Related Experiment Videos
Labeling of lumbar discs using both pixel- and object-level features with a two-level probabilistic model.
Raja' S Alomari1, Jason J Corso, Vipin Chaudhary
1Department of Computer Science and Engineering, State University of New York–Buffalo, Buffalo, NY 14260, USA. ralomari@buffalo.edu
IEEE Transactions on Medical Imaging
|April 10, 2010
Summary
This study introduces a novel two-level probabilistic model for precise lumbar spine disc localization using MRI scans. The model enhances abnormality detection in vertebral column analysis.
Area of Science:
- Medical Imaging
- Biomedical Engineering
- Radiology
Background:
- Accurate detection and labeling of vertebral column structures are crucial for spinal analysis.
- Measurements of appearance, shape, and geometry are vital for identifying local (e.g., herniation) and global (e.g., scoliosis) spinal abnormalities.
Purpose of the Study:
- To propose a two-level probabilistic model for localizing discs in clinical magnetic resonance imaging (MRI) data.
- To capture both pixel-level and object-level features for improved disc localization.
Main Methods:
- A two-level probabilistic model utilizing a Gibbs distribution for pixel-level appearance and spatial information.
- Object-level modeling of disc spatial distribution and inter-disc distances.
- Generalized expectation-maximization algorithm for efficient optimization and disc label convergence.
Main Results:
- The model demonstrated efficiency and robustness by leveraging conditional independence at the pixel-level.
- Encouraging results were achieved on a dataset of 105 clinical MRI cases, including both normal and abnormal lumbar spine findings.
Conclusions:
- The proposed two-level probabilistic model offers an effective approach for lumbar spine disc localization from MRI.
- The method shows promise for enhancing the accuracy and efficiency of spinal abnormality detection.
