PSwinUNet: Bridging Local and Global Contexts for Accurate Medical Image Segmentation with Semi-Supervised Learning

Zhixuan Zhao; Bailin Liu; Hongpei Zhang; Chentao Qian; Yijian Zhang

doi:10.2478/ijanmc-2025-0024

.blurhash-client-img { display: none !important; }

PSwinUNet: Bridging Local and Global Contexts for Accurate Medical Image Segmentation with Semi-Supervised Learning

International Journal of Advanced Network, Monitoring and Controls

Volume 10 (2025): Issue 3 (September 2025)

By: Zhixuan Zhao, Bailin Liu, Hongpei Zhang, Chentao Qian and Yijian Zhang

Open Access

|Sep 2025

Figures & Tables

(a): We present the architecture of our PSwinUNet, a hybrid CNN-Transformer architecture. (b): TheSwin-Transformer block (c): The SCPSA module enhances cross-dimensional interactions from both channel and spatial aspects, compensating.

Lustration of an efficient batch computation approach for self-attention in shifted window configuration.

The visual segmentation results of various methods in the semisupervised experiment with 1/2 labeled data amounts on the BUSI, DRIVE, and CVC-ClinicDB datasets are displayed in Fig. 3. Notably, our PSwinUNet demonstrates relatively superior visualizations compared to other methods.

TABLE I.

THE QUANTITATIVE RESULTS FOR DSC OF VARIOUS METHODS ON 1/8, 1/4, 1/2, AND FULL LABELED DATA AMOUNTS ARE PRESENTED

Method	Dataset	Labeled Data Amount
Method	Dataset	1/8	1/4	1/2	full
UNet	*BUSI DRIVE*	*0.612* *0.686*	*0.744* *0.753*	*0.791* *0.801*	*0.836* *0.844*
UNet	*CVC-ClinicDB*	*0.570*	*0.691*	*0.731*	*0.804*
UNet++	*BUSI DRIVE*	*0.597* *0.691*	*0.783* *0.744*	*0.832* *0.795*	*0.893* *0.852*
UNet++	*CVC-ClinicDB*	*0.602*	*0.686*	*0.751*	*0.804*
SwinUNet	*BUSI DRIVE*	*0.688* *0.723*	*0.761* *0.772*	*0.869* *0.815*	*0.952* *0.846*
SwinUNet	*CVC-ClinicDB*	*0.711*	*0.745*	*0.815*	*0.883*
TransUNe t	*BUSI DRIVE*	*0.712* *0.738*	*0.795* *0.791*	*0.891* *0.842*	*0.943* *0.894*
TransUNe t	*CVC-ClinicDB*	*0.732*	*0.791*	*0.844*	*0.934*
UNet3+	*BUSI DRIVE*	*0.662* *0.663*	*0.736* *0.720*	*0.806* *0.791*	*0.897* *0.862*
UNet3+	*CVC-ClinicDB*	*0.705*	*0.818*	*0.864*	*0.907*
PSwinUN et	*BUSI DRIVE*	*0.781* *0.740*	*0.813* *0.786*	*0.896* *0.872*	*0.960* *0.896*
PSwinUN et	*CVC-ClinicDB*	*0.750*	*0.802*	*0.874*	*0.939*

TABLE II.

ABLATION STUDY ON THE IMPACT OF PSA CONNECTION MODES ON BUSI DATASET.

Connection Modes	Data Amount
Connection Modes	1/8	1/4	1/2	full
None	*0.692*	*0.733*	*0.791*	*0.855*
Channel-only branch	*0.702*	*0.782*	*0.858*	*0.891*
Spatial-only branch	*0.723*	*0.757*	*0.844*	*0.889*
parallel layout	*0.781*	*0.813*	*0.896*	*0.960*
sequential layout	*0.744*	*0.822*	*0.868*	*0.915*

TABLE III.

ABLATION STUDY ON THE IMPACT OF SWIN-TRANSFORMER BLOCK ON BUSI DATASET.

Method	Data Amount
Method	1/8	1/4	1/2	full
Baseline	*0.612*	*0.744*	*0.791*	*0.836*
PSwinUNet(w/o)	*0.662*	*0.796*	*0.845*	*0.887*
PSwinUNet(Ours)	*0.781*	*0.813*	*0.896*	*0.960*

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/ijanmc-2025-0024 | Journal eISSN: 2470-8038

Journal RSS Feed

Language: English

Page range: 33 - 42

Published on: Sep 30, 2025

Published by: Xi’an Technological University

In partnership with: Paradigm Publishing Services

Publication frequency: 4 issues per year

Keywords:

Machine Learning,

PSwinUNet,

CNN-Transformer,

Semi-supervised learning

Related subjects:

Computer sciences,

Computer sciences, other

© 2025 Zhixuan Zhao, Bailin Liu, Hongpei Zhang, Chentao Qian, Yijian Zhang, published by Xi’an Technological University
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Volume 10 (2025): Issue 3 (September 2025)

PSwinUNet: Bridging Local and Global Contexts for Accurate Medical Image Segmentation with Semi-Supervised Learning

Figures & Tables

Figure 1.

Figure 2.

Figure 3.

TABLE I.

TABLE II.

TABLE III.

Paradigm

My account