| Quantitative analysis of premature neonatal brain MRI is essential for studying early human brain development, quantication of brain injury and its impact on early postnatal neurodevelopment. An accurate automatic delineation of the brain structures in the MRI scan remains the first step of any morphological analysis. In such studies, scans are usually acquired at any arbitrary gestational age during a rapid anatomical growth period, and with a limited imaging time. Due to the inter-subject anatomical variations and limited image quality, it is particularly challenging to accurately and automatically segment the tissue structures in such data. The objective of this work was to develop algorithmic tools that enable accurate automatic tissue segmentation and quantitative analysis of premature neonatal brain MRI scans. Multiple methods such as combining atlas-based and patchbased method in two ways for normal brain tissue segmentation, as well as combining spatial and non-spatial dictionary learning for automatic focal lesion labeling were developed and validated to show improved segmentation accuracy. The methodology developed in this work has been used for quantitative image analysis in multiple multi-site clinical studies on brain development after preterm births. |
