Related Experiment Video
Updated: Nov 12, 2025
09:13
Remote Magnetic Navigation for Accurate, Real-time Catheter Positioning and Ablation in Cardiac Electrophysiology Procedures
Published on: April 21, 2013
28.2K
Automatic lumen detection and magnetic alignment control for magnetic-assisted capsule colonoscope system
Sheng-Yang Yen1, Hao-En Huang1, Gi-Shih Lien2,3
1Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan.
Scientific Reports
|March 20, 2021
Summary
We developed an AI-powered magnetic capsule colonoscope system that improves navigation. This advanced system uses computer vision for accurate lumen detection and alignment, significantly reducing cecal intubation time compared to manual methods.
Area of Science:
- Medical Devices
- Artificial Intelligence
- Gastroenterology
Background:
- Colonoscopy remains a critical diagnostic tool, but traditional methods can be invasive and uncomfortable.
- Capsule endoscopy offers a less invasive alternative, but navigation and control present challenges.
- Advancements in computer vision and AI are enabling more sophisticated endoscopic tools.
Purpose of the Study:
- To develop and evaluate a magnetic-assisted capsule colonoscope system integrated with computer vision for improved navigation.
- To assess the performance of deep learning models for lumen identification and alignment control.
- To compare the cecal intubation time between semi-automated and manual capsule colonoscope navigation.
Main Methods:
- Developed a magnetic-assisted capsule colonoscope system incorporating computer vision-based object detection and an alignment control scheme.
- Trained two convolutional neural network models (A and B) on 9080 endoscopic images for lumen identification.
- Utilized models C and D with a simulated dataset of 8414 images for lumen alignment experiments, evaluating recall, precision, mAP, and F1 score.
Main Results:
- Model D achieved the highest predictive performance with R, P, mAP, and F1 scores of 0.964, 0.961, 0.961, and 0.963, respectively.
- The lumen alignment experiment showed mean yaw and pitch adjustments of 21.70° and 13.78°, with a mean control time of 0.902 seconds.
- Semi-automated navigation reduced average cecal intubation time to 7 min 23.61 s, compared to 9 min 28.41 s for manual navigation.
Conclusions:
- The developed deep learning-based automatic lumen detection model demonstrates high performance in validation metrics.
- The magnetic-assisted capsule colonoscope system with AI integration offers a promising advancement for colonoscopy procedures.
- The system significantly reduces navigation time, potentially improving patient comfort and diagnostic efficiency.
