Comparison of non-parametric T2 relaxometry methods for myelin water quantification

• A comparison between ten methods for computing the T 2 distribution was performed. Various regularization techniques and penalty functions were tested. Our findings demonstrated need regularization. list of recommendations selecting optimal algorithms provided. The implemented packaged in a freely distributed toolbox. Multi-component relaxometry allows probing tissue microstructure by assessing compartment-specific relaxation times water fractions, including myelin fraction. Non-negative least squares (NNLS) with zero-order Tikhonov is conventional method estimating smooth distributions. Despite improved estimation provided this compared to non-regularized NNLS, solution still sensitive underlying noise weight. This especially relevant clinically achievable signal-to-noise ratios. In literature inverse problems, various well-established approaches promote solutions, first-order second-order regularization, different criteria weight have been proposed, such as L-curve, Generalized Cross-Validation, Chi-square residual fitting. However, quantitative comparisons available reconstruction distribution, weight, are lacking. study, we evaluated algorithms, resulting from individual combinations three terms estimate plus NNLS. Their performance both simulated data real brain MRI acquired healthy volunteers through scan-rescan repeatability analysis. demonstrate As result work, provide based on data. Moreover, toolbox reproducible research, facilitate further research use promising technique clinical practice.