View in EDS

DEBi‐UKAN: A Dual‐Encoder Bi‐Level Attention U‐KAN Network for Enhanced Cell Image Segmentation.

Saved in:
Bibliographic Details
Title: DEBi‐UKAN: A Dual‐Encoder Bi‐Level Attention U‐KAN Network for Enhanced Cell Image Segmentation.
Authors: Li, Yan1 (AUTHOR) usst_liyan@163.com, Li, Meng1 (AUTHOR), Jia, Chuanlong1 (AUTHOR)
Source: International Journal of Imaging Systems & Technology. May2026, Vol. 36 Issue 3, p1-14. 14p.
Subjects: Cell segmentation, Attention, Artificial neural networks, Diagnostic imaging, Deep learning
Abstract: Accurate cell image segmentation is fundamental to disease research, drug discovery, and quantitative clinical decision‐making. Yet many U‐shaped architectures still struggle when cells present indistinct boundaries, dense spatial arrangements, and pronounced scale variability. Under these conditions, fine structural details are often lost, and feature interactions between shallow and deep layers remain insufficient, which limits the reliability of downstream analysis. To mitigate these issues, this study introduces the Dual‐Encoder Bi‐Level Attention U‐KAN (DEBi‐UKAN), a coordinated segmentation framework for complex cell images. The proposed model employs a dual‐encoder design in which a convolutional branch emphasizes local textures and boundary cues, while a MaxViT branch models long‐range semantic context. Their complementary representations improve the characterization of complex cellular morphology, particularly in regions with weak contrast or severe crowding. Building on this representation, a tokenized KAN component enhances nonlinear function approximation and strengthens deep feature modeling in the high‐level layers. On this basis, a feature fusion module is constructed. It combines direction‐aware attention, residual convolutional refinement, and multibranch channel recalibration to enhance cross‐scale feature alignment, reduce information loss during downsampling, and preserve both edge and interior details. At the bottleneck, bi‐level routing attention is used to selectively emphasize informative regions and strengthen long‐range semantic dependencies. This design promotes more coherent contour delineation and improves robustness under challenging imaging conditions. Extensive quantitative evaluations show that DEBi‐UKAN achieves consistently higher scores than several representative segmentation baselines across multiple metrics, with clear gains in scenarios characterized by low contrast and densely packed cells. These results indicate that DEBi‐UKAN provides a stable and accurate framework for high‐throughput cell image analysis and offers a practical methodological basis for intelligent cell imaging applications. [ABSTRACT FROM AUTHOR]
Copyright of International Journal of Imaging Systems & Technology is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Accurate cell image segmentation is fundamental to disease research, drug discovery, and quantitative clinical decision‐making. Yet many U‐shaped architectures still struggle when cells present indistinct boundaries, dense spatial arrangements, and pronounced scale variability. Under these conditions, fine structural details are often lost, and feature interactions between shallow and deep layers remain insufficient, which limits the reliability of downstream analysis. To mitigate these issues, this study introduces the Dual‐Encoder Bi‐Level Attention U‐KAN (DEBi‐UKAN), a coordinated segmentation framework for complex cell images. The proposed model employs a dual‐encoder design in which a convolutional branch emphasizes local textures and boundary cues, while a MaxViT branch models long‐range semantic context. Their complementary representations improve the characterization of complex cellular morphology, particularly in regions with weak contrast or severe crowding. Building on this representation, a tokenized KAN component enhances nonlinear function approximation and strengthens deep feature modeling in the high‐level layers. On this basis, a feature fusion module is constructed. It combines direction‐aware attention, residual convolutional refinement, and multibranch channel recalibration to enhance cross‐scale feature alignment, reduce information loss during downsampling, and preserve both edge and interior details. At the bottleneck, bi‐level routing attention is used to selectively emphasize informative regions and strengthen long‐range semantic dependencies. This design promotes more coherent contour delineation and improves robustness under challenging imaging conditions. Extensive quantitative evaluations show that DEBi‐UKAN achieves consistently higher scores than several representative segmentation baselines across multiple metrics, with clear gains in scenarios characterized by low contrast and densely packed cells. These results indicate that DEBi‐UKAN provides a stable and accurate framework for high‐throughput cell image analysis and offers a practical methodological basis for intelligent cell imaging applications. [ABSTRACT FROM AUTHOR]
ISSN:08999457
DOI:10.1002/ima.70356