Maximum likelihood ADC parameter estimates improve selection of metastatic cervical nodes for patients with head and neck squamous cell cancer

The aim of this work was to determine whether classification of benign and metastatic cervical nodes based on diffusion weighted imaging (DWI) could be improved by use of a maximum likelihood algorithm for derivation of ADC parameters. A non linear least squares (LSQ) algorithm is usually used to fit parameters to the measured MR signal intensities as a function of b-value. LSQ assumes that the noise in high b-values is normally distributed whereas in reality it follows a Rice distribution. To account for the Rician noise, maximum likelihood (ML) algorithms have been proposed that provide unbiased ADC estimates. In this work the monoexponential, stretched exponential and biexponential models were examined, with their involved parameters calculated using the LSQ and the ML algorithms.