Related Experiment Video
Updated: Aug 1, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
458
LSW-Net: Lightweight Deep Neural Network Based on Small-World properties for Spine MR Image Segmentation
Siyuan He1, Qi Li1,2, Xianda Li1
1School of Computer Science and Technology, Changchun University of Science and Technology, Changchun, China.
Journal of Magnetic Resonance Imaging : JMRI
|April 29, 2023
Summary
A new lightweight neural network (LSW-Net) effectively segments spinal MR images with high accuracy and fewer parameters than U-net. This model is suitable for low-power embedded devices, enabling wider clinical application.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Accurate segmentation of spinal tissues in MR images is crucial for automated analysis.
- Deep neural networks offer efficiency but demand significant computational resources.
Purpose of the Study:
- To develop a lightweight neural network, LSW-Net, leveraging small-world properties.
- To enable efficient spinal MR image segmentation on devices with limited computing power.
Main Methods:
- LSW-Net was developed and trained on 2948 MR images from 386 subjects.
- Performance was compared against mainstream and lightweight models using radiologist segmentations.
- The model was deployed on an NVIDIA Jetson nano for embedded device evaluation.
Main Results:
- LSW-Net achieved comparable accuracy to U-net with 98.5% fewer parameters.
- Segmentation accuracy metrics (DSC, AUC) were similar across two independent datasets.
- No significant difference was found in segmented pixel counts on the embedded device compared to manual segmentation.
Conclusions:
- LSW-Net provides a high-accuracy, parameter-efficient solution for spinal MR image segmentation.
- Its suitability for embedded devices facilitates broader accessibility and application in clinical settings.
