FEBE-Net: Feature Exploration Attention and Boundary Enhancement Refinement Transformer Network for Bladder Tumor Segmentation DOI Creative Commons

Chao Nie,

Chao Xu,

Zheng-Ping Li

et al.

Mathematics, Journal Year: 2024, Volume and Issue: 12(22), P. 3580 - 3580

Published: Nov. 15, 2024

The automatic and accurate segmentation of bladder tumors is a key step in assisting urologists diagnosis analysis. At present, existing Transformer-based methods have limited ability to restore local detail features insufficient boundary capabilities. We propose FEBE-Net, which aims effectively capture global remote semantic features, preserve more information, provide clearer precise boundaries. Specifically, first, we use PVT v2 backbone learn multi-scale feature representations adapt changes tumor size shape. Secondly, new exploration attention module (FEA) fully explore the potential information shallow extracted by backbone, eliminate noise, supplement missing fine-grained details for subsequent decoding stages. same time, enhancement refinement (BER), generates high-quality clues through detection operators help decoder refine adjust final predicted map. Then, efficient self-attention calibration (ESCD), which, with provided BER module, gradually recovers contextual from high-level after low-level attention. Extensive experiments on cystoscopy dataset BtAMU five colonoscopy datasets shown that FEBE-Net outperforms 11 state-of-the-art (SOTA) networks performance, higher accuracy, stronger robust stability, generalization ability.

Language: Английский

FEBE-Net: Feature Exploration Attention and Boundary Enhancement Refinement Transformer Network for Bladder Tumor Segmentation DOI Creative Commons

Chao Nie,

Chao Xu,

Zheng-Ping Li

et al.

Mathematics, Journal Year: 2024, Volume and Issue: 12(22), P. 3580 - 3580

Published: Nov. 15, 2024

The automatic and accurate segmentation of bladder tumors is a key step in assisting urologists diagnosis analysis. At present, existing Transformer-based methods have limited ability to restore local detail features insufficient boundary capabilities. We propose FEBE-Net, which aims effectively capture global remote semantic features, preserve more information, provide clearer precise boundaries. Specifically, first, we use PVT v2 backbone learn multi-scale feature representations adapt changes tumor size shape. Secondly, new exploration attention module (FEA) fully explore the potential information shallow extracted by backbone, eliminate noise, supplement missing fine-grained details for subsequent decoding stages. same time, enhancement refinement (BER), generates high-quality clues through detection operators help decoder refine adjust final predicted map. Then, efficient self-attention calibration (ESCD), which, with provided BER module, gradually recovers contextual from high-level after low-level attention. Extensive experiments on cystoscopy dataset BtAMU five colonoscopy datasets shown that FEBE-Net outperforms 11 state-of-the-art (SOTA) networks performance, higher accuracy, stronger robust stability, generalization ability.

Language: Английский

Citations

0