAUNet five challenging datasets conducted comparative analysis against state‐of‐the‐art models using six evaluation metrics. The experimental results demonstrate that achieves varying levels improvement across datasets. Particularly Kvasir dataset, Mean Dice IOU metrics reached 91.6% 84.3%, respectively, confirming model's outstanding in segmentation.

Language: Английский

Citations

0