Related Experiment Video
Updated: Apr 21, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
986
Interactive prostate segmentation using atlas-guided semi-supervised learning and adaptive feature selection
Sang Hyun Park1, Yaozong Gao2, Yinghuan Shi3
1Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina 27599.
Medical Physics
|November 6, 2014
Summary
This study introduces an interactive prostate segmentation method that significantly improves accuracy for radiation therapy. The new technique reduces manual editing time and variability among clinicians.
Area of Science:
- Medical Imaging
- Computational Anatomy
- Radiation Oncology
Background:
- Accurate prostate segmentation is crucial for effective prostate cancer radiation therapy.
- Manual segmentation of 3D CT images is time-consuming and prone to significant inter- and intra-observer variability.
- Existing automated methods often require extensive manual correction due to performance limitations.
Purpose of the Study:
- To develop a novel interactive segmentation method for prostate cancer radiation therapy.
- To enable flexible and rapid correction of segmentations using minimal user input (scribbles or dots).
- To provide a robust tool that can sequentially refine segmentations from any automated or interactive method.
Main Methods:
- The method formulates segmentation editing as a semisupervised learning problem, integrating user interactions with training data.
- It locally searches for appropriate training labels near user inputs and estimates confident prostate/background voxels.
- Location-adaptive features are extracted using regression forests, and manifold regularization is applied to predict labels for uncertain voxels.
Main Results:
- The interactive method improved automatic segmentation accuracy (Dice similarity coefficient) from 0.78 to 0.865-0.872 on 30 challenging CT images.
- Each editing interaction took less than 3 seconds, demonstrating high efficiency.
- The method showed robust and consistent editing results across different user interactions, outperforming other approaches.
Conclusions:
- The proposed interactive segmentation method delivers robust results with minimal user input, effectively addressing segmentation errors.
- It leverages location-adaptive features and manifold regularization for improved accuracy and consistency.
- This approach is expected to significantly reduce manual editing workload and inter-/intra-observer variability in clinical practice.
