Meeting Banner
Abstract #3169

Automatic Brain MRI Segmentation in Very Preterm Infants

Lili He1, Nehal A. Parikh1

1Center for Perinatal Research, Nationwide Children's Hospital, Columbus, OH, United States

Premature birth and secondary perinatal-neonatal insults profoundly alter brain development and functional outcomes. Accurate gray and white matter tissue quantification of neonatal brain MRI scans can enhance our understanding of this developmental trajectory and serve as diagnostic and prognostic biomarkers. However, accurate quantification remains an extremely challenging task in very pdue to a combination of factors, including lower image contrast (due to incomplete myelination at this isointense stage), lower signal-to-noise ratio (shorter scan times), greater motion artifacts, and lower spatial resolution (smaller head size) as compared to adult brain scans. In this paper, we present a fully automated and computational efficient spatial fuzzy segmentation algorithm, which we first proposed to detect activation regions in functional MRI. The method is validated both qualitatively and quantitatively on simulated and in-vivo extremely low birth weight (ELBW; BW 1000g) infant brain MRI scans at term-equivalent age.

Abstract PDF

Keywords

accuracy accurate activation agreement anatomical approved artifacts atlas auto automated automatic axial balance belonging birth board brain certain challenge class combination components computational computationally conducted consent considered considering constraints construction containing contrast control corrected currently defined describe determined develop development deviation diagnostic dice driven dual efficient eliminate enhance evaluate expect expert extra extremely fields five framework full fully function functional fuzzy gold gray greater ground head heuristically hospital imposing improve in vivo incomplete index infant infants influenced inhomogeneities institutional insults intensive interactions interactive interchanged interclass intra joint lack local manner manual matrix medical membership model motion mutual nationwide need needed neighborhood neighboring neonatal newborn noise operator outcomes paper perinatal potential predominantly prior probabilities profoundly prognostic programs proposed qualitatively quality quantification random real realizations relatively resolution rows segmentation segmentations segmented semi serve sets similarity simulate simulated spatial spin statistical subjective task theory tissue training trajectory transactions truth understanding usually validated validation views white whose