Related Experiment Video
Updated: Nov 23, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
Fully automatic deep learning trained on limited data for carotid artery segmentation from large image volumes
Tianshu Zhou1, Tao Tan2, Xiaoyan Pan1
1Engineering Research Center of EMR and Intelligent Expert System, Ministry of Education, Key Laboratory for Biomedical Engineering of Ministry of Education, College of Biomedical Engineering and Instrument Science, Zhejiang University, Hangzhou, China.
Quantitative Imaging in Medicine and Surgery
|January 4, 2021
Summary
A novel deep learning framework, CarotidNet, automatically segments carotid bifurcations in 3D CTA images. This method shows promise for stroke risk assessment by enabling accurate carotid stenosis quantification.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Imaging
Background:
- Carotid artery disease is a significant risk factor for stroke.
- Accurate segmentation of carotid bifurcations in computed tomography angiography (CTA) is crucial for stenosis quantification and stroke risk assessment.
- Manual segmentation is time-consuming and prone to inter-observer variability.
Purpose of the Study:
- To develop a 3D deep learning framework (CarotidNet) for automated carotid bifurcation segmentation in CTA images.
- To enable precise quantification of carotid stenosis.
- To facilitate improved stroke risk assessment.
Main Methods:
- A two-stage cascade network comprising localization and segmentation phases was designed.
- The framework utilized a 3D U-Net architecture with residual connections and deep supervision.
- Dilated convolutions and a hybrid objective function were incorporated to enhance contextual information capture and address foreground-background imbalance.
Main Results:
- The CarotidNet framework achieved a Dice similarity coefficient of 82.3% on test cases.
- The method successfully segmented tiny carotid bifurcation lumens from large backgrounds without manual intervention.
- The network was trained on 15 cases and evaluated on 41 cases from the MICCAI Challenge 2009 dataset.
Conclusions:
- A novel deep learning-based method for automatic carotid bifurcation segmentation in 3D CTA images was successfully developed.
- This represents the first application of deep learning for 3D carotid bifurcation segmentation in CTA.
- The findings suggest deep learning is a highly promising approach for automated extraction of carotid bifurcation lumens, aiding in clinical applications.
