Automatic WML Segmentation & Quantification using a Machine Learning Approach

Mariano Rincon¹, Per Selnes², Christopher Alfred Larsson³, Tormod Fladby², Atle Fillibom Bjrnerud³

A machine learning method for automatic segmentation of white matter lesions with a high success rate in spite of sub-optimal 2D FLAIR images is proposed. The automatic method facilitates use in large data sets. Each white matter lesion is characterized by a vector of 105 local and global features, which enable robust segmentation and further classification according to different criteria for other analysis. Based on the results obtained, the method warrants further testing in ongoing studies in patients with neurodegenerative disease.