Technical and clinical validation of a novel deep learning-based white matter hyperintensity segmentation tool
Benno Gesierich1, Lukas Pirpamer1, Dominik S Meier1
1Medical Image Analysis Center (MIAC), Basel, Switzerland.
Cerebral Circulation - Cognition and Behavior
|September 22, 2025
Summary
We developed novel white matter hyperintensity (WMH) segmentation algorithms that outperform existing tools. These advanced deep learning models offer improved accuracy and efficiency for analyzing cerebral small vessel disease on MRI scans.
Area of Science:
- Neuroimaging
- Medical image analysis
- Cerebral small vessel disease
Background:
- White matter hyperintensities (WMH) are key indicators of cerebral small vessel disease.
- Existing WMH segmentation tools have limitations impacting usability.
- There is a need for improved, validated WMH segmentation algorithms.
Purpose of the Study:
- To develop and validate a novel WMH segmentation algorithm.
- To address limitations of current WMH segmentation tools.
- To provide a user-friendly and generalizable solution for WMH analysis.
Main Methods:
- Trained deep learning models (MD-GRU, nnU-Net) on a heterogeneous dataset.
- Benchmarked new models against state-of-the-art algorithms using independent datasets.
- Conducted technical validation (bias, precision, repeatability, reproducibility) and clinical validation (treatment effect detection).
Main Results:
- The novel algorithms outperformed benchmarking tools in agreement, bias, and precision.
- nnU-Net demonstrated superior statistical power for detecting treatment effects, reducing sample size by 41%.
- Algorithms showed excellent generalization capabilities across diverse datasets.
Conclusions:
- Two novel WMH segmentation algorithms were developed and validated.
- The algorithms exhibit strong generalization and outperform existing methods.
- Publicly available, user-friendly processing pipelines can be applied widely without re-training.
