Related Experiment Video
Updated: May 24, 2026
Proton Therapy Delivery and Its Clinical Application in Select Solid Tumor Malignancies
Published on: February 6, 2019
Learning image context for segmentation of the prostate in CT-guided radiotherapy
Wei Li1, Shu Liao, Qianjin Feng
1Biomedical Engineering College, Southern Medical University, Guangzhou, People's Republic of China. IDEA Lab, Department of Radiology and BRIC, University of North Carolina at Chapel Hill, 130 Mason Farm Road, Chapel Hill, NC 27599-7513, USA. shibazislw@gmail.com
Accurate prostate segmentation in CT scans is crucial for radiotherapy. This study introduces a patient-specific online learning method using location-adaptive classifiers to overcome challenges like low contrast and motion, achieving precise segmentation.
Area of Science:
- Medical Imaging
- Radiotherapy
- Computational Anatomy
Background:
- Accurate prostate segmentation in CT images is vital for effective external beam radiotherapy for prostate cancer.
- Challenges include low image contrast, prostate motion, and variations in surrounding organs (bladder, rectum).
Purpose of the Study:
- To present an online-learning, patient-specific classification method for precise prostate segmentation in CT images.
- To address challenges of low contrast, motion, and anatomical variations using location-adaptive image context.
Main Methods:
- Developed a method using two sets of location-adaptive classifiers along coordinate directions.
- Classifiers are trained with planning and previous treatment images for patient-specific segmentation.
- Employs recursively trained sub-classifiers using static appearance and iterative context features at multiple scales.
Main Results:
- Evaluated on 161 images from 11 patients with daily 3D CT scans.
- Achieved a mean Dice value of 0.908.
- Reported a mean ± SD of average surface distance of 1.40 ± 0.57 mm, outperforming other methods.
Conclusions:
- The proposed online-learning, location-adaptive method enables precise prostate segmentation in CT images.
- This approach effectively handles common segmentation challenges in radiotherapy planning.
- Demonstrated superior segmentation accuracy compared to existing methods.
