Merge pull request #106 from PennLINC/mri-processing

Add information on common MRI data and processing

docs/CommonMethods/mri_processing.md

@@ -0,0 +1,181 @@
+---
+layout: default
+title: Neuroimaging Data Processing Tools
+parent: Common Methods
+---
+
+There are many tools out there for processing and analyzing neuroimaging data.
+In the PennLINC team, we typically acquire magnetic resonance imaging (MRI)
+data with a handful of modalities, such as functional MRI (fMRI),
+diffusion-weighted imaging (DWI),
+structural MRI (sMRI) (especially T1- and/or T2-weighted imaging),
+and arterial spin labeling (ASL).
+This page summarizes the acquisitions we tend to use,
+as well as the processing tools we use for each.
+
+We almost exclusively acquire our data on a 3T Siemens Prisma MRI scanner.
+
+## Table of Contents
+{: .no_toc .text-delta }
+
+1. TOC
+{:toc}
+
+## Structural Magnetic Resonance Imaging (sMRI)
+
+The two main types of structural images we tend to acquire are
+high-resolution T1-weighted and T2-weighted images.
+
+A high-resolution T1-weighted image is required for most of the processing workflows
+for other modalities, such as fMRI, DWI, and ASL.
+
+We generally try to acquire a high-resolution T2-weighted image as well,
+in order to (1) improve Freesurfer segmentation and (2) to calculate a myelin-weighted map.
+
+### sMRIPrep
+
+We recommend using sMRIPrep to process T1w and T2w images.
+sMRIPrep is a BIDS App that minimally preprocesses structural MRI data,
+runs Freesurfer to segment the data and project it to the surface,
+and warps the data to requested template spaces.
+
+{: .note-title }
+> Tip
+>
+> A number of processing pipelines for other MRI modalities,
+> including fMRIPrep and ASLPrep, will automatically run sMRIPrep.
+> We generally rely on the structural processing from these pipelines
+> instead of running sMRIPrep directly.
+
+### Freesurfer
+
+Freesurfer is a very popular tool for segmenting sMRI data and projecting it to the surface.
+We generally run Freesurfer within one of the BIDS Apps (sMRIPrep, fMRIPrep, ASLPrep, etc.)
+rather than running it separately.
+
+
+## Functional Magnetic Resonance Imaging
+
+### PennLINC-preferred fMRI Protocols
+
+Starting in 2023, we began using a specific fMRI protocol in our studies.
+This is a multi-echo fMRI protocol with complex reconstruction and
+no-excitation noise volumes acquired at the end of each run.
+
+{: .highlight-title }
+> Fun fact
+>
+> This is the same basic protocol used in
+> [Siegel et al. (2024)](https://www.nature.com/articles/s41586-024-07624-5) and
+> [Moser et al. (preprint)](http://biorxiv.org/lookup/doi/10.1101/2023.10.27.564416).
+>
+> We have some openly-available data with this protocol on
+> [OpenNeuro](https://openneuro.org/datasets/ds005250).
+
+#### Multi-echo fMRI
+
+We use multi-echo fMRI instead of the more common single-echo approach.
+While acquiring multiple echoes means that we end up with a longer TR than common
+highly-accelerated protocols, such as the HCP protocol,
+multi-echo data have improved SNR over single-echo data,
+and the echoes can be leveraged to distinguish BOLD and non-BOLD noise with ``tedana``.
+
+For more information on the costs and benefits of multi-echo fMRI please see
+the ME-ICA team's
+[multi-echo data analysis book](https://me-ica.github.io/multi-echo-data-analysis/content/intro.html)
+or the [tedana documentation](https://tedana.readthedocs.io/en/stable/).
+
+#### Complex-valued fMRI
+
+We enable complex reconstruction for our fMRI sequences,
+which produces both magnitude and phase data rather than the more common magnitude-only
+reconstruction.
+While there are many methods to leverage phase information in fMRI data out there
+in the literature,
+in practice we mostly use the phase data for thermal noise removal with NORDIC.
+
+In the future, we will be able to use the phase data for the following:
+
+- Dynamic distortion correction with the MEDIC algorithm.
+- Phase regression (see [`nipreps/fMRIPost-phase`](https://github.com/nipreps/fmripost-phase))
+- Complex-valued ICA (see [`nipreps/fMRIPost-phase`](https://github.com/nipreps/fmripost-phase))
+- Phase jolt and phase jump time series calculation (see [`nipreps/fMRIPost-phase`](https://github.com/nipreps/fmripost-phase))
+
+#### No-excitation Noise Volumes
+
+At the end of the protocol, we acquire 3 volumes without a radiofrequency pulse.
+These no-excitation volumes are then used to characterize the thermal noise levels in the data,
+which improves thermal noise removal by NORDIC.
+
+
+#### Concurrent Physiological Recording
+
+We try to acquire cardiac data with plethysmography and respiration data with
+a chest belt during our fMRI acquisitions.
+In practice, we've had trouble with the chest belt.
+
+These physiological data can be used to remove non-neural noise from the fMRI data,
+as well as for direct analysis.
+
+
+### Processing fMRI Data
+
+In order to process these unique data, we use the following basic workflow:
+
+1. [complex-valued data only] Thermal noise removal with NORDIC.
+2. Minimal preprocessing with fMRIPrep.
+3. [multi-echo data only] Multi-echo denoising with tedana.
+4. Post-processing and connectivity extraction with XCP-D.
+5. Statistical analysis with (usually) custom R code.
+
+
+## Diffusion-Weighted Imaging
+
+### PennLINC-preferred DWI Protocols
+
+We generally use compressed-sensing diffusion spectrum imaging (CS-DSI)
+as our preferred DWI protocol.
+
+{: .highlight-title }
+> Fun fact
+>
+> We have some openly-available data with our preferred CS-DSI protocol on
+> [OpenNeuro](https://openneuro.org/datasets/ds004737).
+> One important modification is that we enable complex reconstruction,
+> which will allow for improved thermal noise removal with MP-PCA.
+
+### Processing DWI Data
+
+We use the following workflow for DWI data:
+
+1. Minimal preprocessing with QSIPrep.
+2. Reconstruction with QSIRecon.
+3. Statistical analysis with (usually) custom R code.
+
+
+## Arterial Spin Labeling
+
+Arterial spin labeling (ASL) is a sequence that tags blood in the neck and
+tracks the flow of that blood into the brain.
+The primary output of an ASL sequence is cerebral blood flow (CBF),
+although some protocols, such as multi-delay ASL,
+can be used to estimate other meaningful blood flow measures,
+such as arterial transit time (ATT), arterial bolus arrival time (aBAT),
+and arterial blood volume (ABV).
+
+### PennLINC-preferred ASL Protocols
+
+Generally, for developmental samples (PennLINC's bread and butter),
+we use a single-delay PCASL protocol developed by Manuel Taso.
+This protocol is extremely quick (~4:30) and provides both ASL data and
+a low-resolution quantitative T1 map.
+
+For aging populations, we might use a multi-delay PCASL protocol,
+but that hasn't come up in our internal studies yet.
+
+### Processing ASL Data
+
+We use the following workflow for ASL data:
+
+1. Preprocessing, CBF estimation, and connectivity extraction with ASLPrep.
+2. Statistical analysis with (usually) custom R code.