Iteratively reweighted L1-fitting for model-independent outlier removal and regularization in diffusion MRI

Diffusion magnetic resonance imaging is negatively affected by subject motion occurring during the image acquisition. The induced data artifacts adversely influence the estimation of microstructural diffusion measures. State-of-the-art procedures for outlier removal detect and reject defective images during model fitting. These methods, however, are tailored only for specific diffusion models and excluding a varying number of diffusion-weighted images might be disadvantageous for the parameter estimation. Therefore, this work proposes a novel method based on an iteratively reweighted Ll-Fitting for model-independent outlier removal with subsequent reconstruction of faulty images by modeling the signal in the continuous SHORE basis. We validate the proposed method on simulation data and clinical in vivo human brain scans and demonstrate its effect on diffusion parameters determined by the kurtosis and NODDI model.