Presented at the Neonatal Society 2018 Summer Meeting.
Galdi P1, Blesa M1, Sullivan G1, Lamb GJ1, Stoye DQ1, Quigley AJ2, Thrippleton MJ3, Bastin ME3, Boardman JP1,3
1 MRC Centre for Reproductive Health, University of Edinburgh
2 Department of Radiology, Royal Hospital for Sick Children, Edinburgh
3 Centre for Clinical Brain Sciences, University of Edinburgh
Background: Multimodal MRI captures information about brain macro- and micro-structure, which can be combined in morphometric similarity networks to derive a detailed “fingerprint” of the anatomical properties of individual brains (1). We aimed to test the hypotheses that these fingerprints can be used to derive a data-driven metric of brain maturation, the Relative Brain Network Maturation Index (RBNMI), and that RBNMI differs between preterm infants at term equivalent age and term infants.
Methods: We combined data from different imaging sequences (diffusion and structural MRI) to extract multiple properties from cortical and sub-cortical brain regions (e.g., regional volumes, diffusion tensor-derived metrics, neurite orientation dispersion and density imaging features) which were used to construct individual morphometric similarity networks (1). A regression model was trained to predict postmenstrual age (PMA) at the time of scanning from inter-regional connections. We then derived the relative brain network maturation index (RBNMI) by measuring the difference between apparent (i.e., predicted) and actual age (2). This approach was validated on data from the Theirworld Edinburgh Birth Cohort (TEBC) and the developing Human Connectome Project (dHCP) (PMA range: 37-45 weeks). Ethical approval for use of TEBC from NRES was obtained.
Results: The best performing model in the age prediction task was based on the following features: regional volume, the ratio of T1-weighted and T2-weighted signal intensity, fractional anisotropy, axial and radial diffusivity, intracellular volume fraction, orientation dispersion index and isotropic volume fraction.
The model consistently predicted preterm infants to be younger than their actual age (the box plots on the left depict the distribution of the RBNMI in the term and preterm populations). The connections involved in age prediction (shown in the chord diagram on the right) were predominantly located in fronto-temporal and subcortical regions, posterior cingulate cortex, brain stem and cerebellum.
Conclusion: Morphometric similarity networks combined information from multiple image features to detect dysmaturity in the developing brain. The RBNMI offers a data driven and tractable metric for quantifying atypical brain development associated with preterm birth.
Acknowledgement: Part of these results were obtained using data made available from the Developing Human Connectome Project funded by the European Research Council under the European Union’s Seventh Framework Programme (FP/2007-2013) / ERC Grant Agreement no. 319456.
Corresponding author: firstname.lastname@example.org
1. Seidlitz, J., et al., Neuron 97.1 (2018): 231-247
2. Brown, C. J., et al., Proceedings of MICCAI2016 (2017): 84-91.