Improved Automatic Segmentation of White Matter Hyperintensities in MRI Based on Multilevel Lesion Features

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Brain white matter hyperintensities (WMHs) are crucial indicators of cerebrovascular and neurodegenerative diseases in the elderly.
• Accurate detection and characterization of WMHs are vital for clinical assessment and disease management.

Purpose of the Study:

• To develop and evaluate a novel, automated approach for segmenting white matter hyperintensities (WMHs) from multi-contrast MRI data.
• To improve segmentation performance using both voxel-based and lesion-based information for enhanced accuracy in volume and object metrics.

Main Methods:

• The proposed AMOS-2D method employs a four-stage "generate-and-test" strategy: pre-processing, Gaussian white matter (WM) modeling, hierarchical multi-threshold WMH segmentation, and support vector machine-based object filtering.
• Utilized volumetric T1-weighted and 2D FLAIR MRI sequences from 28 subjects with diverse WMH loads.
• Expert-defined manual lesion masks served as the ground truth for training and validation.

Main Results:

• The AMOS-2D method achieved performance comparable to inter-expert agreement for both WMH number (Dice Similarity Coefficient [DSC] = 0.637) and volume (DSC = 0.743).
• Demonstrated superior accuracy in WMH detection compared to alternative methods.
• Exhibited consistent performance across a wide range of WMH lesion burdens, indicating robustness.

Conclusions:

• The AMOS-2D approach provides a valuable tool for fully automated WMH segmentation in clinical settings.
• Its high agreement with expert annotations and stability across varying lesion loads support its utility for patients with cerebrovascular and neurodegenerative diseases.