Diffusion kurtosis imaging (DKI) often yields abnormally low mean kurtosis (MK) values that are physically and/or biologically implausible. We aim to characterize the relationship between abnormally low MK and baseline (b0) values. We show that too low b0 signals explain abnormally low MK values. We propose an automatic and threshold free approach for the identification of low MK voxels, along with a correction strategy based on adaptive smoothing. Our results suggest that modifying the b0 is sufficient to resolve the vast majority of low MK values, and is preferred over two other popular correction methods.