A large margin algorithm for automated segmentation of white matter hyperintensity. (May 2018)

• Record Type: Journal Article
• Authors: Qin, Chen; Guerrero, Ricardo; Bowles, Christopher; Chen, Liang; Dickie, David Alexander; Valdes-Hernandez, Maria del C.; Wardlaw, Joanna; Rueckert, Daniel
• Is Part Of: Pattern recognition. Volume 77(2018:May)
• Page Start: 150
• Page End: 159
• Publication Date: 2018-05
• Subjects: Supervised learning -- Semi-supervised learning -- Segmentation -- White matter hyperintensity -- Brain MRI; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.12.016
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A novel large margin method for white matter hyperintensity segmentation is proposed. A supervised large margin algorithm is proposed to learn a global classifier. A semi-supervised large margin classifier is learned for refinement on test subject. The proposed model shows competitive performance on subjects with vascular disease. Abstract: Precise detection and quantification of white matter hyperintensity (WMH) is of great interest in studies of neurological and vascular disorders. In this work, we propose a novel method for automatic WMH segmentation with both supervised and semi-supervised large margin algorithms provided by the framework. The proposed algorithms optimize a kernel based max-margin objective function which aims to maximize the margin between inliers and outliers. We show that the semi-supervised learning problem can be formulated to learn a classifier and label assignment simultaneously, which can be solved efficiently by an iterative algorithm. The model is learned first via the supervised approach and then fine-tuned on a target image by using the semi-supervised algorithm. We evaluate our method on 88 brain fluid-attenuated inversion recovery (FLAIR) magnetic resonance (MR) images from subjects with vascular disease. Quantitative evaluation of the proposed approach shows that it outperforms other well known methods for WMH segmentation.