Related Experiment Video
Updated: Jul 10, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Research on Pectoral Muscle Segmentation Algorithm of CT Image Based on Deep Learning
Ying Wang1, Ping Zhou1, Xingqun Zhao1
1School of Biological Science and Medical Engineering, Southeast University, Nanjing, China.
Studies in Health Technology and Informatics
|November 26, 2023
Summary
This study introduces a deep learning algorithm for segmenting pectoral muscles in CT scans to assess Chronic Obstructive Pulmonary Disease (COPD) severity. The U-Net based method achieves high accuracy, aiding in better diagnosis and management of COPD patients.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Pulmonology
Background:
- Chronic Obstructive Pulmonary Disease (COPD) poses a significant global health threat, with China experiencing the highest mortality rates and under-diagnosis.
- Lung function tests are standard for COPD diagnosis, but CT imaging offers auxiliary diagnostic value by assessing pectoral muscle area.
- Reduced pectoral muscle area in COPD patients correlates with more severe expiratory airflow obstruction, highlighting the need for accurate muscle segmentation.
Purpose of the Study:
- To develop and evaluate a deep learning-based algorithm for accurate pectoral muscle segmentation in CT images.
- To improve the auxiliary diagnosis of COPD severity by quantifying pectoral muscle area.
- To leverage U-Net and its variant U-Net++ for efficient medical image segmentation, even with limited labeled data.
Main Methods:
- A deep learning approach utilizing U-Net and U-Net++ architectures for pectoral muscle segmentation in CT images.
- Implementation of data augmentation techniques to enhance learning from limited labeled medical datasets.
- Validation of the algorithm using a dataset from Jiangsu Province Hospital.
Main Results:
- The proposed algorithm achieved an average Dice coefficient exceeding 94% and an average accuracy rate of 91%.
- Demonstrated accurate segmentation of pectoral muscles in CT images, indicating robust performance.
- The U-Net based approach effectively learned from limited data, proving suitable for medical image segmentation tasks.
Conclusions:
- The developed U-Net based algorithm accurately segments pectoral muscles in CT images, offering a valuable tool for COPD severity assessment.
- This method holds significant potential for improving the auxiliary diagnosis and management of COPD.
- Deep learning segmentation shows promise for extracting detailed information from medical images, advancing diagnostic capabilities.
