Elsevier

NeuroImage

Volume 173, June 2018, Pages 88-112
NeuroImage

The developing human connectome project: A minimal processing pipeline for neonatal cortical surface reconstruction

https://doi.org/10.1016/j.neuroimage.2018.01.054Get rights and content

Abstract

The Developing Human Connectome Project (dHCP) seeks to create the first 4-dimensional connectome of early life. Understanding this connectome in detail may provide insights into normal as well as abnormal patterns of brain development. Following established best practices adopted by the WU-MINN Human Connectome Project (HCP), and pioneered by FreeSurfer, the project utilises cortical surface-based processing pipelines. In this paper, we propose a fully automated processing pipeline for the structural Magnetic Resonance Imaging (MRI) of the developing neonatal brain. This proposed pipeline consists of a refined framework for cortical and sub-cortical volume segmentation, cortical surface extraction, and cortical surface inflation, which has been specifically designed to address considerable differences between adult and neonatal brains, as imaged using MRI. Using the proposed pipeline our results demonstrate that images collected from 465 subjects ranging from 28 to 45 weeks post-menstrual age (PMA) can be processed fully automatically; generating cortical surface models that are topologically correct, and correspond well with manual evaluations of tissue boundaries in 85% of cases. Results improve on state-of-the-art neonatal tissue segmentation models and significant errors were found in only 2% of cases, where these corresponded to subjects with high motion. Downstream, these surfaces will enhance comparisons of functional and diffusion MRI datasets, supporting the modelling of emerging patterns of brain connectivity.

Introduction

The period of rapid cortical expansion during fetal and early neonatal life is a crucial time over which the cortex transforms from a smooth sheet to a highly convoluted surface. During this time, the cellular foundations of our advanced cognitive abilities are mapped out, as connections start to form between distant regions (Ball et al., 2013b, Van Essen, 1997), myelinating, and later pruning, at different rates. Alongside the development of this neural infrastructure, functional brain activations start to be resolved (Doria et al., 2010), reflecting the development of cognition.

Much of what is currently known about the early human connectome has been learnt from models of preterm growth (Ball et al., 2013b, Counsell et al., 2013, Doria et al., 2010, Keunen et al., 2017). Whilst invaluable, it is known that early exposure to the extra-uterine environment has long-term implications (Ball et al., 2012, Ball et al., 2013a, Ball et al., 2017, Counsell et al., 2014, Hintz et al., 2015, Ullman et al., 2015). For this reason, the Developing Human Connectome Project (dHCP) seeks to image emerging brain connectivity for the first time in a large cohort of fetuses and newborn infants.

More broadly, the goals of the dHCP are to pioneer advances in structural, diffusion, and functional Magnetic Resonance Imaging (MRI), in order to collect high-quality imaging data for fetuses and (term/preterm-born) neonates, from both control and at-risk groups. Imaging sets will be supported by a database of clinical, behavioural, and genetic information, all made publicly available via an expandable and user-friendly informatics structure. The dataset will allow the community to explore the neurobiological mechanisms, and genetic and environmental influences, which underpin healthy cognitive development. Models of healthy development will provide a vital basis of comparison from which the effects of preterm birth, and neurological conditions such as cerebral palsy or autism, may become better understood.

The dHCP takes inspiration from the WU-MINN Human Connectome Project (HCP) (Van Essen et al., 2013). Now in its final stages, the HCP has pushed the boundaries of MRI based brain connectomics, collecting 1200 sets of healthy adult functional and structural connectomes, at high spatial and temporal resolution. Data from this project has been used to generate refined maps of adult cortical organisation (Glasser and Van Essen, 2011), and improve understanding of how the functional connectome correlates with behaviour (Smith et al., 2015).

A core tenet, underlying the success of the HCP approach, has been the advocation of surface-based processing and analysis of brain MR images. This is grounded in the understanding that distances between functionally specialised areas on the convoluted cortical sheet are more neurobiologically meaningful, when represented on the 2D surface rather than in the 3D volume (Glasser et al., 2013). Surface-based processing therefore minimises the mixing of data from opposing sulcal banks or between tissue types. Further, surface-based registration approaches (Durrleman et al., 2009, Fischl et al., 1999c, Lombaert et al., 2013, Robinson et al., 2014, Wright et al., 2015, Yeo et al., 2010) improve the alignment of cortical folds and areal features.

Unfortunately, modelling cortical connectome structure in neonates and fetuses is particularly challenging. MRI, especially functional and diffusion protocols, is highly sensitive to head motion during scanning. This is a particular issue for the dHCP, where the goal is to image un-sedated neonatal, and free-moving fetal subjects. Therefore, correcting for this has required the development of advanced scanning protocols and motion correction schemes (Cordero-Grande et al., 2018, Hughes et al., 2017, Kuklisova-Murgasova et al., 2012).

Outside of the challenges facing acquisition and reconstruction, the properties of fetal and neonatal MRI differ significantly from that of adult data. Specifically, baby and adult brains differ vastly in terms of size, with the fetal and neonatal brain covering a volume in the range of 100–600 mL in contrast to an average adult brain volume of more than 1 L (Allen et al., 2002, Orasanu et al., 2014, Makropoulos et al., 2016). Furthermore, the perinatal brain develops rapidly, which results in vast changes in scale and appearance of the brain scanned at different weeks. This, together with the fact that scanning times must be limited for the well-being and comfort of mother and baby, means that spatial and temporal resolution of the resulting images are reduced relative to adults. Furthermore, immature myelination of the white matter in neonatal and fetal brains results in inversion of MRI contrast when compared to adult brain scans (Prastawa et al., 2005). This requires image processing to be performed on T2-weighted rather than T1-weighted structural MRI.

Combined, these vast differences in image properties considerably limit the translation of conventional adult methods for image processing to fetal and neonatal cohorts. In particular, the popular FreeSurfer framework (Fischl, 2012), utilised within the HCP pipelines (Glasser et al., 2013), fails on neonatal data as it relies solely on fitting surfaces to intensity-based tissue segmentation masks (Dale et al., 1999). These pipelines have been optimised to work with adult MRI data, and are not compatible with neonatal image intensity distributions, which are significantly different and vary drastically within different weeks of development. For this reason, simple adoption of the adult HCP processing pipelines has not been possible.

Instead, this paper presents a refined surface extraction and inflation pipeline, that will accompany the first data and software release of the dHCP. This proposed framework builds upon a legacy of advances in neonatal image processing. This includes the development of specialised tools for tissue segmentation that address the difficulties in resolving tissue boundaries blurred through the presence of low resolution and partial volume. A variety of techniques have been proposed for tissue segmentation of the neonatal brain in recent years: unsupervised techniques (Gui et al., 2012), atlas fusion techniques (Weisenfeld and Warfield, 2009, Gousias et al., 2013, Kim et al., 2016), parametric techniques (Prastawa et al., 2005, Song et al., 2007, Xue et al., 2007, Shi et al., 2010, Cardoso et al., 2013, Makropoulos et al., 2012, Wang et al., 2012, Wu and Avants, 2012, Beare et al., 2016, Liu et al., 2016), classification techniques (Anbeek et al., 2008, Srhoj-Egekher et al., 2012, Chiţă et al., 2013, Wang et al., 2015, Sanroma et al., 2016, Moeskops et al., 2016) and deformable models (Wang et al., 2011, Wang et al., 2013, Dai et al., 2013, Wang et al., 2014). A review of neonatal segmentation methods can be found in Devi et al., 2015, Makropoulos et al., 2017. The majority of these techniques have been applied to images with a lower resolution than those acquired within the dHCP, and typically to images of preterm-born subjects.

Once segmentations are extracted, surface mesh modelling approaches are, to an extent, agnostic of the origin of the data; allowing, in principle, the application of a wide variety of cortical mesh modelling approaches to neonatal data, including those provided within the FreeSurfer (Dale et al., 1999, Fischl, 2012), BrainSuite (Shattuck and Leahy, 2002), BrainVISA (Rivière et al., 2009), and CIVET (MacDonald et al., 2000, Kim et al., 2005, Kim et al., 2016) packages. In general, these methods fit surfaces to boundaries of tissue segmentation masks, which, allowing for some need for topological correction, relies on the accuracy of the segmentation. In neonatal imaging data, the use of T2 images however leads to segmentation errors not seen in adult data. This is the misclassification of CSF as white matter, caused by the fact that CSF and white matter appear bright in neonatal T2 images, whereas in adult T1 data white matter is bright and CSF is dark. If not fully accounted for during segmentation, these errors will be propagated through to surface reconstruction (Xue et al., 2007).

In what follows, we present a summary of our proposed pipeline. This brings together existing tools for neonatal segmentation (refined to minimise the propagation of misclassification errors through to surface extraction) with new tools for cortical extraction that combine information from segmentation masks and T2-weighted image intensities, in order to compensate for the effects of partial volume and improve correspondence with true tissue boundaries. Re-implementations of existing tools for surface inflation and projection to the sphere are provided to minimise software overhead for users.

The processing steps are as follows: 1) acquisition and reconstruction of T1 and T2 images (Cordero-Grande et al., 2018, Hughes et al., 2017, Kuklisova-Murgasova et al., 2012); 2) tissue segmentation and regional labelling (Makropoulos et al., 2012, Makropoulos et al., 2014, Makropoulos et al., 2016); 3) cortical white and pial surface extraction (Schuh et al., 2017); 4) inflation and projection to a sphere (for use with spherical alignment approaches) (Fischl et al., 1999a, Elad et al., 2005); and 5) definition of cortical feature descriptors, including descriptors of cortical geometry and myelination (Glasser et al., 2013). Manual quality control is performed by two independent expert raters to assess the quality of the acquired images, segmentations, and reconstructed cortical surfaces. Assessment of these data sets shows, that with very few exceptions (3%) the protocol is able to extract cortical surfaces which fit closely the expectations for observed anatomy.

Section snippets

Project overview

The goal of the dHCP is to create a dynamic map of human brain connectivity from 20 to 45 weeks post-conceptional age (PMA) from healthy, term-born neonates, infants born prematurely (prior to 36 weeks PMA), and fetuses. The infants are being scanned using optimised protocols for structural (T1 and T2-weighted) images, resting state functional MRI (fMRI), and multi-shell High Angular Resolution Diffusion Imaging (HARDI). Imaging data will be combined with genetic, cognitive, and environmental

The neonatal structural pipeline

The first stage of the project has been to optimise acquisition protocols and collect data for the neonatal cohort (Cordero-Grande et al., 2018, Hughes et al., 2017, Kuklisova-Murgasova et al., 2012). The methods in this paper therefore reflect neonatal structural processing protocols, and are designed to accompany the first data release of neonatal subjects.

The workflow of the neonatal processing pipeline is summarised in Fig. 1. The motion-corrected, reconstructed T2-weighted image is first

Quality control (QC)

The quality of the pipeline was assessed by manually scoring a sub-set of randomly selected images from the cohort. We performed three independent scorings for the different parts of the pipeline: image reconstruction, segmentation and surface reconstruction. 160 images were used for the image reconstruction and segmentation QC. A sub-set of 43 images was then used for the surface reconstruction QC due to the more manually intensive and time consuming scoring of cortical details. All images and

Comparison to HCP pipelines

The dHCP pipeline has been inspired by the HCP (Glasser et al., 2013). However, there are several key differences between the pipelines, as highlighted in Table 3. Neonatal subjects are imaged during natural sleep. Therefore, total scanning time for all structural, diffusion and rfMRI scans is markedly reduced relative to the HCP. Furthermore, due to concerns about the effects of motion, scans are acquired in stacks that must be reconstructed and motion corrected prior to analysis. Finally, T1

Discussion

This paper presents a fully automatic pipeline for brain tissue segmentation and cortical surface modelling of neonatal MRI. All methods have been tuned on images collected using the dHCP protocol, and take advantage of improvements in image quality (gained from advances in acquisition, reconstruction and motion correction (Cordero-Grande et al., 2018, Hughes et al., 2017, Kuklisova-Murgasova et al., 2012)) to offer topologically correct (genus-0) surface representations of images acquired

Acknowledgements

The research leading to these results has received funding from the European Research Council under the European Unions Seventh Framework Programme (FP/2007–2013)/ERC Grant Agreement no. 319456. We are grateful to the families who generously supported this trial. The work was supported by the NIHR Biomedical Research Center at Guys and St. Thomas NHS Trust.

References (134)

  • B. Fischl

    FreeSurfer

    Neuroimage

    (2012)
  • B. Fischl et al.

    Cortical surface-based analysis II: inflation, flattening, and a surface-based coordinate system

    Neuroimage

    (1999)
  • B. Fischl et al.

    Cortical surface-based analysis II: inflation, flattening, and a surface-based coordinate system

    Neuroimage

    (1999)
  • H. Glasel et al.

    A robust cerebral asymmetry in the infant brain: the rightward superior temporal sulcus

    Neuroimage

    (2011)
  • M.F. Glasser et al.

    The minimal preprocessing pipelines for the human connectome project

    Neuroimage

    (2013)
  • I.S. Gousias et al.

    Magnetic resonance imaging of the newborn brain: manual segmentation of labelled atlases in term-born and preterm infants

    Neuroimage

    (September 2012)
  • D.N. Greve et al.

    Accurate and robust brain image alignment using boundary-based registration

    Neuroimage

    (2009)
  • L. Gui et al.

    Morphology-driven automatic segmentation of MR images of the neonatal brain

    Med. Image Anal.

    (December 2012)
  • I. Išgum et al.

    Evaluation of automatic neonatal brain segmentation algorithms: the NeoBrainS12 challenge

    Med. Image Anal.

    (February 2015)
  • H. Kim et al.

    Neocivet: towards accurate morphometry of neonatal gyrification and clinical applications in preterm newborns

    Neuroimage

    (September 2016)
  • J.S. Kim et al.

    Automated 3-D extraction and evaluation of the inner and outer cortical surfaces using a Laplacian map and partial volume effect classification

    Neuroimage

    (August 2005)
  • M. Kuklisova-Murgasova et al.

    Reconstruction of fetal brain mri with intensity matching and complete outlier removal

    Med. Image Anal.

    (2012)
  • C. Laule et al.

    Myelin water imaging of multiple sclerosis at 7 t: correlations with histopathology

    Neuroimage

    (2008)
  • G. Li et al.

    Measuring the dynamic longitudinal cortex development in infants by reconstruction of temporally consistent cortical surfaces

    Neuroimage

    (2014)
  • G. Li et al.

    Consistent reconstruction of cortical surfaces from longitudinal brain mr images

    Neuroimage

    (2012)
  • G. Li et al.

    Construction of 4d high-definition cortical surface atlases of infants: methods and applications

    Med. Image Anal.

    (October 2015)
  • G. Li et al.

    Simultaneous and consistent labeling of longitudinal dynamic developing cortical surfaces in infants

    Med. Image Anal.

    (December 2014)
  • M. Liu et al.

    Patch-based augmentation of Expectation-Maximization for brain MRI tissue segmentation at arbitrary age after premature birth

    Neuroimage

    (February 2016)
  • D. MacDonald et al.

    Automated 3-D extraction of inner and outer surfaces of cerebral cortex from MRI

    Neuroimage

    (September 2000)
  • A. Makropoulos et al.

    Regional growth and atlasing of the developing human brain

    Neuroimage

    (January 2016)
  • Y. Meng et al.

    Spatial distribution and longitudinal development of deep cortical sulcal landmarks in infants

    Neuroimage

    (October 2014)
  • E. Orasanu et al.

    Brain volume estimation from post-mortem newborn and fetal MRI

    Neuroimage: Clin.

    (October 2014)
  • S. Parisot et al.

    Group-wise parcellation of the cortex through multi-scale spectral clustering

    Neuroimage

    (2016)
  • B. Alexander et al.

    A new neonatal cortical and subcortical brain atlas: the melbourne children's regional infant brain (m-crib) atlas

    Neuroimage

    (2016)
  • J.S. Allen et al.

    Normal neuroanatomical variation in the human brain: an MRI-volumetric study

    Am. J. Phys. Anthropol.

    (August 2002)
  • P. Anbeek et al.

    Probabilistic brain tissue segmentation in neonatal magnetic resonance imaging

    Pediatr. Res.

    (Feb. 2008)
  • S. Arslan et al.

    Joint spectral decomposition for the parcellation of the human cerebral cortex using resting-state fmri

  • X. Artaechevarria et al.

    Combination strategies in multi-atlas image segmentation: application to brain MR data

    IEEE Trans. Med. Imag.

    (August 2009)
  • G. Ball et al.

    Multimodal image analysis of clinical influences on preterm brain development

    Ann. Neurol.

    (2017)
  • G. Ball et al.

    The effect of preterm birth on thalamic and cortical development

    Cerebr. Cortex

    (2012)
  • G. Ball et al.

    Development of cortical microstructure in the preterm human brain

    Proc. Natl. Acad. Sci. Unit. States Am.

    (2013)
  • A.J. Barkovich et al.

    Normal maturation of the neonatal and infant brain: MR imaging at 1.5 T

    Radiology

    (Jan. 1988)
  • R.J. Beare et al.

    Neonatal brain tissue classification with morphological adaptation and unified segmentation

    Front. Neuroinf.

    (2016)
  • J. Bozek et al.

    In vivo cortical myelination of the neonatal brain in the developing human connectome project

  • J. Bozek et al.

    Construction of a neonatal cortical surface atlas using multimodal surface matching

  • J. Bozek et al.

    Construction of a Neonatal Cortical Surface Atlas Using Multimodal Surface Matching in the Developing Human Connectome Project

    (2017)
  • S.M. Chiţă et al.

    Automatic segmentation of the preterm neonatal brain with MRI using supervised classification

  • L. Cordero-Grande et al.

    Three-dimensional motion corrected sensitivity encoding reconstruction for multi-shot multi-slice MRI: application to neonatal brain imaging

    Magn. Reson. Med.

    (2018)
  • S. Counsell et al.

    Diffusion imaging in the developing brain

    Diffusion MRI: From Quantitative Measurement to In vivo Neuroanatomy

    (2013)
  • S.J. Counsell et al.

    New imaging approaches to evaluate newborn brain injury and their role in predicting developmental disorders

    Curr. Opin. Neurol.

    (2014)
  • Cited by (0)

    1

    These authors contributed equally.

    View full text