Related Experiment Video
Updated: May 15, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Precise segmentation of multiple organs in CT volumes using learning-based approach and information theory
Chao Lu1, Yefeng Zheng, Neil Birkbeck
1Image Analytics and Informatics, Siemens Corporate Research, Princeton, New Jersey, USA.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|January 5, 2013
Summary
This study introduces a new method for automatically segmenting pelvic organs like the prostate and bladder in CT scans. The approach enhances accuracy and is significantly faster than existing techniques.
Area of Science:
- Medical Imaging
- Computer Vision
- Computational Anatomy
Background:
- Accurate segmentation of pelvic organs (prostate, bladder, rectum) is crucial for cancer diagnosis and treatment planning.
- Existing segmentation methods struggle with diverse CT imaging protocols and patient variability.
- Challenges include varying resolutions, contrast agents, and the presence of medical implants.
Purpose of the Study:
- To develop a novel, accurate, and efficient method for automatic pelvic organ segmentation in 3D CT volumes.
- To address the challenges posed by diverse imaging protocols and patient data.
- To improve segmentation performance through the integration of information theory.
Main Methods:
- Incorporation of information theory into a learning-based segmentation framework.
- Application of marginal space learning (MSL) for organ localization in diverse CT volumes.
- Utilizing steerable features for robust boundary detection in heterogeneous textures.
- Employing Jensen-Shannon divergence within the boundary inference process for improved mesh fitting.
Main Results:
- The proposed method achieves excellent segmentation accuracy on a challenging dataset of 188 diverse CT volumes.
- The approach demonstrates a significant speed improvement, running approximately 80 times faster than previous state-of-the-art methods.
- Successfully handles variations in scanning protocols, contrast, and implants.
Conclusions:
- The novel information-theoretic, learning-based approach provides accurate and efficient pelvic organ segmentation.
- This method offers a valuable tool for computer-aided diagnosis, treatment planning, and image-guided radiotherapy.
- Potential to enhance clinical workflows in managing pelvic cancers.
