Related Experiment Video
Updated: May 9, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Left ventricular segmentation from MRI datasets with edge modelling conditional random fields.
Janto F Dreijer1, Ben M Herbst, Johan A du Preez
1Department of Applied Mathematics, Stellenbosch University, Stellenbosch, South Africa. jantod@gmail.com
BMC Medical Imaging
|August 1, 2013
Summary
This study presents a novel method for automatic segmentation of cardiac ventricles in MRI scans, successfully integrating shape and motion data to overcome challenges posed by papillary muscles. The approach achieves high accuracy, improving cardiac image analysis.
Area of Science:
- Medical imaging analysis
- Cardiovascular image processing
- Computational anatomy
Background:
- Automatic segmentation of the left cardiac ventricle in short-axis MRI is challenging due to factors like papillary muscles near the endocardium.
- Traditional threshold-based methods struggle with complex cardiac structures.
Purpose of the Study:
- To develop and validate an automated method for segmenting left cardiac ventricle contours in short-axis MRI.
- To address the difficulties in segmentation caused by papillary muscles and other anatomical variations.
Main Methods:
- Modeling endo- and epicardium as related series of radii, incorporating shape and motion features.
- Utilizing a discriminatively trained Conditional Random Field (CRF) with loopy belief propagation for segmentation inference.
- Employing Powell's method for CRF parameter optimization and frame alignment error minimization.
Main Results:
- The algorithm demonstrates robustness against papillary muscle inclusion by effectively integrating shape and motion information.
- Achieved high segmentation accuracy on the Sunnybrook dataset, with average Dice metrics of 0.91 ± 0.02 (inner) and 0.93 ± 0.02 (outer).
Conclusions:
- The developed model successfully integrates shape and motion for robust segmentation of cardiac ventricle contours, even with papillary muscles present.
- Identified challenges in patients with hypertrophy where the blood pool may not be visible at end-systole, suggesting areas for future improvement.
