Unsupervised Models for White Matter Fiber-Bundles Analysis in Multiple Sclerosis

A major challenge of neuroimaging research consists in identifying new markers that can accurately characterize pathological processes and predict clinical outcomes. Achieving this goal is particularly crucial in Multiple Sclerosis (MS), the primary cause of neurological disability in young adults and remains without well-known etiology. MS is a chronic demyelinating inflammatory disease of the central nervous system, characterized by white matter (WM) lesions that are well detected by conventional MRI. However, T2 lesion load is moderately correlated with the patient clinical status leading to the development of more sensitive techniques such as diffusion tensor imaging (DTI).
DTI is a promising technique for white matter WM fiber-tracking and microstructural characterization of axonal/neuronal integrity and connectivity. By measuring water molecules motion in the three directions of space, numerous parametric maps can be reconstructed based on eigenvalues of the diffusion tensor. Among these, fractional anisotropy (FA), mean diffusivity (MD), and radial (λr) diffusivity have extensively been used to investigate brain diseases. In MS, DTI has proved to be sensitive enough to detect microscopic changes occurring in WM lesions, normal appearing white matter (NAWM). Indeed, several studies have demonstrated how diffusion markers change in lesions when compared to NAWM of MS patients and to NAWM of healthy controls. Furthermore, these pathological events may occur along afferent or efferent WM fiber pathways, leading to antero- or retrograde degeneration. Thus, for a better understanding of MS pathological processes spatial progression, an accurate and sensitive characterization of WM fibers along their pathways is needed.
By merging the spatial information of fiber tracking with the diffusion metrics derived from the tensor, WM fiber-bundles could be modeled and analyzed along their profile.
In this work, we will describe different semi-supervised and unsupervised methods to extract and analyze longitudinal changes occurring along WM fiber-bundles. In the first part, will introduce a new string-based formalism for unsupervised and semi supervised extraction of WM fiber-bundles. In the second part we will describe a distribution-based method to detect longitudinal changes. In the third part we show how non-negative matrix factorization can be used to longitudinal analysis of WM fiber-bundle. Finally, we will generalize the proposed formalism using tensor- based factorization.