Related Experiment Video
Updated: Mar 8, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
A general framework to learn surrogate relevance criterion for atlas based image segmentation.
1Department of Radiation Oncology, University of California, Los Angeles, CA 90095, USA.
Physics in Medicine and Biology
|August 16, 2016
Summary
This study introduces a new framework for learning image relevance criteria in multi-atlas segmentation. It improves the selection of relevant data for more accurate image segmentation results.
Area of Science:
- Medical Image Analysis
- Computer Vision
Background:
- Multi-atlas based image segmentation faces challenges with large, heterogeneous datasets.
- Current methods use image similarity as a surrogate for geometric relevance, which is often inaccurate.
Purpose of the Study:
- To develop a general framework for learning image-based surrogate relevance criteria.
- To improve the selection of relevant data (fusion set) for more accurate image segmentation.
Main Methods:
- A unified formulation for surrogate relevance criteria was developed.
- Neighborhood relationships among atlases were modeled using oracle relevance knowledge.
- Surrogates were trained to distinguish geometrically relevant neighbors from irrelevant ones.
Main Results:
- The proposed framework was validated in corpus callosum segmentation.
- Learned surrogates outperformed benchmark surrogates in inferring oracle values.
- Superiority was demonstrated in selecting relevant fusion sets for segmentation.
Conclusions:
- The developed framework effectively learns surrogate relevance criteria.
- This approach enhances data relevance assessment in multi-atlas image segmentation.
- The method shows significant improvements over existing benchmark surrogates.
