Related Experiment Video
Updated: Jul 23, 2026
06:54
MR Molecular Imaging of Prostate Cancer with a Small Molecular CLT1 Peptide Targeted Contrast Agent
Published on: September 3, 2013
11.8K
LM-UNet: Lightweight Mamba-UNet Prostate MRI image segmentation network
Kuncai Xu1, Shuai Zhou1, Yan Chen1
1College of Intelligent Engineering, Guiyang Institute of Information Science and Technology, Guiyang, Guizhou, PR China.
Plos One
|March 23, 2026
Summary
A new Lightweight Mamba-UNet (LM-UNet) improves prostate MRI segmentation accuracy by using parallel vision mamba and multi-level skip connections. This method enhances lesion margin segmentation and reduces computational demands compared to traditional networks.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate prostate magnetic resonance imaging (MRI) segmentation is crucial for patient care and treatment.
- Traditional UNet networks struggle with fuzzy boundaries and low contrast in prostate MRI, limiting segmentation accuracy.
Purpose of the Study:
- To develop a Lightweight Mamba-UNet (LM-UNet) for enhanced prostate MRI segmentation.
- To improve segmentation accuracy, especially for lesion margins, while reducing model complexity.
Main Methods:
- Proposed LM-UNet integrates parallel vision mamba (PV-Mamba) for long-range feature correlation and efficient multi-scale attention (EMA) for feature aggregation.
- Incorporated edge feature extraction (EFE) and edge feature fusion (EFF) for encoder-level feature fusion.
- Implemented multi-stage and multi-level skip connections (MMSC) for improved encoder-decoder feature fusion.
Main Results:
- LM-UNet demonstrated superior performance on the PROMISE12 dataset compared to seven other segmentation methods.
- Achieved reductions in parameter count and computational memory requirements.
- Showcased precise segmentation of lesion margins in prostate MRI scans.
Conclusions:
- The proposed LM-UNet significantly improves prostate MRI segmentation accuracy.
- LM-UNet offers a computationally efficient and effective solution for clinical applications.
- The novel architecture addresses limitations of traditional UNet models in handling complex image features.
