Automatic magnetic resonance brain volume measurements to assess the effect of minimised blood sample extraction on brain development in extremely preterm infants

Sammanfattning: Aim: To evaluate a pipeline for automatic MRI brain segmentation and volumetric tissue structure quantification in a cohort of extremely preterm infants. Methods: 13 extremely preterm neonates underwent MRI-scanning at a term equivalent age. Seven of these subjects had suitable MRI sequences and were analyzed in the pipeline. Segmentation of MRI brain images into brain tissue types and parcellation of the brain into its different anatomical structures was done using Developing Brain Region Annotation With Expectation-Maximization (DRAW-EM), a Medical Image Registration Toolkit package. Volumetric quantification of the regions of interest were done on the segmented images utilizing the FMRIB Software Library (FSL), a software library containing statistical tools for image analysis of MRI brain data. Thalami, hippocampi, white matter and cortical gray matter were analyzed. Result: Six of the analyzed subjects had little or no imaging artefacts and one had severe imaging artefacts due to motion. One of the subjects with little or no artefacts had a prominent pathology in the left lateral ventricle. Segmentation in the patients with little or no imaging artefacts was overall satisfactory, with possible room for improvement in cortical gray matter segmentation based on one subject. Deep white matter was not demarcated by the program. The pipeline had difficulty segmenting the patient with ventricular pathology, but did surprisingly well with the patient with severe motion artefacts. Conclusion: The pipeline is able to analyze images without severe pathology and little to no imaging artefacts. Even with severe artefact and no motion correction code, the segmentation in itself was satisfactory. Prominent ventricular pathologies were challenging for the program to handle and may require manual intervention. Segmentation of deep white matter needs to be implemented. Further improvements are needed to address issues regarding accurate segmentation of cortical gray matter around sulci.