White matter hyperintensities segmentation using the ensemble U-Net with multi-scale highlighting foregrounds.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Image Analysis

Background:

• White matter hyperintensities (WMHs) are MRI indicators linked to aging, cognitive decline, and dementia.
• Accurate segmentation of WMHs is crucial for understanding their clinical impact.

Purpose of the Study:

• To develop and evaluate a U-Net with multi-scale highlighting foregrounds (HF) for improved WMH segmentation.
• To assess the clinical utility of WMH volumes derived from the proposed method.

Main Methods:

• Implementation of a U-Net architecture incorporating multi-scale highlighting foregrounds (HF).
• Evaluation on the WMH Segmentation Challenge dataset.
• Assessment of clinical utility using the Alzheimer's Disease Neuroimaging Initiative database.

Main Results:

• The U-Net with HF significantly enhanced WMH voxel detection, especially at boundaries and in small clusters.
• Achieved top performance among 39 methods in the WMH Segmentation Challenge, with the highest Dice similarity index and F1-score.
• Automatically computed WMH volumes correlated significantly with cognitive performance and improved classification of cognitive states.

Conclusions:

• The U-Net with HF offers a robust and accurate solution for white matter hyperintensity segmentation.
• The method demonstrates significant clinical relevance in diagnosing and monitoring neurodegenerative diseases like Alzheimer's.
• Publicly available implementation facilitates broader research and clinical application.