Prewhitening and Maximum Likelihood Estimation for Root-Sum-of-Squares Images

INTRODUCTION: In recent years, phased array coils have seen widespread adoption across a variety of MR applications. Multiple receivers can be exploited to improve SNR and/or reduce scan time through parallel imaging methods. Although the matched-filter is the theoretically optimal method for combining coils (highest SNR and no bias) [1], it requires accurate coil sensitivity maps and access to complex image data. Since these are not always available, the alternative Root-Sum-of-Squares (RSS) is the most commonly used method to combine signals originating from different coils. Similar to single-channel or quadrature reception, the assumption of Gaussian noise is valid for RSS images at high SNR. At low SNR, there is a magnitude bias that increases with the number of coils [2]. This can cause systemic errors in quantitative measurements that depend on data points close to the noise floor (e.g. Diffusion, Relaxometry, Magnetization Transfer, etc.).