Multi-kernel support vector machine and Levenberg-Marquardt classification approach for neonatal brain MR images

Abstract

This paper focuses on the development of an accurate neonatal brain MRI segmentation algorithm and its clinical application to characterize normal brain growth and explore the neuro-anatomical correlates of cognitive impairments. The segmentation of MR images of the neonatal brain is a fundamental step in the study and assessment of infant brain development. The highest level of development techniques for adult brain MRI segmentation are not suitable for neonatal brain, because of substantial contrasts in structure and tissue properties between newborn and fully developed brains. Current newborn brain MRI segmentation techniques either depend on manual interaction or require the utilization of atlases or templates, which unavoidably presents a bias of the results towards the population that was utilized to derive the atlases. In this paper, we proposed an atlas-free approach for the segmentation of neonatal brain MRI, based on the neural network approach. The segmentation is primary stage to obtain quantitative analysis of regional brain tissues. These measurements allow characterization of the regional brain growth and inspect the correlations with clinical factors.