Regularized Spectral Lineshape Deconvolution

Yan Zhang¹, Shizhe Li¹, Jun Shen¹

The process of lineshape deconvolution is an inverse problem. A new referencing deconvolution method is proposed, which uses Tiknohov regularization to restrain the noise amplification. To determine the optimal regularization, the noise to signal ratio in frequency domain was defined as a function of the regularization parameter. It was found that this function yielded a well-defined L-curve with the transition point that marks the optimal regularization parameter. The method was validated on 1H spectral data which were acquired on human brain with single voxel at 3T. The spectral quality was markedly improved after the data were processed with the proposed method.