Related Experiment Video
Updated: May 26, 2026
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
Robust white matter lesion segmentation in FLAIR MRI
April Khademi1, Anastasios Venetsanopoulos, Alan R Moody
1Department of Electrical and Computer Engineering, University of Toronto, Toronto, ON, Canada. akhademi@ieee.org
IEEE Transactions on Bio-Medical Engineering
|December 29, 2011
Summary
This study presents an automated method for segmenting white matter lesions (WMLs) in MRI scans, improving accuracy by accounting for partial volume averaging. This offers a reliable alternative to manual measurements for conditions like stroke.
Area of Science:
- Medical Imaging
- Neuroimaging
- Biomedical Engineering
Background:
- White matter lesions (WMLs) are crucial indicators of neurological conditions such as stroke and carotid disease.
- Accurate quantification of WML volume is essential for diagnosis and monitoring, but manual methods are time-consuming and error-prone.
- Automated segmentation offers an objective, efficient, and reliable approach to WML volume measurement.
Purpose of the Study:
- To develop and validate an automated white matter lesion (WML) segmentation scheme for fluid attenuation inversion recovery (FLAIR) magnetic resonance imaging (MRI).
- To achieve subvoxel precision in WML volume computation by accurately modeling and correcting for partial volume averaging (PVA) artifacts.
- To provide an automated tool for lesion load studies, analyzing WML volumes per brain hemisphere.
Main Methods:
- A novel WML segmentation method for FLAIR MRI that computes lesion volume with subvoxel precision.
- Modeling of partial volume averaging (PVA) using a localized edge strength measure in 3D, transformed to a global representation for noise robustness.
- Determination of the PVA fraction (α) in mixture voxels to refine segmentation accuracy.
Main Results:
- High WML segmentation performance on simulated and real FLAIR images, achieving 98.9% and 83% overlap with ground truth, respectively.
- Demonstrated superior performance compared to existing automated WML segmentation methods.
- Successful implementation of automated lesion load studies analyzing WML volumes in each brain hemisphere separately.
Conclusions:
- The proposed automated WML segmentation method provides accurate and reliable WML volume quantification from FLAIR MRI.
- The technique effectively addresses partial volume averaging artifacts, outperforming traditional methods.
- This approach offers significant advantages, including no requirement for distributional assumptions, parameters, or training samples, and uses a single MR modality.
