Towards high-resolution magnetic resonance spectroscopic imaging: spatiotemporal denoising and echo-time selection

Magnetic resonance spectroscopic imaging (MRSI) enables acquisition of spatial-spectral nuclear spin distributions. Compared to conventional MRI, the additional spectral information provides a powerful tool for in vivo study of biological tissues. However, considerable practical challenges, which cannot be overcome using the traditional Fourier method, remain in obtaining spatial-spectral data with both high resolution and high signal-to-noise ratio (SNR). In addition, the choice of echo time (TE) during data acquisition affects the SNR and complexity of the model used to describe measured data. TE selection remains a complicated and controversial issue in proton MR spectroscopy. This dissertation addresses two issues, that is the problem of low SNR and the problem of TE selection. To address SNR limitations of MRSI, we first investigate a new scheme for denoising MRSI data, incorporating both an anatomically adapted spatial-smoothness constraint and an autoregressive spectral constraint within the penalized maximum-likelihood framework. Theoretical analysis is provided to characterize the denoising performance of this approach. Results demonstrate the ability of the spectral constraint to suppress noise in non-metabolite regions. However, this nonuniform spectral denoising effect may not be visually accepted, limiting the practical use of this denoising approach. We then further propose to denoise MRSI data by exploiting low-rank properties. These are two low-rank structures of MRSI data, one due to partial separability and the other due to linear predictability of MRSI data. Denoising is performed by arranging the measured data in appropriate matrix forms (i.e., Casorati and Hankel) and applying low-rank approximations by the singular value decomposition. Experimental results obtained from in vivo MRSI data demonstrate that the combination and a particular order of the two proposed low-rank approximations provide an effective way to denoise MRSI data in the case of typical severe noise contamination. To analyze the problem of TE selection for MRS data acquisition, we reconsider this problem from an estimation theoretic perspective. Specifically, we analyze the Cram??er-Rao lower bound (CRB) on estimated spectral parameters as a function of TE, which serves as a metric to quantify the reliability of the estimation procedure. This new approach provides a quantitative method for identifying potentially useful TEs in contrast to a common heuristic choice of TE. In addition, unlike empirical studies which face practical limitations on acquisition time, the CRB analysis enables easy evaluation of an arbitrarily large range of TEs