Aug 2026· arXiv.org· Vol abs/2608.22281· 0 citations· 20 references
Computer ScienceEngineering
TL;DR
This paper proposes a lightweight hybrid architecture built upon the Swin-UNet framework that integrates a parallel CNN encoder to complement the shallow layer reasoning of Swin Transformers with local texture features and introduces cross-layer skip connections that explicitly propagate shallow CNN features into the decoder.
Abstract
Medical image segmentation requires high accuracy and robustness, yet practical commercial deployment also demands privacy preservation and computational efficiency. In this context, the U-Net architecture, which can be inherently decoupled into independent encoder and decoder components, serves as a natural commercial choice. However, pure Transformer-based variants like Swin-UNet often suffer from insufficient local detail capture and limited interpretability. In this paper, we propose a lightweight hybrid architecture built upon the Swin-UNet framework. Our model integrates a parallel CNN encoder to complement the shallow layer reasoning of Swin Transformers with local texture features. To bridge the semantic gap and enhance fine-grained spatial detail recovery, we design an asymmetric feature fusion strategy and introduce cross-layer skip (XSkip) connections that explicitly propagate shallow CNN features into the decoder. We further incorporate novel loss functions and an auxiliary supervision head (Aux-Head) to strengthen training stability, boundary delineation, and intermediate feature interpretability. Extensive experiments on the Synapse multi-organ segmentation dataset demonstrate that our approach achieves state-of-the-art competitive Dice scores and Hausdorff distances, offering an accurate, efficient, and interpretable solution for clinical deployment.
Robust medical image classification remains a persistent challenge due to heterogeneous acquisition protocols, label noise, and subtle inter-class variability. In chest radiography, these difficulties are compounded by shortcut learning from non-anatomical artifacts and projection variability, which severely undermine...
Saeed Al Gharib, F. Dornaika, J. Charafeddine et al.· Journal of imaging informati...· 0 citations
Various task-specific medical segmentation models have demonstrated strong performance, their heavy reliance on extensive annotations and highly specialized architectures limits their versatility, scalability, and clinical applicability. Recently, adapting vision foundation models, particularly the segment anything mod...
Accurate nuclei segmentation in histopathology images is crucial for downstream biological analysis. While Transformer based models excel at capturing global contexts, they struggle with blurred boundaries and structural adhesion in dense cellular regions. To address this, we propose Swin-UNet-EARM, introducing an Edge...
Wu-Hao Zhong, Song Liu, Xin-Rui Wan et al.· Conference on Advanced Algor...· 0 citations
Accurate dense prediction in medical images remains challenging because anatomical targets often exhibit large-scale variation, low-contrast boundaries, heterogeneous textures, and fine structural details that are easily degraded during feature extraction. Although CNNs provide strong local inductive biases and Transfo...
Lan-Xiang Ma, Zong-Jian Yang, Jing-Hua Zhu et al.· The Visual Computer· 0 citations
Accurate RGB medical segmentation must combine high overlap, reliable boundaries, low computational cost, and robustness to color and illumination shifts. We present CAPB-Net, a lightweight color-adaptive boundary network that places an image-adaptive color polynomial (IACP) front-end before a compact U-shaped encoder...
Shi-Cheng Zhou, Alan Wee-Chung Liew· International Conference on...· 0 citations
A novel HCSA-RMSDC U-Net is designed in this work, which is characterized by an asymmetric architecture with a computationally heavy encoder and a lightweight decoder that supports the complementary roles of the HCSA and RMSDC mechanisms in enhancing feature representation and multi-scale contextual modeling.
Xiao-Shuai Cao, Xi-Ge Song, Jin-Can Zhang et al.· Frontiers in Neuroscience· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.