Automatic Quantification Of In Vivo Cerebral~1H CSI Data

As a multi-voxel magnetic resonance spectroscopic imaging technique, chemicalshift imaging (CSI) can provide the concentrations of some important metabolites inbiological tissues. With the development of nuclear magnetic resonance (NMR) tech-nology, CSI will be applied more broadly in the future. At present,1H CSI has beenapplied to clinical diagnosis of brain diseases. However, much more efort should bemade to achieve accurate quantification of the metabolites from in vivo cerebral1HMRS data, because the bio-magnetic phenomena are very complex in brain tissues,making the spectra very complex. What’s more, the raw data are very possible badquality due to some nonideal acquisition conditions, such as the field inhomogeneity,the body movement and so on. Therefore, in vivo cerebral1H MRS data should bepreprocessed a lot before their quantification and analysis. The existing processingmethods regard the CSI data as some individual single-voxel spectra, ignoring theirmulti-voxel features. This dissertation presents a whole automatic quantification sys-tem which can achieve stable and meaningful results by employing prior knowledgeand multi-voxel information. Many pre-processing methods are studied in this dis-sertation, including apodization, residual water removal, phase and baseline correc-tion, and so on. Among them, the peak removal method with Hankel Lanczos sin-gular value decomposition (HLSVD) has been improved, and a new frequency adjustmethod using multi-voxel information has been proposed, and a criterion based on thevariances of fitting baselines has been set up to estimate whether the pre-processing re-sults are acceptable. With these pre-processing methods, an automatic pre-processingflow is designed, which can result in very good spectra with flat baselines, accurate fre-quency coordinates, good SNR and spectral resolution, and with water signal removedcompletely. The good pre-processing results contribute to the right quantification andanalysis. As a time-domain computationally efcient algorithm, HLSVD is chosen tomake automatic quantification. In order to deal with the drawback of the black-box algorithm, a post-processing approach is proposed to achieve physically meaningfulquantification results that are necessary for the following automatic data analysis. Theauthor also studied to use the peak values of absolute spectrum instead of the area val-ues of real spectrum to quantify metabolites’ peaks, and the result is that peak valuesare more stable for in vivo cerebral1H CSI data acquired in current conditions. Alot of experiments with phantom data and in vivo data have been done to estimate thewhole automatic quantification system, and the results prove that it can provide stablerelative concentrations of some interesting metabolites. This automatic quantificationsystem for in vivo cerebral1H CSI data can be used to help clinical diagnosis of somecerebral diseases, and it is especially potential for preoperative diagnosis of glioma,improvement of tumor tissue segmentation, and assessment of treatment.