Related Experiment Video
Updated: Sep 17, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Catheter detection and segmentation in X-ray images via multi-task learning.
Lin Xi1, Yingliang Ma2,3, Ethan Koland4
1School of Computing Sciences, University of East Anglia, Norwich, NR4 7TJ, UK. l.xi@uea.ac.uk.
Summary
This study introduces a deep learning model for real-time detection and segmentation of surgical catheters in X-ray images. The method enhances surgical guidance by accurately locating devices during minimally invasive heart procedures.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Minimally invasive heart surgeries require precise guidance.
- Surgical devices like catheters need accurate localization in X-ray fluoroscopic images.
- Automated detection and segmentation can improve image guidance.
Purpose of the Study:
- To develop a real-time convolutional neural network (CNN) model for surgical device localization and segmentation.
- To integrate electrode detection and catheter segmentation into an end-to-end deep learning framework.
- To enhance image guidance in minimally invasive cardiac procedures.
Main Methods:
- A CNN model with a ResNet architecture and multiple prediction heads was developed.
- A multi-task learning strategy was employed for simultaneous electrode detection and catheter segmentation.
- A novel multi-level dynamic resource prioritization method was introduced to balance task difficulty during training.
Main Results:
- The method was validated on public and private datasets for single-task and multi-task scenarios.
- Significant improvements were demonstrated compared to state-of-the-art methods.
- Achieved a mean J of 64.37/63.97 for detection and segmentation multi-task on validation and test sets.
Conclusions:
- The proposed approach offers a balance between accuracy and efficiency.
- The model is suitable for real-time surgical guidance applications.
- Automated detection and segmentation of surgical devices can enhance minimally invasive procedures.
