| The rapid development of magnetic resonance spectroscopy (MRS) greatly facilitates non-invasive measurement of brain metabolites, which makes it a versatile diagnostic procedure for biomedical research. The validity and dependability of MRS data relies on the accuracy and efficiency of data post-processing and quantification analysis. Throughout the years, various quantification methods have been proposed and implemented in both 31P and 1H spectrum analysis. The frequency variation of certain chemical compounds of interest and serious baseline distortions remain the primary challenges for post-processing in large volume in vivo 31P MRS. This work aims to undertake these problems by developing a Hankel Singular Value Decomposition (HSVD) based adaptive prior knowledge algorithm that can intelligently guide itself to an optimal result. This algorithm uses so called interference signals to optimize prior knowledge iteratively for parameter optimization. The purpose of this approach is to improve the quantification quality of MRS signals from different brain locations as well as from different experimental environments. To achieve this goal, we developed an algorithm termed Iterative Reduction of Interference Signal - HSVD (IRIS-HSVD). The Monte Carlo evaluations of the algorithm were conducted with simulated data using relevant in vivo parameters. The performance of this algorithm was compared to those of other automatic methods including HSVD and HTLS-PK. Examples of in vivo 31P data obtained from brains of healthy subjects on a 4T MRI scanner were also presented, which demonstrated the superiority of the new method as compared to AMARES, a widely used program in the NMR community. |
