Related Experiment Video
Updated: May 26, 2026
04:22
Pipeline for Multi-Scale Three-Dimensional Anatomic Study of the Human Heart
Published on: June 28, 2024
Hierarchical scale-based multiobject recognition of 3-D anatomical structures.
Ulas Bagci1, Xinjian Chen, Jayaram K Udupa
1Center for Infectious Disease Imaging, Department of Radiology and Imaging Sciences, National Institutes of Health, Bethesda, MD 20892, USA.
IEEE Transactions on Medical Imaging
|December 29, 2011
Summary
This study introduces a novel method for segmenting anatomical structures in 3D medical images by automatically locating and delineating them. The approach enhances accuracy by integrating shape models and intensity-weighted object extraction, improving medical image analysis.
Area of Science:
- Medical Imaging
- Computer Vision
- Computational Anatomy
Background:
- Accurate segmentation of anatomical structures in medical images is crucial for diagnosis and treatment planning.
- Current methods often struggle with automatic localization and delineation, requiring manual intervention or complex optimization processes.
Purpose of the Study:
- To develop a generalized framework for anatomy segmentation that addresses automatic structure localization and delineation.
- To improve the accuracy and robustness of medical image segmentation across different modalities.
Main Methods:
- Proposed an intensity-weighted ball-scale object extraction concept for automatic anatomical structure recognition without search or optimization.
- Integrated graph-cut (GC) segmentation with a prior shape model for delineation.
- Evaluated the framework on 3D abdominal CT scans and 2D/3D foot MR images.
Main Results:
- Effective recognition significantly improved delineation accuracy.
- Incorporating a large number of anatomical structures in the shape model dramatically enhanced recognition and delineation.
- Ball-scale analysis provided valuable object-image relationship insights.
- Intensity variations across images degraded object recognition performance.
Conclusions:
- The proposed method offers an effective approach for automated anatomical structure segmentation in medical imaging.
- The framework demonstrates generalizability across CT and MR modalities, highlighting its robustness.
- Future work should address the impact of intensity non-standardness on recognition performance.
