Related Experiment Video
Updated: Aug 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
MSU-Net: Multi-scale Sensitive U-Net based on pixel-edge-region level collaborative loss for nasopharyngeal MRI
Yuanquan Hao1, Huiyan Jiang2, Zhaoshuo Diao3
1Northeastern University, Shenyang 110819, China.
Computers in Biology and Medicine
|April 28, 2023
Summary
This study introduces a novel multi-scale sensitive U-Net (MSU-Net) for segmenting nasopharyngeal carcinoma (NPC) in radiotherapy. The MSU-Net achieves higher accuracy and effective segmentation performance compared to existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Radiotherapy is a standard treatment for early nasopharyngeal carcinoma (NPC).
- Accurate segmentation of cancerous lesions in the nasopharynx is critical for effective radiotherapy planning.
- Existing U-Net models struggle with the diverse anatomical variations in nasopharyngeal structures, necessitating multi-scale information processing.
Purpose of the Study:
- To develop an advanced segmentation model for nasopharyngeal carcinoma (NPC) that addresses the limitations of current methods in handling anatomical variability.
- To improve the accuracy and efficiency of automatic lesion segmentation in radiotherapy for NPC patients.
Main Methods:
- A novel multi-scale sensitive U-Net (MSU-Net) was developed, incorporating specialized modules for feature extraction.
- A spatial continuity information extraction module (SCIEM) was designed to leverage contextual information from adjacent slices for detecting small lesions.
- A multi-scale semantic feature extraction module (MSFEM) was implemented to capture features across different receptive fields.
- A collaborative loss function (LCo-PER) was introduced to optimize segmentation across various lesion sizes.
Main Results:
- The proposed MSU-Net achieved a global Dice score of 84.50%, Precision of 97.48%, Recall of 84.33%, and IOU of 82.41% on the testing dataset.
- The method demonstrated superior performance compared to state-of-the-art techniques in NPC segmentation.
- The novel feature fusion modules and loss function effectively handled multi-scale information and anatomical variations.
Conclusions:
- The developed MSU-Net provides a significant advancement in automatic segmentation for nasopharyngeal carcinoma (NPC) in radiotherapy.
- The model's ability to process multi-scale information and utilize spatial continuity enhances segmentation accuracy, particularly for varying lesion sizes.
- This approach offers a promising tool for improving radiotherapy planning and patient outcomes in NPC treatment.
