Fitting to Magnitude Diffusion MRI Data Using a Least Squares Algorithm Gives Biased ADC Values and Is Less Able to Characterise Necrosis

Simon Walker-Samuel¹, Matthew Orton¹, Lesley D. McPhail¹, Simon P. Robinson¹

The noise in magnitude MR data is Rice-distributed, and it has been shown previously that fitting using a least-squares algorithm leads to biased ADC estimates. In this study, the magnitude of this bias in orthotopic PC3 tumours is investigated and compared with a robust maximum likelihood approach. It is found that least-squares over-estimates ADC by an average of 23.4 12 %.. More significantly, regions of histologically-confirmed necrosis are not identifiable in ADC maps from the least-squares algorithm, but can be clearly observed in maximum-likelihood maps.