Related Experiment Video
Updated: Jul 25, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.7K
A Lightweight Convolutional Neural Network Based on Dynamic Level-Set Loss Function 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
|June 29, 2023
Summary
A new lightweight model, Dynamic Level-set Net (DLS-Net), offers effective spine MR image segmentation with fewer parameters. This approach enhances computer-aided diagnosis (CAD) for spine disorders, improving diagnostic accuracy and applicability.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Spine Disorders
Background:
- Spine MR image segmentation is crucial for computer-aided diagnosis (CAD) of spinal disorders.
- Current convolutional neural networks offer effective segmentation but demand high computational resources.
- Developing lightweight models is essential for wider clinical application and efficiency.
Purpose of the Study:
- To design a lightweight model for high-performance spine MR image segmentation.
- To utilize a dynamic level-set loss function for improved segmentation accuracy.
- To enhance the efficiency of CAD algorithms for spine disorder diagnosis.
Main Methods:
- A novel Dynamic Level-set Net (DLS-Net) was developed and evaluated.
- DLS-Net was compared against mainstream and lightweight segmentation models using five-fold cross-validation.
- A CAD algorithm for lumbar disc assessment was developed using DLS-Net segmentation results.
Main Results:
- DLS-Net achieved comparable segmentation accuracy to U-net++ with significantly fewer parameters (1.48%).
- Segmentation results showed no significant difference compared to manual labels for discs and vertebrae.
- The CAD algorithm utilizing DLS-Net segmentation demonstrated higher diagnostic accuracy (87.47%) than using non-cropped images (61.82%).
Conclusions:
- The proposed DLS-Net offers an efficient and accurate solution for spine MR image segmentation.
- Its lightweight nature and high performance facilitate wider application in CAD systems.
- DLS-Net contributes to improved accuracy in diagnosing spinal conditions like disc degeneration and herniation.
