Robust myelin water quantification using spatially regularized nonnegative least square algorithm

Introduction: A quantitative measurement of the myelin content of white matter (WM) can be used as a significant predictor of the prognoses of the clinically isolated syndrome and for earlier diagnosis of WM diseases such as multiple sclerosis (MS). One approach developed to provide valuable information on myelin content is to measure myelin water fraction (MWF) by analyzing T2 decay curves using a nonnegative least square (NNLS) algorithm [1,2]. This technique has successfully demonstrated the merit of quantitative measurement of MWF in the study of MS. The NNLS algorithm is commonly used with a regularization in the T2 spectral domain, which will be referred to as rNNLS hereafter. The rNNLS algorithm showed an improved performance, but it is still prone to the noise in the measurements [3]. The purpose of our current study is to develop a new analysis approach, referred to as a spatially regularized NNLS (srNNLS) algorithm, for robust MWF estimation with a reduced sensitivity to the noise. In the srNNLS algorithm, the regularization is expanded into the spatial domain in addition to the T2 spectral domain.