Contrast-based fully automatic segmentation of white matter hyperintensities: method and validation.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Neurology

Background:

• White matter hyperintensities (WMH) are common in elderly individuals on MRI.
• WMH are linked to stroke and dementia, necessitating accurate quantification.
• Current WMH assessment often involves time-consuming manual segmentation.

Purpose of the Study:

• To introduce WHASA (White matter Hyperintensities Automated Segmentation Algorithm), a novel automated method for WMH segmentation.
• To evaluate WHASA's performance against manual segmentation and existing automated methods.
• To assess WMH segmentation in multicentre studies across different MRI scanners.

Main Methods:

• WHASA utilizes non-linear diffusion filtering and watershed segmentation to enhance contrast.
• Segmentation relies on automatically computed thresholds and anatomical information.
• The algorithm processes FLAIR and T1 MRI sequences.

Main Results:

• WHASA achieved high accuracy with an intraclass correlation coefficient (ICC) of 0.96 and a mean similarity index (SI) of 0.72 compared to manual segmentation.
• The algorithm outperformed unsupervised methods and performed comparably to supervised methods (kNN, SVM).
• WHASA demonstrated robust performance across diverse patient data from multiple scanners without requiring a training set.

Conclusions:

• WHASA provides an accurate and efficient automated solution for WMH segmentation.
• The algorithm is suitable for multicentre studies and various lesion loads.
• WHASA offers a significant advancement over manual and existing automated WMH assessment techniques.