The frequency of premature birth has risen to almost one in every eight births. Although most preterm infants survive, there is growing evidence of adverse neurodevelopmental outcomes due to alterations in brain structure and function. However, the relationship between morphometric changes and final neurological outcome is unknown. This project aims to locate specific structural bio-markers of delayed or decreased brain folding that correlate with decreased neurocognitive abilities, by implementing different quantifications of gyrification. Gyrification, or curvature of the brain, has been identified as a promising structural marker for neurodevelopment. Curvature measurements are heavily dependent on size of the surface which drastically increases during this brief developmental time period; therefore our framework includes normalization and ratio based measures for improved consistency. Our image analysis framework proposes a novel atlas-based automatic tissue segmentation which utilizes age-specific tissue probability maps to serve as a source of spatial priors. We propose to adapt an atlas-based segmentation technique from fetal to preterm analysis with the addition of lobe segmentations that increase the precision of segmentation and registration. This will allow for brain surface extraction by tessellation of tissue maps to reconstruct topology correct representations of the inner and outer cortical surfaces, in which our unique vertex-wise gyrification analysis will be mapped onto a population-average surface by volumetric unbiased template-free groupwise registration. Statistical modeling will be performed for each vertex to detect local and regional patterns of folding and asymmetry. This project has the opportunity to directly relate these measures of early brain folding against later measures of developmental outcomes derived from the Bayley's Scale of Infant Development (BSID) with the goal of predicting neurocognitive outcomes. Achieving this goal will allow for the possibility of earlier clinical diagnostics permitting timely neurodevelopmental interventions.

The application of Magnetic Resonance Imaging (MRI) in medicine and research has seen a stable growth in magnetic field strength. It is therefore essential to determine if motion correction algorithms developed for low field MRI show comparable results for higher field MRI, by observing quantifiable parameters like signal and contrast to noise ratios (SNR and CNR). The goal of this research study is to study the variations in signal-to-noise ratio (SNR) in 3 Tesla and 1.5 Tesla fetal brain Magnetic Resonance Images for slice-intersection based motion-correction (SIMC) algorithms. The SIMC approach aligns the individual MRI slices by considering slice-to-slice matching directly and improves the final image alignment accuracy, thereby reducing motion related image artifacts. In order to perform suitable analysis, SNR is first measured in raw MRI data by evaluating the signal levels in transient tissue regions of the brain against background noise and then compared with the SNR for motion-corrected data in 3T and 1.5T fetal brains. Grey and white matter regions from adult brains are used as a reference to determine the transient tissue regions of interest. Since a higher SNR is produced for 3T in comparison to 1.5T fetal brain MRI for raw data, we expect significant increase in the SNR values for motion-corrected 3T data as well. We also expect that the SIMC method significantly improves motion-related image artifacts, while preserving the image quality for high-field data. The results from this study can help establish 3T field strength as one suitable for imaging fetal brains, due to higher signal and higher image contrast. Moreover, it can help ascertain the efficacy of the SIMC method for a range of MRI field strengths.

Recent techniques now allow us to study the developing fetal brain anatomy in 3D from the reconstruction of motion corrected sets of multiple 2D images. The use of iterative, as opposed to interpolation based, reconstruction to form a 3D image from the slice data can provide improved anatomical detail. However, the lack of strategy on deciding the regularization parameter has prevented us from getting the optimal result. Here we implemented the L-curve method, which took into account of the trade-off between the fit and the data error, on choosing the optimal regularization parameter. We reviewed the 3D reconstruction quality from clinically acquired data through the examination of tissue contrast and spatial resolution and compared the results against interpolation methods. In this study, we analyzed a series of T2-weighted MRI scans from the developing fetuses and focused on the main fetal brain sectors –cortical plate (CP), deep gray nuclei (DG), germinal matrix (GMAT), subplate and intermediate zone (SP+IZ), and ventricles (VENT). Through these evaluations, we were able to obtain the optimal iterative reconstructed images that work best with different numbers of image stacks, image resolution, image signal strength as well as presenting clear anatomical details within critical brain regions.