Fetal developing Human Connectome Project functional MRI data release: methods and data structures.
Presented During:
Tuesday, June 25, 2024: 12:00 PM - 1:15 PM
COEX
Room:
Grand Ballroom 103
Poster No:
2219
Submission Type:
Abstract Submission
Authors:
Vyacheslav Karolis1,2, Lucilio Cordero-Grande3,1, Anthony Price4, Emer Hughes4, Vanessa Kyriakopoulou1, Alena Uus5, Sean Fitzgibbon6, Seyedeh-Rezvan Farahibozorg7, Tomoki Arichi1,8, Mary Rutherford9, Eugene Duff10,2, Daniel Rueckert11,8, A. Edwards4, Stephen Smith12, Joseph Hajnal4
Institutions:
1King's College London, London, United Kingdom, 2FMRIB, University of Oxford, Oxford, United Kingdom, 3Universidad Politécnica de Madrid & CIBER-BBN, Madrid, Spain, 4King's College London, London, England, 5King's College London, London, Other, 6FMRIB, University of Oxford, Oxford, Oxfordshire, 7FMRIB, University of Oxford, Oxford, Non-US/Other, 8Imperial College London, London, United Kingdom, 9King's College London, London, London, 10Imperial College London, London, England, 11Technical University of Munich, Munich, Bavaria, 12University of Oxford, Oxford, Oxfordshire
First Author:
Vyacheslav Karolis
King's College London|FMRIB, University of Oxford
London, United Kingdom|Oxford, United Kingdom
King's College London|FMRIB, University of Oxford
London, United Kingdom|Oxford, United Kingdom
Co-Author(s):
Lucilio Cordero-Grande
Universidad Politécnica de Madrid & CIBER-BBN|King's College London
Madrid, Spain|London, United Kingdom
Universidad Politécnica de Madrid & CIBER-BBN|King's College London
Madrid, Spain|London, United Kingdom
Tomoki Arichi
King's College London|Imperial College London
London, United Kingdom|London, United Kingdom
King's College London|Imperial College London
London, United Kingdom|London, United Kingdom
Eugene Duff
Imperial College London|FMRIB, University of Oxford
London, England|Oxford, United Kingdom
Imperial College London|FMRIB, University of Oxford
London, England|Oxford, United Kingdom
Daniel Rueckert
Technical University of Munich|Imperial College London
Munich, Bavaria|London, United Kingdom
Technical University of Munich|Imperial College London
Munich, Bavaria|London, United Kingdom
Introduction:
Advances in fetal fMRI represent, for the first time, an opportunity for neuroscience to study functional brain connectivity at the time of its emergence [1,2]. The unique challenges of in utero imaging require a community-wide effort to develop tailored methods for image preprocessing and analysis. The progress however has been hampered by the lack of openly available datasets that could be exploited by researchers across disciplines. The dHCP closes this gap by releasing the first open-access and largest-to-date fetal fMRI dataset at https://nda.nih.gov, processed using state-of the art methods.
Methods:
275 completed resting-state fetal fMRI scans (255 unique subjects, 137 male, 116 female, 2 unknown) were acquired with Philips Achieva 3T system and a 32-channel cardiac coil, using a single-shot EPI (TR/TE = 2200/60) sequence, with slice grid = 144 x 143, 48 slices, isotropic resolution = 2.2 mm, multi-band (MB) factor = 3, and SENSE factor = 1.4 [3]. Each scan consists of 350 volumes.
The data underwent 4 stages of preprocessing (Fig 1A): 1) image reconstruction based on soft SENSE ESPIRIT for considering motion or fat-shift induced model inconsistencies [4,5]; 2) dynamic shot-by-shot B0 field correction based on phase unwrapping of complex data using weighted iterative least squares for solving the Poisson equation with iterative correction of residuals after rewrapping [6]; 3) rigid motion correction where volume-to-volume motion estimates are used to initialise slice-to-volume (SVR) corrections with motion states defined jointly for simultaneously excited slices, using a simplified version of [7]; 4) optimised temporal denoising.
Specific artefacts were targeted with denoising: 1) failure of SVR corrections in the presence of large motion; 2) potential effects of residual leakages which could not be suppressed with SENSE reconstruction; 3) residual effects of distortion corrections; 4) motion-induced artefacts, including spin history artefacts, manifesting themselves as spatially non-stationary travelling waves. To address these, the pipeline utilises novel types of 4D (voxelwise) regressor maps, in addition to traditionally used volume censoring and data-derived white matter and CSF timecourse regressors.
The dHCP data release also includes an advanced volumetric mapping infrastructure between native and template spaces using one interpolation step (Fig 1B), that enables group-level analyses and synthesis of the fMRI data with structural and diffusion data, acquired in the same cohort, and the wider neonatal cohort of the dHCP [8].
The data underwent 4 stages of preprocessing (Fig 1A): 1) image reconstruction based on soft SENSE ESPIRIT for considering motion or fat-shift induced model inconsistencies [4,5]; 2) dynamic shot-by-shot B0 field correction based on phase unwrapping of complex data using weighted iterative least squares for solving the Poisson equation with iterative correction of residuals after rewrapping [6]; 3) rigid motion correction where volume-to-volume motion estimates are used to initialise slice-to-volume (SVR) corrections with motion states defined jointly for simultaneously excited slices, using a simplified version of [7]; 4) optimised temporal denoising.
Specific artefacts were targeted with denoising: 1) failure of SVR corrections in the presence of large motion; 2) potential effects of residual leakages which could not be suppressed with SENSE reconstruction; 3) residual effects of distortion corrections; 4) motion-induced artefacts, including spin history artefacts, manifesting themselves as spatially non-stationary travelling waves. To address these, the pipeline utilises novel types of 4D (voxelwise) regressor maps, in addition to traditionally used volume censoring and data-derived white matter and CSF timecourse regressors.
The dHCP data release also includes an advanced volumetric mapping infrastructure between native and template spaces using one interpolation step (Fig 1B), that enables group-level analyses and synthesis of the fMRI data with structural and diffusion data, acquired in the same cohort, and the wider neonatal cohort of the dHCP [8].
Results:
263 scans were fully processed. Fig 2A-C shows the outputs of the dHCP fetal preprocessing steps for an exemplar subject following MB-SENSE reconstruction. Fig 2A shows the effect of distortion corrections on the brain geometry in comparison to the raw reconstructed image. Figure 2B shows the effect of slice-to-volume reconstruction on temporal signal-to-noise ratio compared to to volume-to-volume alignment, the default approach in ex-utero image preprocessing. Fig 2C shows the effect of temporal denoising on the temporal evolution of the signal for the same subject. Finally, capabilities of the registration infrastructure are exemplified by the results of group-level ICA analysis, as shown in Fig 2D.
Conclusions:
The dHCP fetal fMRI dataset is designed to promote fetal MRI from its current status as a niche research field to its deserved and timely place in the community-wide effort to build a life-long connectome of the human brain. By releasing the data at different pre-processing stages, we ensure that researchers with diverse scientific background can benefit from this dataset, starting from reconstruction and preprocessing method developers, who can utilise raw or partially pre-processed data to benchmark performance of their models, to modelers of neurodevelopment, who can use the fully pre-processed data and advanced registration infrastructure to probe key questions about early brain development.
Lifespan Development:
Lifespan Development Other 2
Modeling and Analysis Methods:
Methods Development
Motion Correction and Preprocessing
Neuroinformatics and Data Sharing:
Databasing and Data Sharing 1
Keywords:
Data Organization
Development
FUNCTIONAL MRI
Open Data
1|2Indicates the priority used for review
Provide references using author date format
2) Ferrazzi, G. et al. (2013). Resting State fMRI in the moving fetus: A robust framework for motion, bias field and spin history correction. Neuroimage 101, 555-568.
3) Price, A. N. et al. (2019). The developing Human Connectome Project (dHCP): fetal acquisition protocol. Proceedings of the annual meeting of the International Society of Magnetic Resonance in Medicine (ISMRM). 244.
4) Zhu, K. et al. (2016). Hybrid-space SENSE reconstruction for simultaneous multi-Slice MRI. IEEE Transactions on Medical Imaging, 35, 1824-36.
5) Uecker, M. et al. (2014). ESPIRIT - An eigenvalue approach to autocalibrating parallel MRI: where SENSE meets GRAPPA. Magn Reson Med, 71, 990-1001.
6) Zhao, Z., Zhan, H., Ma, C., Fan, C. & Zha, H. (2020). Comparative study of phase unwrapping algorithms based on solving the Poisson equation. Measurement Science and Technology, 31, 065004.
7) Cordero-Grande, L., Hughes, E. J., Hutter, J., Price, A. N. & Hajnal, J. V. (2018). Three-dimensional motion corrected sensitivity encoding reconstruction for multi-shot multi-slice MRI: Application to neonatal brain imaging. Magn Reson Med, 79, 1365-1376.
8) Edwards, A. D. et al. (2022). The Developing Human Connectome Project Neonatal Data Release. Frontiers in Neuroscience 16, doi:10.3389/fnins.2022.886772