Related Experiment Video
Updated: Jan 12, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Lightweight deep training network for lymph nodes segmentation from head and neck CT images
Fan Lu1, Xiao-Long Li2, Binbin Jiang3
1School of Future Science and Engineering, Soochow University, Suzhou, Jiangsu, China.
Medical Physics
|November 4, 2025
Summary
We developed LNSNet, an efficient deep learning model for segmenting head and neck lymph nodes (LNs) in CT scans. This lightweight network reduces computational complexity while improving segmentation accuracy for better disease diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate lymph node (LN) segmentation is crucial for diagnosing head and neck diseases.
- Manual LN identification in CT images is challenging due to size variation, complex shapes, and blurred boundaries.
- Existing 3D convolutional methods for LN segmentation have high computational costs.
Purpose of the Study:
- To develop an efficient and lightweight volumetric convolutional neural network (LNSNet) for head and neck lymph node segmentation.
- To address the limitations of high computational complexity in current segmentation methods.
Main Methods:
- LNSNet utilizes a 3D Volume Block combining Volumetric Partial Convolution (VPConv) and point-wise convolution to reduce computational load and parameters.
- A Lightweight Boundary Enhancement Module (LBEM) and depthwise separable convolution are incorporated to boost segmentation accuracy.
Main Results:
- The model was evaluated on 678 3D LNs from 123 head and neck cancer patients.
- LNSNet demonstrated fewer parameters and lower computational complexity compared to state-of-the-art models.
- Achieved a Dice Similarity Coefficient (DSC) of 73.81%, with Average Surface Distance (ASD) of 0.92 mm and Hausdorff Distance 95th (HD95) of 2.52 mm.
Conclusions:
- LNSNet enhances computational efficiency and robustness in lymph node segmentation.
- The reduced parameter count and complexity make LNSNet suitable for practical clinical applications.
