Development of software tools for improved 1H magnetic resonance spectroscopic imaging

Proton magnetic resonance spectroscopic imaging (1H-MRSI) provides a non-invasive, spatially resolved evaluation of brain metabolism. However, there are some limitations of 1H-MRSI preventing its wider use in the clinics, including the spectral quality issues, partial volume effect, chemical shift artifact, and low spatial resolution. Additionally, it is necessary to create metabolite maps for analyzing spectral data along with other MRI modalities. In the first part of this study, a MATLAB-based open-source data analysis software for three-dimensional 1H-MRSI, called Oryx-MRSI, which includes modules for visualization of raw 1H-MRSI data and LCModel outputs, chemical shift correction, tissue fraction calculation, metabolite map production, and registration onto standard MNI152 brain atlas while providing automatic spectral quality control, is presented. Oryx-MRSI implements region of interest analysis at brain parcellations defined on MNI152 brain atlas. All generated metabolite maps are stored in NIfTI format. Oryx- MRSI is publicly available at Oryx-MRSI link along with six example datasets. In the second part of this study, we investigated metabolic changes of mild cognitive impairment in Parkinson's disease (PD-MCI) using 1H-MRSI data analyzed with Oryx-MRSI. Sixteen healthy controls (HC), 26 cognitively normal Parkinson's disease (PD-CN) patients, and 34 PD-MCI patients were scanned in this prospective study. Neuropsychological tests were performed, and three-dimensional 1H-MRSI was obtained at 3T. Metabolic parameters and neuropsychological test scores were compared between PD-MCI, PD-CN, and HC. The correlations between neuropsychological test scores and metabolic intensities were also assessed. Supervised machine learning algorithms were applied to classify HC, PD-CN, and PD-MCI groups based on metabolite levels. PD-MCI had a lower corrected total N-acetylaspartate over total creatine ratio (tNAA/tCr) in the right precentral gyrus, corresponding to the sensorimotor network (p = 0.01), and a lower tNAA over myoinositol ratio (tNAA/mI) at a part of the default mode network, corresponding to the retrosplenial cortex (p = 0.04) than PD-CN. The HC and PD-MCI patients were classified with an accuracy of 86.4% (sensitivity = 72.7% and specificity = 81.8%) using bagged trees. 1H-MRSI revealed metabolic changes in the default mode, ventral attention/salience, and sensorimotor networks of PD-MCI patients, which could be summarized mainly as 'posterior cortical metabolic changes' related with cognitive dysfunction. In the last part of this thesis, low spatial resolution problem of 1H-MRSI was tackled. The application of super resolution convolutional neural networks (SRCNN) and Enhanced Deep Residual Networks for Single Image Super-Resolution (EDSR) for increasing the spatial resolution of 1H-MRSI were investigated. Fluid attenuated inversion recovery (FLAIR), T1-weighted and T2-weighted magnetic resonance imaging (MRI) data and a fused MRI, which contained the three different structural MR images in each RGB channel, were used in training the SRCNN scheme. Additionally, T1- weighted MRIs were used for training the EDSR network. The spatial resolution of 1H-MRSI images were increased using both SRCNN and EDSR models trained with the anatomical MR images. The results of the proposed technique were compared with bicubic interpolation in terms of peak signal to noise ratio. The spectral pattern of the super resolved images were similar to that of the original high resolution MRI. Our results indicated that deep learning based super resolution models would contribute to reconstructing higher resolution 1H-MRSI. This thesis contributed to the literature in terms of developing an open-source MRSI software called Oryx-MRSI, which provides an unprecedented detailed data anal- ysis pipeline for 1H-MRSI, identifying metabolic correlates of PD-MCI, which might aid the clinicians for the diagnosis of MCI, and implementing deep learning based super resolution approaches that might increase the spatial resolution of 1H-MRSI without a high cost of scan time.