Related Experiment Video
Updated: Oct 26, 2025
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
Triplanar ensemble U-Net model for white matter hyperintensities segmentation on MR images
Vaanathi Sundaresan1, Giovanna Zamboni2, Peter M Rothwell3
1Wellcome Centre for Integrative Neuroimaging, Oxford Centre for Functional MRI of the Brain, Nuffield Department of Clinical Neurosciences, University of Oxford, UK; Oxford-Nottingham Centre for Doctoral Training in Biomedical Imaging, University of Oxford, UK; Oxford India Centre for Sustainable Development, Somerville College, University of Oxford, UK. Electronic address: https://www.ndcn.ox.ac.uk/team/vaanathi-sundaresan.
This study introduces an advanced AI method for accurately segmenting white matter hyperintensities (WMHs) in brain MRIs. The novel approach improves WMH quantification, aiding in the understanding of neurological diseases.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Artificial Intelligence
Background:
- White matter hyperintensities (WMHs) are linked to cerebrovascular and neurodegenerative diseases.
- Accurate WMH quantification is crucial for clinical impact assessment in diverse populations.
- Automated WMH segmentation faces challenges due to heterogeneity, artifacts, and varied pathologies.
Purpose of the Study:
- To develop an accurate automated method for segmenting white matter hyperintensities (WMHs).
- To improve WMH quantification by addressing challenges in deep and periventricular regions.
- To evaluate the proposed method against existing techniques and a benchmark challenge.
Main Methods:
- An ensemble triplanar network was proposed, integrating predictions from three imaging planes.
- Network loss functions incorporated anatomical WMH distribution information.
- The method was evaluated on five datasets, including MICCAI WMH Segmentation Challenge 2017 data.
Main Results:
- The proposed method achieved robust and comparable performance in both deep and periventricular WMH regions.
- It outperformed several existing methods, including FSL BIANCA.
- Performance was on par with top-ranked deep learning methods in the MWSC 2017 challenge.
Conclusions:
- The ensemble triplanar network offers an accurate and robust solution for WMH segmentation.
- Incorporating anatomical priors in loss functions enhances segmentation efficiency and handles contrast variations.
- This method advances automated WMH quantification for clinical and research applications.
