Related Experiment Video
Updated: Mar 17, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
852
A parallel UNet integrating KAN and mamba for medical image segmentation
Jiyuan Liu1, Jiabao Wu1, Liming Xu1,2
1School of Computer Science, China West Normal University, Nanchong, 637009, China.
Scientific Reports
|March 16, 2026
Summary
KMP-UNet, a novel medical image segmentation model, efficiently captures long-range context using Mamba and nonlinear features with Kolmogorov-Arnold Networks (KANs). This compact framework achieves high accuracy on benchmark datasets.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Medical image segmentation is crucial for clinical diagnosis, aiding in lesion and organ boundary delineation.
- Current UNet models, dominated by Convolutional Neural Networks (CNNs), struggle with long-range contextual information.
- Transformer-based methods offer context but often incur high computational costs due to quadratic attention mechanisms.
Purpose of the Study:
- To develop an efficient and effective medical image segmentation framework that overcomes the limitations of existing models.
- To introduce KMP-UNet, a novel U-shaped architecture integrating Mamba and Kolmogorov-Arnold Networks (KANs).
- To enhance feature representation and contextual modeling for improved segmentation performance.
Main Methods:
- Proposed KMP-UNet, a parallel U-shaped network combining a Mamba-based branch for linear-complexity contextual modeling and a Kolmogorov-Arnold Network (KAN) branch for nonlinear feature representation.
- Introduced a task-oriented fusion block and a skip refinement module to leverage hierarchical encoder-decoder features.
- Implemented a compact model with approximately 1.0 million parameters.
Main Results:
- Evaluated KMP-UNet on four public datasets: ISIC2017, ISIC2018, CVC-ClinicDB, and BUSI.
- Achieved competitive segmentation performance, including 0.9038 Dice Similarity Coefficient (DSC) and 0.9600 accuracy on the ISIC2018 dataset.
- Demonstrated the effectiveness of the proposed fusion and refinement modules through extensive ablation studies.
Conclusions:
- KMP-UNet presents an efficient and powerful approach for medical image segmentation.
- The integration of Mamba and KANs offers a promising direction for balancing contextual understanding and nonlinear feature extraction.
- The compact design and strong performance highlight KMP-UNet's potential for clinical applications.
