Presented at the Neonatal Society 2014 Spring Meeting.
Dickie DA1, Job D1, Sparrow S2, Piyasena C3, Wilkinson G4, Wardlaw J1, Boardman JP2
1 Centre for Clinical Brain Sciences, University of Edinburgh, UK
2 MRC / University of Edinburgh Centre for Reproductive Health, UK
3 Centre for Cardiovascular Science, University of Edinburgh, UK
4 Royal Hospital for Sick Children, NHS Lothian, UK
Background: Preterm birth is associated with alterations in brain structure detected by group wise comparisons with term controls using voxel based null hypothesis significance testing, e.g. t-tests. However, this approach does not describe how ‘normal’ an individual subject is. Percentile rank atlases have been developed to measure the ‘normality’ of individual subjects. Hypothesis: voxel-based ranking against an atlas created from term controls will detect alterations in individual preterm infants.
Methods: Subjects: 43 preterm VLBW infants (postmenstrual age [PMA] at birth <32 weeks’ gestation; birth weight <1500g) were recruited with approval from the NHS Research Ethics Service. Image acquisition: Siemens Verio 3T system was used to acquire T1 weighted MP-RAGE images in the sagittal plane with 1x1x1mm resolution; 13 subjects were excluded due to severe motion artefact. Processing: semi-automatic brain extraction and spatial and intensity normalisation to the Montreal Neurological Institute 36-44 weeks’ PMA normal infant atlas. Deviation from normal was calculated by % difference in each voxel: (Ui-Xij/Ui)x100, where Ui is voxel i in the atlas and Xij is voxel i in subject j. Funding body: Theirworld.
Results: The output for each subject is a fully ranked image (i) and an abnormal mask visualised in atlas space at a specified threshold (ii). In the example: (i) green regions show where the preterm infant voxel value = atlas, and yellow-red is where preterm voxel value < atlas; (ii) regions where the individual voxel is <12.5th percentile with respect to atlas. Lower voxel values indicate CSF where tissue is present in controls. Voxel-based ranking detected lateral ventricle enlargement in 30/30 preterms, with 37% showing localisation of enlargement to the temporal horns of the lateral ventricles.
Conclusion: Voxel based ranking against a term control neonatal atlas detects regional structural variation associated with preterm birth at the level of the individual. Our method and data will be made available online: http://www.sinapse.ac.uk/research-resources/brains-project
Corresponding author: ddickie1@staffmail.ed.ac.uk
References
Avants et al Med Image Anal 2008
Boardman et al NeuroImage 2006
Dickie et al PLoS ONE 2013
Fonov et al Neuroimage 2009