Improved SENSE reconstruction by pre-scaled TLS method

Abstract

In SENSE reconstruction, the unfolding process usually involves solving an over-determined linear system of equations (LSE) using a least squares (LS) estimation. In this work it is proposed that when the coil sensitivity maps can not be accurately calibrated the pre-scaled total least squares (TLS) method is more suitable for unfolding aliased pixels. By carefully choosing the relative scaling parameter for the LSE, the TLS method leads to more accurate estimation than ordinary LS method and hence the reconstructed images has less aliasing artifacts and better SNR.