Removal of FM Sidebands Artifacts in NWS MRS by QZ-bac Algorithm

Introduction Non-water-suppressed (NWS) MRS techniques [1,2] have been recently investigated for the benefits that the un-suppressed water signal in NWS MRS serving as an internal reference for reliable metabolic quantification while avoiding the possible distortion of metabolite signal by water suppression (WS) pulses. However there are still some problems originating from the huge water peak. First the main water peak needs to be removed by some post processing procedures. For this part several time domain SVD-based [2] water-modeling methods have been developed to deal with the huge water resonance. Secondly, the Frequency-Modulated (FM) sideband artifacts related to gradient oscillation have to be eliminated. Otherwise, the quantification of metabolites will be severely hampered by the sideband contaminations [3,4]. The FM sidebands artifacts originating from the frequency modulation of water signal caused by the oscillating spoiler gradients can be reduced by increasing the echo time [1] or other experimental method [4,5]. However post-processing methods is more desirable for the advantage that artifacts can be eliminated without modifying the pulse sequences. In this study, we develop a post-processing algorithm, named QZ-Based Artifacts Cancellation (QZ-bac), to remove the FM sidebands in NWS MRS. we will demonstrate the performance of QZ-bac by a computer simulation and in vivo experiment. The QZ-bac method is a purely post-processing algorithm, which is expected to improve the feasibility and popularity for current NWS MRS at both short and long echo time. Theory and Methods QZ-bac algorithm: The algorithm is based on the Filter diagnonalization Method (FDM) [6] and can be summarized as follow. First, we modify the real/imaginary signals with weighting by ) 1 ( Δ + and ) 1 ( Δ − respectively. the modified signal is the combination of original signal ) (t s , the modified sidebands ) ( ~ ) 1 ( t s Δ − , and the Skew-Hermitian ghost ) (t s ⋅ Δ , related to signal from down-field. By FDM processing we can filter out the undesired ghost from ) (t s ⋅ Δ such that only the up field signals are included. This process can be carried out for many times until all the sideband term ) ( ~ ) 1 ( t s Δ − can be effectively reduced. Simulation: Three singlets were simulated to mimic the three metabolites, N-acetyl aspartate (NAA), Choline and Creatine, for in vivo condition. The complex Lorentzian spectral line was generated with additional water signal. The water is 10 times in the magnitude. The simulated spectrum was then modulated by 6.8Hz Gaussian decay and FM with following parameters: the modulation index mf was 0.2% , modulation frequency (ωG/2π) was 191 Hz. In order to validate the efficiency of algorithm, we perform Monte Carlo simulation with 400 realizations. To evaluate the performance under different SNR, normally distributed complex noises were generated using true random seeds. The spectrum processed by QZ-bac was then compared with the spectrum without noise between spectral ranging from 3.3 ppm to 1.9 ppm. NWS MRS Experiments: In vivo experiments were performed on healthy subjects on a 3T system (Tim TRIO, Siemens Medical Solutions, Erlangen, Germany). PRESS sequence was used with parameters: TE=30ms, TR=2000ms, NEX=130, Voxel size=20×20×20 mm. For each subject two scans including NWS and WS were performed sequentially at the same localization. WS scans were then taken as reference standard for the comparison. . Before data processing zero order phase correction and frequency shift correction was done according to the water peak. The water subtraction and associated FM sidebands artifacts cancelation were then carried out using the Matrix Pencil Method [1] and QZ-bac algoritm, with rank=20 and limited between ±100Hz. Tails of FID were also removed by fitting algorithm. Results and Discussion