A Supervised Hybrid Classifier for Brain Tissues and White Matter Lesions on Multispectral MRI

Abstract

Accurate quantification of brain tissues is a very challenging problem in neuroimaging, such as quantification of gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF), and white matter lesions (WMLs). However, on many cases brain tissues and white matter lesions cannot be segmented and separated simultaneously by current techniques. Recently, a TRIO algorithm (TRIOA) is proposed to integrate three algorithms, Independent Component Analysis (ICA), Support Vector Machine (SVM) and Fisher's Linear Discriminant Analysis (FLDA) (ICA+SVM+IFLDA) which can effectively classify GM, WM and CSF by considering MR images as multispectral images in the native coordinate space. This paper further extends TRIOA in conjunction with Band Expansion Process (BEP), called Extended TRIOA (ETRIOA), to classify brain tissues as well as WMLs simultaneously. The accuracy assessment of the ETRIOA was evaluated by using the similarity index. The conducted experimental results demonstrate the clinical applicability of ETRIOA in simultaneous classification of GM, WM, CSF and WMLs.