SDCFlows: convergent formalization of MRI field mappings for reliable distortion correction
Poster No:
1692
Submission Type:
Abstract Submission
Authors:
Oscar Esteban1, Matthew Cieslak2, Mathias Goncalves3, Eilidh MacNicol4, Russell Poldrack5, Christopher Markiewicz3
Institutions:
1Lausanne University Hospital and University of Lausanne, Lausanne, VD, 2UPenn, Philadelphia, PA, 3Stanford University, Stanford, CA, 4Neuroimaging, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom, 5Stanford University, Palo Alto, CA
First Author:
Co-Author(s):
Eilidh MacNicol
Neuroimaging, Institute of Psychiatry, Psychology and Neuroscience, King’s College London
London, United Kingdom
Neuroimaging, Institute of Psychiatry, Psychology and Neuroscience, King’s College London
London, United Kingdom
Introduction:
Echo-planar imaging (EPI) allows very fast acquisition of whole-brain data enabling standard functional and diffusion MRI. However, small deviations in parts-per-million from the nominal B₀ caused by steps in magnetic susceptibility at tissue interfaces introduce misplacements in the registered location of voxels of up to centimeters. Geometrical distortions of the imaged specimen become apparent along the phase-encoding direction. In humans, geometrical distortions are prominent near the ventromedial prefrontal cortex and temporal lobes due to the proximity of air in the nasal sinuses and ear canals. The expected voxel shifts can be estimated by mapping B₀ deviations (so-called fieldmaps). Abundant literature has supported the consolidation of a few families of estimation: using ad-hoc B₀ inhomogeneity mapping with specific MRI schemes¹ (most often called "phase-difference"), measuring the point spread function², acquiring two or more short EPI schemes with varying phase-encoding polarity and/or direction³ (called "pepolar" in the following), and by image registration to an undistorted reference such as T₂-weighted image⁴ (often referred to as "fieldmapless"). Here, we introduce a BIDS-App to execute SDCFlows (Susceptibility Distortion Correction Flows), an open-source utility that leverages BIDS⁵ and several existing software tools to provide standardized, best-effort B₀ inhomogeneity map estimates and corresponding distortion correction.
Methods:
Data. The development of the core libraries employed several datasets openly available at OpenNeuro⁶. Bugfixes and debugging were mostly based on minimally viable datasets provided by users of fMRIPrep⁷, which is the principal mode of distribution and deployment of SDCFlows. To evaluate the new BIDS-App, we employed the Human Connectome PHantom (HCPh) dataset⁸.
Implementation. The new BIDS-App implementation is based on the NiPreps framework, akin to fMRIPrep. Therefore, the user interface follows the corresponding specifications and requires the input dataset to be BIDS-compliant. At the output, preprocessed fieldmaps ready to correct EPI volumes are generated under BIDS-Derivatives conventions. The software first processes the input with PyBIDS⁹, and identifies the estimable field maps leveraging the BIDS metadata. These estimable maps are represented by FieldmapEstimation objects. Then, the get_workflow() member is called on each object, which generates the corresponding processing and estimation workflow depending on the type of fieldmap. All the workflows are then loaded into a NiPype¹⁰ computational graph, which can be executed in parallel and distributable through a number of platforms.
Implementation. The new BIDS-App implementation is based on the NiPreps framework, akin to fMRIPrep. Therefore, the user interface follows the corresponding specifications and requires the input dataset to be BIDS-compliant. At the output, preprocessed fieldmaps ready to correct EPI volumes are generated under BIDS-Derivatives conventions. The software first processes the input with PyBIDS⁹, and identifies the estimable field maps leveraging the BIDS metadata. These estimable maps are represented by FieldmapEstimation objects. Then, the get_workflow() member is called on each object, which generates the corresponding processing and estimation workflow depending on the type of fieldmap. All the workflows are then loaded into a NiPype¹⁰ computational graph, which can be executed in parallel and distributable through a number of platforms.
Results:
Single representation model for all estimation methodologies. SDCFlows represents all map estimations with a B-Spline basis, ensuring the smoothness of the solution, easier to manipulate (e.g., map onto a target image), and a compressed representation with fewer parameters than the corresponding dense field. We generated 131 fieldmap estimations corresponding to 50 phase-difference and 81 pepolar estimators, amounting to all the available estimable fieldmaps in the dataset. By sharing a common basis, different estimation strategies can be compared and combined. The single representation model also enabled the implementation of a B0FieldTransform object that permits head-motion correction and susceptibility distortion correction with a single interpolation step (Fig. 1).
A BIDS-Apps interface and the NiPreps transparency principles. SDCFlows can easily be integrated into the neuroimaging pipeline thanks to its BIDS-Apps compliance and generates comprehensive diagnostic figures (Fig. 2) for their aggregation within other tools such as fMRIPrep.
A BIDS-Apps interface and the NiPreps transparency principles. SDCFlows can easily be integrated into the neuroimaging pipeline thanks to its BIDS-Apps compliance and generates comprehensive diagnostic figures (Fig. 2) for their aggregation within other tools such as fMRIPrep.
·Figure 2
Conclusions:
The standardization of distortion estimation and correction offered by SDCFlows will equip researchers with a reliable, off-the-shelf solution to solidify their neuroimaging workflow.
Modeling and Analysis Methods:
Exploratory Modeling and Artifact Removal 1
Image Registration and Computational Anatomy
Methods Development 2
Neuroinformatics and Data Sharing:
Informatics Other
Keywords:
Computational Neuroscience
Computing
Data analysis
Data Registration
FUNCTIONAL MRI
Informatics
Open-Source Software
1|2Indicates the priority used for review
·Figure 1 - https://www.nipreps.org/sdcflows/master/notebooks/SDC%20-%20Theory%20and%20physics.html
Provide references using author date format
2. Zaitsev, M., Hennig, J. & Speck, O. Point spread function mapping with parallel imaging techniques and high acceleration factors: Fast, robust, and flexible method for echo-planar imaging distortion correction. Magn. Reson. Med. 52, 1156–1166 (2004).
3. Chiou, J. Y. & Nalcioglu, O. A simple method to correct off-resonance related distortion in echo planar imaging. in Proc. Intl. Soc. Mag. Reson. Med. vol. 8 1711 (2000).
4. Studholme, C., Constable, R. T. & Duncan, J. S. Accurate alignment of functional EPI data to anatomical MRI using a physics-based distortion model. IEEE Trans. Med. Imaging 19, 1115–1127 (2000).
5. Poldrack, R. A. et al. The Past, Present, and Future of the Brain Imaging Data Structure (BIDS). Preprint at https://doi.org/10.48550/arXiv.2309.05768 (2023).
6. Markiewicz, C. J. et al. The OpenNeuro resource for sharing of neuroscience data. eLife 10, e71774 (2021).
7. Esteban, O. et al. fMRIPrep: a robust preprocessing pipeline for functional MRI. Nat. Methods 16, 111–116 (2019).
8. Provins, C. et al. Reliability characterization of MRI measurements for analyses of brain networks on a human phantom. Nat. Methods (Stage 1 accepted-in-principle), (2023).
9. Yarkoni, T. et al. PyBIDS: Python tools for BIDS datasets. J. Open Source Softw. 4, 1294 (2019).
10. Gorgolewski, K. et al. Nipype: a flexible, lightweight and extensible neuroimaging data processing framework in python. Front. Neuroinformatics 5, 13 (2011).