Improved segmentation of neonatal brain MRI scans by addressing motion artifacts with data interpolation

Verschuur, A, Boswinkel, V, Tax, Chantal ORCID: https://orcid.org/0000-0002-7480-8817, van Osch, J, Nijholt, I, Slump, C, de Vries, L, van Wezel-Meijler, G, Leemans, A and Boomsma, M 2022. Improved segmentation of neonatal brain MRI scans by addressing motion artifacts with data interpolation. Presented at: Proceedings of the 13th International Newborn Brain Conference: Neuro-imaging studies, 10-12/02/2022. Journal of Neonatal-Perinatal Medicine. IOS Press, 10.3233/NPM-229001

PDF - Published Version Available under License Creative Commons Attribution. Download (6MB)

Abstract

negatively affect segmentation. The purpose of this study was to investigate whether motion-affected slices can be replaced by interpolated slices to enhance segmentation of neonatal brain MRI scans. METHODS: From August 2017 to November 2019, moderate-late preterm infants were enrolled in a prospective cohort study entitled Brain Imaging in Moderate-late Preterm infants (BIMP-study). Around term equivalent age, MRI of the brain was performed using a 3 Tesla MRI. T2-weighed (voxel size 0.35x0.35x2mm) transverse images were automatically segmented into eight brain structures with a neonatal segmentation toolbox [1]. Upon visual inspection, scans with motion artifacts that affected segmentation (25/112; motion group) and scans without motion artifacts (27/112; control group) were selected and used for analysis. Slices with motion artifacts were re-estimated using shape-preserving cubic spline interpolation [2, 3], followed by automatic segmentation of the interpolated scan. Analysis was performed in three stages. Firstly, scans from the control group were used to test interpolation reliability: 18/54 axial slices of these scans were interpolated. Segmentation results of uninterpolated and interpolated scans were compared using the Sørensen-Dice coefficient. Secondly, uninterpolated and interpolated volumes of the motion group were compared using the Wilcoxon Signed-Ranks test. Thirdly, interpolated volumes of the motion group were compared to uninterpolated volumes of the control group using the Mann-Whitney U test. RESULTS: In the control group, Sørensen-Dice coefficients ranged between 0.87 and 0.97. In the motion group, interpolation resulted in a significant decrease of cortical (Z=-2.9, p=0.004) and deep gray matter (Z=-3.30, p<0.001), and a significant increase of white matter (Z=2.84, p=0.005) volumes. No significant differences were found between interpolated volumes of the motion group and uninterpolated volumes of the control group. CONCLUSION: Shape preserving cubic spline interpolation enables reliable segmentation of motion-affected MRI scans in moderate-late preterm infants.

Item Type:	Conference or Workshop Item (UNSPECIFIED)
Date Type:	Published Online
Status:	Published
Schools:	Physics and Astronomy
Publisher:	IOS Press
ISSN:	1934-5798
Date of First Compliant Deposit:	26 October 2022
Date of Acceptance:	12 April 2022
Last Modified:	26 Oct 2022 16:00
URI:	https://orca.cardiff.ac.uk/id/eprint/153507

Actions (repository staff only)

Edit Item