Effects of Noise Estimation Error on the Accuracy and Precision of Maximum Likelihood Estimation of Apparent Diffusion Coefficient

Jing Yuan¹, Qinwei Zhang¹, David Ka Wai Yeung¹, Anil T. Ahuja¹, Ann D. King¹

Maximum likelihood estimation (MLE) is preferable as an unbiased estimator compared to least-squares estimation (LSE) due to the Rician noise distribution of magnitude MR image. Accurate noise estimation is essential for MLE that affects accuracy and precision of MLE, but is often hampered by many factors in practice. We investigate the effects of erroneous noise estimation on ADC estimation by MLE through simulation and ADC mapping of clinical DWI images. Results show that MLE accuracy and precision are significantly reduced by noise estimation error at low signal-to-noise ratios (SNRs), but exhibit fairly good robustness to such errors at high SNRs>10.