Automatized spleen segmentation in non-contrast-enhanced MR volume data using subject-specific shape priors
Oliver Gloger1, Klaus Tönnies2, Robin Bülow3
1Institute for Community Medicine, University of Greifswald, Walther-Rathenau-Str. 48, 17475 Greifswald, Germany.
Physics in Medicine and Biology
|June 2, 2017
Summary
This study introduces an automated 3D spleen segmentation framework using T1-weighted MRI. The novel approach accurately delineates spleen tissue and calculates volume, achieving high precision in segmentation.
Area of Science:
- Medical Imaging
- Radiology
- Computational Anatomy
Background:
- Accurate spleen segmentation is crucial for volumetric analysis in medical imaging.
- Low contrast between spleen and adjacent tissues in non-contrast-enhanced MRI poses a challenge for automated segmentation.
Purpose of the Study:
- To develop and validate a fully automated 3D spleen segmentation framework using T1-weighted MRI.
- To address the challenge of low inter-tissue contrast in non-contrast-enhanced spleen imaging.
Main Methods:
- A modular framework incorporating prior knowledge, including subject-specific 3D shape models.
- Utilized support vector machines for classification of spleen regions and shape types.
- Employed a 3D level set segmentation method guided by both image and shape-driven forces.
Main Results:
- Achieved a mean Dice coefficient of approximately 0.91, indicating high segmentation accuracy.
- Reported a low volumetric mean error of 6.3% for spleen volumetry.
- Demonstrated successful spleen delineation in native T1-weighted MR volume data.
Conclusions:
- The developed framework provides a robust solution for automated 3D spleen segmentation from T1-weighted MRI.
- Integration of diverse prior shape knowledge enhances segmentation performance.
- The approach is suitable for spleen delineation and volumetry in epidemiological studies.
