Related Experiment Video
Updated: Jun 4, 2025
07:46
Author Spotlight: Revolutionizing Remote Surgery with Augmented Reality and Robotics for Enhanced Precision and Accessibility
Published on: August 9, 2024
648
Automated measurement of cardiothoracic ratio based on semantic segmentation integration model using deep learning
Jiajun Feng1, Yuqian Huang2, Zhenbin Hu3
1Department of Medical Imaging, Guangzhou First People's Hospital, School of Medicine, South China University of Technology, No.1 Panfu Road, Yuexiu District, Guangzhou, 510030, Guangdong, China.
Medical & Biological Engineering & Computing
|December 21, 2024
Summary
A new semantic segmentation model accurately measures cardiothoracic ratio (CTR) and detects heart enlargement on chest X-rays. This AI tool is faster and more consistent than manual measurements, aiding radiologists.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Cardiothoracic ratio (CTR) and heart enlargement are key indicators of cardiac health.
- Manual measurement of CTR on chest radiographs is time-consuming and prone to variability.
- Automated methods are needed to improve efficiency and consistency in cardiac assessment.
Purpose of the Study:
- To develop and evaluate a semantic segmentation model for automated CTR measurement and heart enlargement detection.
- To compare the model's performance against reference standards and manual radiologist measurements.
- To assess the model's diagnostic efficacy and speed in clinical scenarios.
Main Methods:
- Developed a semantic segmentation model using a combined dataset of 1406 chest radiographs.
- Employed a soft voting integration method to enhance segmentation accuracy of heart and lungs.
- Utilized Bland-Altman analysis, Pearson's correlation, and Wilcoxon signed-rank test for consistency and comparison.
- Evaluated diagnostic performance using accuracy, sensitivity, specificity, and Area Under the Curve (AUC).
Main Results:
- The integrated model showed strong correlation (r=0.98) and consistency with reference standards for CTR.
- No significant difference in CTR measurements between the model and reference standards across various patient groups.
- Achieved high diagnostic performance for heart enlargement: 96.0% accuracy, 79.5% sensitivity, 99.1% specificity, and 0.988 AUC.
- Automated measurements were significantly faster (approx. 2 seconds) than manual radiologist calculations (approx. 25.75 seconds).
Conclusions:
- The semantic segmentation integration model provides an effective, fast, and accurate method for CTR measurement and heart enlargement detection.
- This automated approach enhances the consistency of CTR measurements and reduces radiologist workload.
- The model demonstrates significant potential for improving diagnostic efficiency and accuracy in cardiovascular imaging.
