Related Experiment Video
Updated: Jun 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
LIT-Unet: a lightweight and effective model for medical image segmentation
Ru Wang1,2, Qiqi Kou3, Lina Dou4
1School of Information and Control Engineering, China University of Mining and Technology, Xuzhou, 221116, China.
Radiological Physics and Technology
|September 20, 2024
Summary
A new lightweight network, LIT-Unet, offers efficient and accurate automatic medical image segmentation. This advanced model aids doctors in diagnosis and treatment planning by improving segmentation accuracy on key datasets.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate medical image segmentation is crucial for clinical diagnosis and treatment planning.
- Existing segmentation models can be computationally intensive or lack sufficient accuracy.
- There is a need for simple, efficient, and accurate automatic segmentation models.
Purpose of the Study:
- To design a simple and efficient automatic segmentation model for medical images.
- To facilitate more accurate diagnosis and treatment planning for medical professionals.
- To evaluate the proposed model's performance against state-of-the-art methods.
Main Methods:
- Proposed a hybrid lightweight network named LIT-Unet, featuring a symmetric encoder-decoder U-shaped architecture.
- Utilized the Synapse multi-organ segmentation dataset and the Automated Cardiac Diagnosis Challenge (ACDC) dataset for performance evaluation.
- Employed Dice Similarity Coefficient (DSC) and 95% Hausdorff Distance (HD95) as evaluation metrics.
Main Results:
- On the Synapse dataset, LIT-Unet achieved a DSC of 80.40%, outperforming TransUnet by 3.8%, with an HD95 of 20.67%.
- On the ACDC dataset, LIT-Unet obtained an optimal average DSC of 91.84% compared to other networks.
- Deformable Token Merging (DTM) improved DSC by 1.62% compared to patch expanding, demonstrating the model's effectiveness.
Conclusions:
- The proposed hierarchical LIT-Unet demonstrates significant accuracy in medical image segmentation.
- The model's lightweight nature and effectiveness were confirmed through ablation experiments.
- LIT-Unet is expected to provide a reliable basis for clinical diagnosis and treatment.
