A multimodal ultrahigh-field MRI processing pipeline

2278 

Abstract Submission 

Raul Rodriguez Cruces¹, Alexander Ngo², Donna Gift Cabalo¹, Jessica Royer³, Youngeun Hwang⁴, Peer Herholz⁵, Nicole Eichert⁶, Yezhou Wang¹, Oualid Benkarim², Jordan DeKraker⁷, Christine Tardif⁸, Boris Bernhardt²

¹McGill University, Montreal, Quebec, ²Montreal Neurological Institute and Hospital, Montreal, Quebec, ³Montreal Neurological Institute and Hospital, Montreal, QC, ⁴McGIll University, Montreal, Quebec, ⁵McConnell Brain Imaging Centre, The Neuro (Montreal Neurological Institute-Hospital), Montreal, QC, ⁶University of Oxford, Oxford, Oxfordshire, ⁷McGill University, Montreal, Canada, ⁸McConnell Brain Imaging Centre, Montreal Neurological Institute and Hospital, Montreal, Quebec

Raul Rodriguez Cruces  
McGill University
Montreal, Quebec

Alexander Ngo  
Montreal Neurological Institute and Hospital
Montreal, Quebec

Donna Gift Cabalo  
McGill University
Montreal, Quebec

Jessica Royer  
Montreal Neurological Institute and Hospital
Montreal, QC

Youngeun Hwang  
McGIll University
Montreal, Quebec

Peer Herholz  
McConnell Brain Imaging Centre, The Neuro (Montreal Neurological Institute-Hospital)
Montreal, QC

Nicole Eichert  
University of Oxford
Oxford, Oxfordshire

Yezhou Wang  
McGill University
Montreal, Quebec

Oualid Benkarim  
Montreal Neurological Institute and Hospital
Montreal, Quebec

Jordan DeKraker  
McGill University
Montreal, Canada

Christine Tardif  
McConnell Brain Imaging Centre, Montreal Neurological Institute and Hospital
Montreal, Quebec

Boris Bernhardt  
Montreal Neurological Institute and Hospital
Montreal, Quebec

Ultrahigh-field Magnetic Resonance Imaging (MRI) at 7T provides a unique opportunity to study the brain's microstructure, function, and connectivity in vivo at unprecedented resolutions. However, the increasing complexity and multimodality of the data demand processing methods that can effectively integrate information across modalities and spatial scales. In this study, we propose a standardized workflow for processing multiparametric 7T MRI acquisitions, combining state-of-the-art surface generation and multiparametric registrations. This pipeline automatically generates multiscale connectomes and surface maps, derived from quantitative MRI microstructural similarity, geodesic distance mapping, functional connectivity, and diffusion MRI tractography. Additionally, it is backwards compatible with datasets acquired a conventional field strength, such as 3T MRI. In line with open science practices, the workflow is available as part of BIDS apps, containerized and hosted on Docker hub and all code is available through GitHub.

To create a functional workflow for 7T images, we implemented various enhancements to our previous pipeline, micapipe (Rodriguez-Cruces, 2022). Regarding structural processing (F1.A), we introduced a denoising algorithm to process MP2RAGE images as structural data and generated brain masks using a deep learning-based algorithm (mri_synthstrip, Hoopes, 2022). Additionally, we normalized the white matter to achieve an intensity homogeneous T1-weighted structural image (T1nativepro). For surface generation (F1.B), we adopted FastSurfer (Henschel, 2020), a deep learning-based tool that proved to be faster and more convenient for high-resolution surface generation compared to traditional methods. To map quantitative images to different surface spaces at the pial, mid-thickness, and white matter surfaces, all derived from FastSurfer, we incorporated new tools that first mapped them to the native space of the structural image (F1.C). Our microstructural profile covariance (MPC) module (F1.D) now applies surface sampling directly from the original qMRI space, enhancing reliability and avoiding potential interpolation issues. Finally, we introduced a novel approach for more accurate registration between modalities, utilizing label-based modality agnostic registration (F1.E, Billot, 2023). This technique combines deep learning-based segmentation and numerical solutions to generate precise warpfields, even for modalities with high signal-to-noise ratio and signal dropout, such as DWI and functional acquisitions (F1.E). Finally, our functional module handles resting state and multiple task acquisitions and includes Time-echo dependent analysis for multi-echo processing (TEDANA; DuPre, 2021). Individual and group-level quality control (QC) can be run at any point during the processing. The QC procedure will generate a pdf report file for each subject containing visualizations of intermediate files for volume visualization, cross-modal co-registrations, and surface parcellations. Moreover, it allows inspection of inter-regional matrices such as structural connectomes, functional connectomes, microstructural profile covariance, and geodesic distance matrices.

This standardized workflow for processing multiparametric 7T MRI acquisitions presents a comprehensive solution for studying the brain's microstructure, function, and connectivity. Leveraging Ultrahigh-field Magnetic Resonance Imaging at 7T, our pipeline integrates state-of-the-art surface generation and multiparametric registrations to automatically generate multiscale connectomes and surface maps.

Our optimized 7T MRI pipeline is the first standardized workflow for processing multiparametric acquisitions from a BIDS directory to ready-to-use matrices and surface maps (F2). This advancement greatly benefits the development of multimodal brain models. Leveraging higher resolution images obtained from 7T, it offers full support to explore new frontiers in brain research.

Methods Development ²

Workflows ¹

FUNCTIONAL MRI

MRI

Open-Source Code

Open-Source Software

STRUCTURAL MRI

Workflows