Early detection of white matter hyperintensities using SHIVA-WMH detector
Ami Tsuchida1,2, Philippe Boutinaud3, Violaine Verrecchia1,2
1GIN, IMN-UMR5293, Université de Bordeaux, CEA, CNRS, Bordeaux, France.
Human Brain Mapping
|December 5, 2023
Summary
A new deep-learning tool, SHIVA-WMH, accurately detects white matter hyperintensities (WMH) across all severities. This advancement aids research into WMH, a marker for stroke and dementia risk.
Area of Science:
- Neuroimaging
- Medical image analysis
- Cerebrovascular disease research
Background:
- White matter hyperintensities (WMH) are key indicators of cerebral small vessel disease.
- WMH are linked to increased risks of stroke, dementia, and mortality.
- Existing automated WMH segmentation tools struggle with detecting small, sparse lesions.
Purpose of the Study:
- To introduce SHIVA-WMH, a novel deep-learning tool for automated WMH segmentation.
- To evaluate SHIVA-WMH's performance across a wide spectrum of WMH severity.
- To provide a publicly available tool for WMH research.
Main Methods:
- Developed a deep-learning model trained on manually segmented WMH across diverse severities.
- Validated SHIVA-WMH on three independent datasets: MRI-Share, MICCAI 2017 WMH challenge, and UK Biobank.
- Compared SHIVA-WMH performance against three established WMH segmentation tools.
Main Results:
- SHIVA-WMH demonstrated high detection efficiency in subjects with both small punctate and large confluent WMH.
- Achieved superior voxel-level (0.66) and lesion cluster-level (0.71) Dice scores compared to reference tools.
- Outperformed LPA, PGS tool, and HyperMapper across diverse WMH loads.
Conclusions:
- SHIVA-WMH offers robust and accurate WMH segmentation across varying disease severity.
- The tool's open availability will advance research on WMH emergence and progression.
- SHIVA-WMH facilitates broader investigations into WMH in diverse populations.
