The developing Human Connectome Project (dHCP) automated resting-state functional processing framework for newborn infants

Abstract

The developing Human Connectome Project (dHCP) aims to create a detailed 4-dimensional connectome of early life spanning 20-45 weeks post-menstrual age. This is being achieved through the acquisition of multi-modal MRI data from over 1000 in- and ex-utero subjects combined with the development of optimised pre-processing pipelines. In this paper we present an automated and robust pipeline to minimally pre-process highly confounded neonatal resting-state fMRI data, robustly, with low failure rates and high quality-assurance. The pipeline has been designed to specifically address the challenges that neonatal data presents including low and variable contrast and high levels of head motion. We provide a detailed description and evaluation of the pipeline which includes integrated slice-to-volume motion correction and dynamic susceptibility distortion correction, a robust multimodal registration approach, bespoke ICA-based denoising, and an automated QC framework. We assess these components on a large cohort of dHCP subjects and demonstrate that processing refinements integrated into the pipeline provide substantial reduction in movement related distortions, resulting in significant improvements in SNR, and detection of high quality RSNs from neonates.

Keywords: Connectome; Developing Human Connectome Project; Functional MRI; Neonate; Pipeline; Quality control.

Fig. 1

Joint distributions of post-menstrual age-at-birth (weeks) and post-menstrual age-at-scan (weeks) for the dHCP-538 (left) and dHCP-40 (right) cohorts.

Fig. 2

Schematic of the dHCP fMRI neonatal pre-processing pipeline. The schematic is segregated into the 4 main conceptual processing stages by coloured background; fieldmap pre-processing (red), susceptibility and motion correction (orange), registration (green), and denoising (purple). Inputs to the pipeline are grouped in the top row, and the main pipeline outputs are grouped in the lower right. Blue filled rectangles with rounded corners indicate processing steps, whilst black rectangles (with no fill) represent data. The critical path is denoted by magenta connector arrows. (dc) = distortion corrected; (mcdc) = motion and distortion corrected.

Fig. 3

Distribution of the z-scored normalised mutual information between the source image and the reference image (both in reference space) for each of the primary registration stages fieldmap-to-structural, functional-to-sbref (distorted), functional-to-sbref (undistorted), sbref-to-structural, and template-to-structural. More positive NMI z-scores indicate more similarity and more negative NMI z-scores indicate less similarity.

Fig. 4

Examples of fieldmap, sbref, and template images resampled to the native structural reference space. The outline of the native structural white matter discrete segmentation is overlaid in green. Examples were selected at the 5th, 50th and 95th percentile of normalised mutual information between the source image and the reference image (both in reference space). Note: the 5th percentile fieldmap is dual-echo-time-derived and therefore lacking tissue contrast, whilst the 50th and 95th percentile fieldmaps are spin-echo-EPI-derived.

Fig. 5

Upper: 40-week T2w and T1w dHCP templates. Middle: group mean and standard deviation (N = 512) of structural T2w in template space. Lower: group mean and standard deviation (N = 512) of functional (mean) in template space.

Fig. 6

Exemplar single-volume of an EPI from a single-subject with intra-volume movement contamination from a left-right head movement (upper) and a front-back head movement (lower), before (Raw) and after motion and susceptibility distortion correction (MCDC), and after FIX denoising (Denoised).

Fig. 7

Five exemplar volumes of an EPI from a single-subject with susceptibility-by-movement distortion due to head motion. The rigid data in the top row have been rigid-body motion corrected, and anterior distortions can be observed in volumes 1154 and 1156 where the front of the brain extends beyond the reference line (green-dashed line). The anterior distortions are diminished after motion and susceptibility distortion correction (MCDC), and more so after denoising.

Fig. 8

Left: mean tSNR (N = 40) for raw EPI (RAW), rigid-body motion correction (RIGID), slice-to-volume motion correction (S2V), and S2V + susceptibility-by-movement distortion correction (S2V+MBS). Centre and right: difference maps and t-statistics for RIGID tSNR minus RAW tSNR (upper), S2V tSNR minus RIGID tSNR (middle) and S2V+MBS tSNR minus S2V tSNR (lower). Only significant results shown. Multiple comparison correction was achieved by FDR correction with a 1.67% threshold (5% divided by the number of tests). The slice coordinates for the difference maps and t-statistic maps were selected by the maximum t-statistic.

Fig. 9

Exemplar spatial maps (left), time-courses (centre), and power spectra (right) for independent components (IC) from a single subject. Each row is a different IC that was manually classified as stereotypical for signal, multi-band artefact, head movement, arteries, sagittal sinus, and unclassified noise. Framewise displacement is plotted in the last row as a reference for the amount and timing of movement for this subject.

Fig. 10

Distribution across decompositions of percentage of components (per decomposition) classified as signal or noise by FIX.

Fig. 11

Correlation of the number of ICs classified by FIX as signal (left) and noise (right) with head movement, where mean framewise displacement is used as a surrogate for motion contamination. Age is the post-menstrual age-at-scan in weeks.

Fig. 12

Left: mean tSNR (N = 512) for raw EPI (RAW), motion and distortion corrected EPI (MCDC), and denoised EPI. Centre and right: difference maps and t-statistics for MCDC minus RAW tSNR (upper), and denoised minus MCDC tSNR (lower). Only significant results shown. Multiple comparison correction was achieved by FDR correction with a 2.5% threshold (5% divided by the number of tests).

Fig. 13

Unbiased group RSN template maps created from MCDC and Denoised data.

Fig. 14

Mean paired-difference (Denoised-MCDC, and MCDC-Raw) of spatial similarity to the unbiased group template per map. Asterisks indicate significant differences. (Lower) Distribution of paired differences (Denoised-MCDC, and MCDC-Raw) of spatial similarity to the unbiased group template pooled over all spatial maps.

Fig. 15

Distribution of paired differences (Denoised-MCDC, and MCDC-Raw) of network matrix similarity to the unbiased group network matrix.

Fig. 16

Voxplots for raw, motion and distortion corrected (MCDC), and denoised fMRI from a single exemplar subject. Voxplots are adapted from (Power, 2017) with the modification that each heat-map is converted to a z-score and a diverging colormap is used. Mean and trend were removed from each heat-map. GM=grey matter, WM=white matter, SC=sub-cortical, CB=cerebellum, BS=brainstem.

Fig. 17

Group z-distributions of QC metrics. More negative z-scores indicate poorer quality on the respective metric. Z-scores less than -2.5 (indicated by red dashed line) are flagged as failing the pipeline and require further inspection.

Fig. 18

PROFUMO modes qualitatively assessed as corresponding to adult resting-state networks. Hierarchical clustering based on spatial correlation between the modes. Warm colours are positive, and cool colours are negative.