Related Experiment Video
Updated: May 11, 2026
09:41
A Pipeline for 3D Multimodality Image Integration and Computer-assisted Planning in Epilepsy Surgery
Published on: May 20, 2016
12.0K
LFE-UNet: A Lightweight Full-Encoder U-shaped Network for Efficient Semantic Segmentation in Medical Imaging
Qinghua Zhang1,2, Yulei Hou3, Changchun He1
1Department of Neurosurgery, Huazhong University of Science and Technology Union Shenzhen Hospital, Shenzhen 518052, China.
Current Medical Imaging
|May 12, 2025
Summary
This study introduces LFE-UNet, a lightweight network for medical image segmentation that achieves high accuracy with fewer parameters. It demonstrates efficient feature extraction and robustness to noise, optimizing the balance between performance and computational cost.
Area of Science:
- Medical Image Analysis
- Deep Learning Architectures
- Computer Vision
Background:
- Semantic segmentation is crucial for medical imaging, but U-Net variants increase computational demands.
- High parameter counts in advanced segmentation models necessitate powerful hardware for training.
Purpose of the Study:
- To develop a lightweight U-Net (LFE-UNet) that balances segmentation accuracy and parameter efficiency.
- To fully leverage encoder feature extraction capabilities in a resource-constrained model.
Main Methods:
- Proposed a lightweight full-encoder U-shaped network (LFE-UNet) with full-encoder skip connections.
- Reduced basic channels to 8 for enhanced architectural efficiency.
- Integrated LFE-UNet with ResNet34 for performance evaluation.
Main Results:
- Achieved a Dice score of 0.97385 on the ISBI LiTS 2017 liver dataset.
- Obtained high scores on the BraTS 2018 brain tumor dataset (average 0.87510, WT 0.93759, TC 0.87301, ET 0.81469).
- Analyzed the impact of channel/layer counts, noise, and loss functions on parameter efficiency and accuracy.
Conclusions:
- LFE-UNet demonstrates that high segmentation accuracy is achievable with significantly fewer parameters.
- Emphasizes the importance of full-scale encoder feature utilization.
- Highlights the influence of loss function choice and noise on segmentation performance.
