Related Experiment Video
Updated: Apr 11, 2026
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.6K
Automatic segmentation of MR brain images of preterm infants using supervised classification
Pim Moeskops1, Manon J N L Benders2, Sabina M Chiţ3
1Image Sciences Institute, University Medical Center Utrecht, The Netherlands; Department of Neonatology, University Medical Center Utrecht, The Netherlands.
Neuroimage
|June 10, 2015
Summary
This study presents an automated algorithm for segmenting brain tissues in preterm infants using MRI scans. The method accurately identifies white matter, grey matter, and cerebrospinal fluid, aiding in the assessment of brain development.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Developmental Neuroscience
Background:
- Preterm birth can lead to impaired brain development, necessitating quantitative assessment of brain structures.
- Accurate segmentation of brain tissues in magnetic resonance (MR) images is crucial for evaluating preterm brain development.
- Existing methods may require extensive manual annotation, limiting their efficiency.
Purpose of the Study:
- To develop and validate an automated algorithm for segmenting unmyelinated white matter (WM), cortical grey matter (GM), and cerebrospinal fluid (CSF) in the extracerebral space of preterm infant brains.
- To assess the algorithm's performance using T2-weighted MR images only.
- To evaluate the impact of training data annotation completeness on segmentation accuracy.
Main Methods:
- A supervised voxel classification algorithm was employed in three stages: initial labelling of easily classifiable voxels, dedicated analysis of remaining voxels, and multi-class classification for inconsistency resolution.
- The system was optimized to use only T2-weighted MR images for automatic segmentation.
- Performance was evaluated using Dice coefficients on images from preterm infants at 30 and 40 weeks postmenstrual age (PMA), comparing training with fully versus partially annotated data.
Main Results:
- The algorithm achieved high average Dice coefficients: 0.95 for WM, 0.81 for GM, and 0.89 for CSF at 30 weeks PMA when trained on fully annotated data.
- Training with partially annotated data yielded comparable results: 0.95 for WM, 0.78 for GM, 0.87 for CSF at 30 weeks PMA, and 0.92 for WM, 0.80 for GM, 0.85 for CSF at 40 weeks PMA.
- Segmentation performance closely approximated that of a second human expert across all tested conditions.
Conclusions:
- The developed algorithm provides robust and accurate automatic segmentation of WM, GM, and CSF in preterm infant brains.
- The method demonstrates resilience to variations in infant age and MR acquisition protocols.
- Reducing the need for extensive training data annotation enhances the method's applicability and efficiency in clinical research.
